mirror of
https://github.com/THU-MAIC/OpenMAIC.git
synced 2026-10-17 05:29:46 +08:00
@openmaic/dsl@0.10.1
75
Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 wyuc
|
e3fd494658 |
feat(tts): add Qwen TTS voice cloning (#1160)
* feat(tts): add Qwen TTS voice cloning via the registration adapter * feat(settings): add Qwen TTS clone-voice manager * fix(tts): harden Qwen voice cloning integration * fix(tts): enforce voice-bound Qwen synthesis * fix(tts): close Qwen voice clone tail issues * fix(tts): close final voice clone review issues * fix(agents): harden voice token parsing * fix(settings): hide the voice-clone model from manual selection * fix(agents): pin the narrator voice to the user's selected voice * fix(agents): keep narrator fallback alive when the pinned voice is unusable * fix(i18n): correct the narration fallback notice copy * fix(tts): bound the narrator voice retry to a single fallback hop * fix(tts): tolerate transient vendor errors in the voice existence precheck |
||
|
wyucandClaude Fable 5
|
5a3e81c4d9 |
feat(document): ingest uploaded sources into the asset pool and extract by id (#1153 part 0) (#1154)
* feat(document): declare a version on every extractor provider (#1153 part 0) * feat(api): accept asset-id input in the extract-document route (#1153 part 0) * feat(upload): ingest selected course materials into the asset pool (#1153 part 0) * fix(document): harden asset-id extraction inputs and close upload ingest leaks (#1153 part 0) * fix(document): scope extraction fallback and harden the asset-id path (#1153 part 0) - G1: byte fallback only on pre-extraction failures; PARSE_FAILED (422/500) is surfaced without re-running the paid extractor - G2: freeze the course-material set for the duration of generate-prep (guarded add/remove + disabled toolbar affordances, session built from the click-time snapshot, belt-and-braces drain) - G3: bound the pre-generation ingest await (~15s); timed-out ingests go byte-path and their late-resolving ids are released - G4: reject non-object JSON bodies (null/array/string/number) as 400 - G5: JSON-path provider pre-validation with generic messages, generic JSON-path PARSE_FAILED/422 responses, control-char-free log values - G6: size cap enforced via identify() before resolve() materializes bytes; too_large maps to the existing 413 * fix(document): mirror provider-hint semantics on the asset-id form and freeze prep-time controls (#1153 part 0) * fix(document): keep the asset-id form's responses free of caller input (#1153 part 0) Address review findings: answer the extractor-selection failure with a generic static message on the asset-id form (the registry's interpolated message carries the caller's MIME type; multipart keeps it verbatim, both sides now pinned by tests), dedupe same-batch material additions by content fingerprint inside the pure updater, document the network-error retry exception in shouldRetryWithByteUpload's docstring, and fix a test name asserting 500 but named 400. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> --------- Co-authored-by: Claude Fable 5 <noreply@anthropic.com> |
||
|
LING DUANandwyuc
|
0d308ed78d |
feat(pi): add Native Child Web Search (#1133)
* feat(pi): add native child web search * fix(pi): align native web search with provider system * fix(pi): harden native web search configuration and cancellation * fix(pi): guide Native Child web search selection --------- Co-authored-by: wyuc <wang-yc24@mails.tsinghua.edu.cn> |
||
|
wyuc
|
e15780835a |
feat(media): convert legacy references to allocated asset ids (#1101)
Lands epic #1007 part 2(c): the app-side reference converter, the `audioUrl` removal, and the `DSL_VERSION` bump as one delivery unit. `convertDocumentAssetRefs` walks a loaded document (slide canvases, whiteboards, video-manifest keys, speech actions) and rewrites legacy handles to allocated pool ids: placeholders and `audioId`s with local bytes are ingested, a co-present `audioUrl`/`audioId` pair collapses to one asset, a definitively dead URL converts to an emptied reference, and a transient failure keeps both handles for a later retry. Unavailable bytes leave the document untouched for a later open, and a converted document returns by identity with no writes. Conversion runs lazily on document open and at fetch time for server-generated classrooms, with conversion and the first document save sharing the per-stage document lock: once the document commits it owns every allocation, and every failure exit rolls its pass ledger back. Legacy URL probes are bounded by a per-probe timeout and a shared 60-second pass budget. `SpeechAction.audioUrl` is removed from the DSL and every consumer reads the converted shape; the server classroom generator keeps its derived-id + serving-URL pair as an explicitly typed pre-conversion transport, consumed before persistence. `DSL_VERSION` becomes 0.2.0 with the first real ladder entry, and `@openmaic/dsl` goes to 0.9.0 per the release rule. Deferred follow-ups are tracked in #1129: production principal derivation with document/asset ownership aligned, conditional document writes, and hoisting the aggregate probe budget into the export and video-timeline paths. Refs #1007 |
||
|
wyuc
|
3b4567076f |
refactor(generation): consume the package everywhere and delete lib/generation (#1090)
Part D of #1057 — the switch. All consumers import @openmaic/generation; lib/generation deleted (net -11,900 lines). PBL wiring preserved (single-call -> classified fallback -> injected loop), streaming outline route deduplicated onto buildOutlinePrompt, serverExternalPackages wired, and the change verified by a real-model four-kind end-to-end run against the production build with a rendered-classroom screenshot check. Closes #1057. |
||
|
|
5d5acd6215 |
feat: claude search (#393)
* Added search tool support for Claude * New locales for i18n * Updated translations * Prevent default web search base URLs from being prepopulated * fixed coding style issues * fixed coding style issues * fixing lint errors * fix: add SSRF protection to fetchPageContent in claude.ts Agent-Logs-Url: https://github.com/joseph-mpo-yeti/OpenMAIC/sessions/099e53fb-e1db-49f4-af7f-4f17cb3d364a Co-authored-by: joseph-mpo-yeti <55380155+joseph-mpo-yeti@users.noreply.github.com> * fix: address all remaining review feedback from PR review thread Agent-Logs-Url: https://github.com/joseph-mpo-yeti/OpenMAIC/sessions/e1ab8fe2-fe0d-45a5-a0af-30945b99c268 Co-authored-by: joseph-mpo-yeti <55380155+joseph-mpo-yeti@users.noreply.github.com> * fix: rename ambiguous variable names in settings store for clarity Agent-Logs-Url: https://github.com/joseph-mpo-yeti/OpenMAIC/sessions/e1ab8fe2-fe0d-45a5-a0af-30945b99c268 Co-authored-by: joseph-mpo-yeti <55380155+joseph-mpo-yeti@users.noreply.github.com> * fixing lint errors * fix: address 5 items from second PR review thread Agent-Logs-Url: https://github.com/joseph-mpo-yeti/OpenMAIC/sessions/09469d80-8e91-449b-8bc2-9c11f83555df Co-authored-by: joseph-mpo-yeti <55380155+joseph-mpo-yeti@users.noreply.github.com> * fixing model verification error * fix: address 3 items from third PR review thread Agent-Logs-Url: https://github.com/joseph-mpo-yeti/OpenMAIC/sessions/0aa49d19-df5b-4a08-a7e1-2295ae9cfc38 Co-authored-by: joseph-mpo-yeti <55380155+joseph-mpo-yeti@users.noreply.github.com> * fixing lint errors * fix: resolved Copilot review comments * fix: returning string promise for ssrf-guard * fix: returning string promise for ssrf-guard * fix: updated locales, added allowed_callers for claude search tools, handle duplicated claude model keys * fix: updated locales, added allowed_callers for claude search tools, handle duplicated claude model keys * updated tests * using Anthropic SDK for search and connection test in search settings. * uses @ai-sdk with a custom fetch including allowed_callers in payload instead of adding Anthropic SDK as a direct dependency * fixed format * removed @anthropic-ai/sdk as direct dependency * fix: resolve TypeScript errors in claude web-search tests Cast mock.calls through unknown to avoid tuple-index type errors on Vitest mock objects whose parameter types are inferred as empty tuples. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * feat: add Claude as a web-search provider on the multi-provider framework Forward-port of the original Claude web-search work onto main's searchWeb dispatcher, per review guidance on #393: - lib/web-search/claude.ts: searchWithClaude via the AI SDK Anthropic provider and Anthropic's native web_search server tool, called through callLLM for usage accounting. Returns the model's answer plus cited sources deduplicated by URL. The page-content enrichment fetch from the original PR is removed entirely, resolving the P1 SSRF finding (no server-side fetches of attacker-influenced result URLs remain). - allowed_callers: ["direct"] is still injected at the fetch layer; the AI SDK's provider-defined web_search tools do not emit it. - Tool version picked from the model: web_search_20260209 for 4.6+ Opus/Sonnet, basic web_search_20250305 for Haiku/older; maxResults maps to the tool's max_uses. - Registry/dispatcher/route wiring: claude provider entry, claudeModelId passthrough (web-search route, classroom generation, generation preview), WEB_SEARCH_CLAUDE_* env config with server-pinned model via WEB_SEARCH_CLAUDE_MODELS, and client base-URL allowlisting for api.anthropic.com. - Settings: per-provider modelId with a model selector in the Claude panel; locale keys added to all ten languages. - Tests: new adapter suite plus claude cases in the dispatcher, constants, route, provider-config, web-search-config, and settings store suites. Closes #392 Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01E9KqAGmKCtBwMC4HtT4JKE * fix(web-search): trim whitespace around the Claude API key and base URL Keys are pasted by hand into Settings, and nothing in the client key path trimmed them: a trailing newline or space was sent verbatim as the x-api-key header value and came back from Anthropic as 'invalid x-api-key', with nothing in the message pointing at whitespace. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> * fix(web-search): correct Claude adapter tool selection, base URL, and timing Addresses the four P2 review findings on #393: - Route dated Claude 4.0 ids (claude-sonnet-4-20250514, claude-opus-4-20250514) to the basic web_search_20250305 tool. The blocklist regex missed them, so they were sent web_search_20260209, which pre-4.6 models reject. Replaced with an allowlist of the models that have dynamic filtering, so unrecognized ids also fall back to the basic tool -- the safe direction, since 4.6+ models accept both. - Pin allowed_callers: ["direct"] only on web_search_20250305. Injecting it into every request disabled the code-execution caller that provides dynamic filtering on web_search_20260209. - Normalize the bare Anthropic root to /v1 before handing the base URL to the SDK, which appends /messages verbatim and would otherwise 404. The adapter is the single choke point for the client allowlist, the server env var, YAML, and the default; other hosts pass through untouched. - Return responseTime in seconds, matching every sibling adapter's WebSearchResult contract instead of exposing milliseconds for Claude. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> --------- Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com> Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com> |
||
|
wyuc
|
0ae21aa559 |
refactor(pbl): retire v1 write path and dead UI, converge on v2 (#1060)
* refactor(pbl): retire v1 write path and dead UI (#1058) Remove the legacy generator, renderer branches, chat API, and compatibility mirror writes. Keep stored v1 scenes readable through documented legacy adapters, and surface ordinary PBL planner double failures instead of degrading. * fix(pbl): contain planner failure to the scene, support legacy scenes in action regeneration, tighten legacy guards - Contain typed planner failures per scene while preserving retry and skip behavior. - Upgrade legacy PBL content before regenerating scene actions. - Apply video-export import boundaries to legacy readers. - Harden legacy rendering and empty-state handling. - Restore legacy upgrade compatibility coverage. - Remove retired translations and refresh fallback narration and legacy cover documentation. * fix(pbl): keep tolerant legacy configs renderable, guard invalid chat timestamps * chore(dsl): bump to 0.6.3 for README accuracy fix The README no longer references the deleted PBLProjectConfig symbol; the version gate requires a bump for any publishable package input change. * fix(pbl): harden legacy upgrade and persistence contract * fix(pbl): align action prompt with v2 workflow and validate project containers * fix(i18n): remove retired PBL v1 keys from the new fr-FR locale The fr-FR locale landed on main after this branch removed the v1 UI key families; prune the same 38 keys so all 10 locales align with en-US. * fix(generation): sync packaged pbl-actions template with the v2 wording The generation package that landed on main carries a byte-identical copy of the app's pbl-actions template; sync the v2 rewording and bump the package version. * fix(pbl): normalize legacy content and preserve planner status * fix(pbl): propagate provider status and ground action prompts * fix(pbl): classify planner failures for retry, default the project summary prompt input * fix(pbl): fall back to legacy data when v2 payloads are damaged, harden summary building * refactor(pbl): resolve hybrid content through a validity-aware resolver * fix(pbl): keep one v2 acceptance criterion across validator, resolver and persistence Persistence gates on the same container predicate as the validator and the renderer's resolver: payloads the renderer will not treat as v2 are inert bytes everywhere, avoiding runtime side effects for documents the write barrier would reject. An explicit null proficiencyAssessment stays processable. * fix(pbl): require runnable projects and harden legacy fallback * test(pbl): exercise the real save reload and hydration chain * fix(pbl): require an instructor and usable microtasks before v2 takes precedence * fix(pbl): fall back to the loop planner unless the provider failed or the user cancelled * fix(pbl): require role and task identities before v2 takes precedence Runtime normalization can repair status and create the instructor thread, but it cannot synthesize the role id that binds the thread, the microtask ids used by lookup and update paths, or the labels the workspace renders. The runnable predicate and its video-export mirror now require them, so identity-less payloads fall back to usable legacy data. |
||
|
|
ceb05adb41 |
feat(media): allocate generated assets through the registry (#1007 part 2, step b) (#1039)
* feat(media): establish shared asset ownership primitives Introduce global browser asset-pool ownership, asset-reference collection, stage reclamation planning, and lease-based URL access. Carry allocated media identity through storage and generation boundaries with regression coverage. * fix(media): enforce safe resolution across every consumer Route image, video, thumbnail, presentation, and video-export consumers through one resolution state machine. Prevent opaque allocated or generated references from reaching render and export sinks, with fallback and ownership tests. * fix(media): protect document-owned assets across mutations Allocate pool bytes before compatibility writes and document commits, then roll back uncommitted generations safely. Preserve document ownership across edits, retries, imports, speech generation, scene changes, and stage deletion. * fix(media): scope retries and tighten ownership guard Scope retries to the target scene and slide and refuse ambiguous shared-reference mutations. Expand retry rendering coverage and keep direct pool URL resolution behind the shared lease owner. * fix(storage): avoid nested lock during stage cleanup Execute prepared reclamation plans against an explicitly deleted document so the compatibility cascade cannot re-enter the per-document lock. Cover deletion of stage-owned media rows even when the document has no references. * refactor(media): confine reclamation to stage deletion Remove inline pool and compatibility-row cleanup from element, speech, scene, and audio replacement flows. Keep whole-stage reclamation behind explicit stage ownership, preserve stage-less legacy audio rows, and document deferred document-truth sweeping. * fix(media): close retry and resolution gaps Restore shared source tasks after successful forks, scope retries across both whiteboard locations, and wait for parallel TTS workers before rollback. Resolve background media through import, rendering, and PPTX export paths while keeping retry controls visible over last-good bytes. * test(media): execute the consumer safety matrix Replace source-substring checks with resolver seam execution across all six UI consumers, stage hydration, and both exporters. Pin each rollback layer independently and harden the ownership guard against aliased pool imports. * chore(packages): publish the additive DSL field Bump the DSL patch version for the optional speech-action field. Keep the transitional reclamation policy app-owned and document the legitimate transaction rollback removals there. * fix(media): make retry rollback task-safe Delay allocation task re-keying until final document reconciliation succeeds, restore shared source tasks on failed forks, and require exact stage-whiteboard targets before falling through from a missed scene. * fix(media): clear private assets and refresh leases Delete the asset-pool database during the confirmed local-data wipe. Notify the app lease layer after same-id replacement so mounted consumers re-resolve current bytes without reaching into storage internals. * test(media): pin closure safety guards Exercise the CSS allocation boundary, unfiltered legacy-row ownership, delete-and-undo byte survival, and real distinct consumer seams. Restore the video-only manifest overwrite condition and share the direct video resolution hook across both element variants. * test(media): narrow rollback element assertion Narrow the reconciled slide element to an image before checking its source so the rollback regression remains type-safe under the full root compiler configuration. * fix(media): close lease refresh races Gate the first batch publication by unique resolved refs, then publish every replacement snapshot without mutating the prior React state object. Serialize invalidation behind pending releases, evict rejected refreshes, and register replacement observation at the pool boundary. * fix(media): reopen pool after clear failures Always evict the singleton once its store has been closed, including blocked and failed database deletion paths. Report blocked deletion as deferred and prove a later write uses a fresh live store. * fix(media): isolate shared retry progress Track forked regeneration under the selected element until it receives a fresh asset identity, leaving the shared source task and bytes untouched. Surface targeted failures through renderer task lookup and skip the redundant fork reconciliation lock. * fix(media): close final asset retry gaps Keep blocked asset clears fail-loud until a successful retry, with actionable settings guidance. Clear failed shared-fork state across durable and live key spaces, and pin renderer lookups, lease publication identity, and committed-ref rollback protection. * fix: preserve actionable retry failures Localize the blocked cache-clear recovery hint across every supported locale and pin the deferred-error mapping. Retain durable fork failure rows while retries run, deleting them only after successful generation so unstructured failures survive reload. * fix(media): hydrate legacy stored video thumbnails Home-page recent-video thumbnails regressed for legacy Dexie mediaFiles rows keyed by gen_vid placeholders: the reworked hydration resolved the row's bytes through the sealed resolver but never surfaced the stored blob (and its poster) as object URLs for the preview card, so the CI recent-video-thumbnail e2e specs found no visible element. Hydration now materializes legacy stored video rows into blob URLs for both the element src and poster while keeping the resolver invariants: opaque refs still never reach a DOM src, and concrete addresses are never blanked. Unit pins cover the seam so the vitest suite catches this class without a browser. * fix(media): resolve sole restored legacy video Classroom playback restored tasks by exact document media references. Legacy gen_vid references can outlive the key used by the one persisted video row, leaving the player on a placeholder even though bytes were restored. Select the sole completed stage video only for legacy sequential refs after exact and reconciled matches. Keep exact failures authoritative, refuse ambiguous candidates, and cover success, ambiguity, and failure precedence in unit tests. * fix(media): scope legacy video recovery Decide restored legacy video recovery once from the complete document and record it through the shared task lookup consumed by playback, editing, and resolved slides. Keep ambiguous documents as placeholders, apply the same decision to thumbnail hydration, and preserve exact failure precedence. * fix(media): exclude claimed video recovery tasks Model restored legacy recovery as a two-pass match across document video elements and task rows. Remove tasks claimed by exact, targeted, or placeholder lookup before applying the sole-candidate fallback, and cover the ownership/cardinality matrix. * fix(media): unify video element resolution Centralize source, task, poster, and legacy recovery decisions for every video consumer. Ensure direct URLs win over opaque refs and play_video waits on element-targeted retry tasks. * refactor(media): route legacy recovery through resolver Let document-aware consumers request legacy video recovery through the unified element binding API, keeping thumbnail hydration on a single decision path. * fix(media): preserve import refs and prefer pool bytes Recognize unambiguous extensionless relative media addresses during classroom import. Resolve allocated export and thumbnail assets from the shared pool before falling back to lagging compatibility rows. * fix(export): preserve concrete video sources Route PPTX video elements through the shared media binding resolver so an unresolved opaque reference cannot replace a playable source. Complete browser and persisted-store cleanup when asset-pool deletion is deferred, while retaining distinct hard-failure behavior. * fix(media): fork retries without exclusive ownership Enumerate logical asset owners across every persisted document before allowing global pool replacement. Thread explicit targets through fresh-id rewrites and cover cross-document aliases plus unreadable ownership. * fix(media): guard global asset reclamation Share a fail-closed persisted-document liveness check between stage deletion and retry replacement. Preserve cross-document pool aliases while deleting stage-owned compatibility rows and cover enumeration failures. * fix(media): cover complete slide asset references Route slide media traversal through a shared mutable slot contract so backgrounds participate in export, thumbnail hydration, collection, and rewrite lifecycles. Snapshot complete surviving-document refs once per reclamation and preserve manifest-only owners while retaining fail-closed behavior. * fix(media): preserve exclusive retry asset ids Allow targeted retries to replace exclusively owned pool assets in place. Keep shared and unprovable ownership paths on fresh allocations, and pin production-shaped retries plus compatibility-row cleanup. * fix(media): revalidate asset bindings at completion Recheck repository-wide ownership before replacing generated media and fork scoped retries when exclusivity changed. Route video export selection through the unified resolver and keep concrete posters independent of task state. * fix(media): count unflushed owners and broadcast replacements The completion-time exclusivity proof read only the persisted document, but slide duplication updates the Zustand aggregate synchronously and schedules persistence behind a debounce. A retry finishing inside that window saw a single persisted owner and replaced the bytes behind a reference the duplicate also held. The proof now also counts owners in the live stage snapshot when that snapshot represents the stage being retried, so an unflushed duplicate forks instead. Same-id replacement notifications were realm-local, so a second tab showing the same classroom kept its lease pinned to the superseded blob URL. The notification now travels over a BroadcastChannel; each receiving realm runs its own observers against its own pool, so a spoofed message can at most force a re-resolve. A missing or failing channel never fails the replacement. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> * fix(media): bind replacement listeners and spare shared audio rows A realm that only renders never sends a replacement, so binding the channel from the sender path left passive tabs deaf to peers. Binding now happens where the observer is registered, when the asset-pool module loads, and the receiving realm resolves its own pool lazily so a cleared or unavailable pool degrades to the next resolve instead of throwing. Observer notifications ran under Promise.all, so a rejection surfaced after BrowserAssetStore.replace had already committed and turned a durable success into a reported failure. They are settled individually now; the callback is wrapped because a synchronous throw would otherwise escape before allSettled sees the array. audioFiles rows are keyed globally by audioId, so deleting a stage removed the sole row for an id a surviving document still referenced — playback and both export paths read that table directly and cannot fall back to the preserved pool blob. Rows are now filtered against surviving references, while a failed enumeration still withholds only the irreversible pool removal. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> * fix(media): serve replaced bytes from the pool everywhere Unknown survivor liveness deleted every planned audioFiles row. Those rows are keyed globally by audioId, and playback plus both export paths read the table directly, so losing one is as irreversible for them as removing the pool entry. Unknown liveness now preserves the rows too, leaving bounded garbage for a later pass that can prove exclusivity. The earlier pool-first change covered PPTX, video collection and thumbnail hydration but missed classroom ZIP export, which still serialized the stale compatibility row after a lagged same-id replacement, shipping media the classroom no longer renders. Auditing every direct reader of the media and audio tables surfaced the same gap in playback: speech regeneration also replaces bytes under a stable id and does not roll the pool back when the compatibility write fails, so the player kept serving superseded narration. It now resolves the pool first and falls back to stored rows for legacy and imported audio. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> * fix(audio): replace exclusively owned speech clips in place regenerateSpeechAudio accepted the action's audioId but always passed undefined as replaceAssetId, so regenerating an exclusively owned pool-backed clip allocated a new asset every time, rewrote the action and orphaned the previous pool entry and compatibility row until stage reclamation — contradicting the stable-id path generateAndStoreTTS already implements for media. Ownership is now established before synthesis, and the rule itself moved to the shared reference module so media retries, poster replacement and speech regeneration consume one implementation instead of restating it. An exclusively owned clip keeps its id and has its bytes replaced; a shared clip, a legacy id with no pool entry, or unprovable ownership still gets a fresh allocation so other holders keep their audio. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> * chore(media): drop the import left behind by the ownership move Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> * fix(media): resolve audio pool-first and fence peer realms Stable-id TTS regeneration commits replaced narration to the pool before the audioFiles mirror write, so a failed mirror leaves the row stale. AudioPlayer already resolved pool-first, but audioObjectUrl, collectAudioFiles and the video timeline dependencies still read Dexie directly and would serve the superseded clip. All allocated-audio readers now share one resolver, with Dexie kept as the fallback for legacy and imported rows. The exclusivity proof modelled unflushed owners in the active realm only, so another tab duplicating the same asset during its save debounce could still be observed as a single owner and have its bytes replaced globally. A peer's pending state cannot be read across realms, so presence is probed instead: any realm holding the stage forces the fork path. A deferred asset-pool deletion no longer reloads the page. The database is still on disk and the guidance asks the user to close the other tab and retry, which the reload discarded. The decision moved into a shared helper so the rule is pinned rather than living inline in the component. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> * fix(media): fail closed when presence cannot be probed The presence helper documented that an unanswerable probe must count as a peer, but every unavailable path — no BroadcastChannel, a constructor that threw, a send that failed, and the window before the pool's asynchronous binding completes — returned false, so the ownership proof cleared a single local owner and replaced globally shared bytes in place. Probing now returns present, absent or unknown, and only a probe that was actually sent and went unanswered is absent; the ownership decision treats unknown exactly like present. The pool declares its binding intent synchronously so a probe issued during the load-time window waits for the bind instead of concluding that presence is unavailable, and releases that gate if the import fails. Coverage reaches the write boundary: with presence unknown, a production-shaped targeted retry forks to a fresh id instead of calling replace, and the original bytes stay intact for a peer's unflushed owner. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> * chore(dsl): bump to 0.6.3 after the release dedupe took 0.6.2 Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> * fix(i18n): add the deferred-clear guidance to fr-FR The locale landed on main after this branch added the key, so the alignment check flagged it as the one missing translation. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 5 <noreply@anthropic.com> Co-authored-by: 杨慎 <117187635+cosarah@users.noreply.github.com> |
||
|
Percyandwyuc
|
d895b8959e |
feat(classroom): route generate-classroom per stage instead of one model (#1048)
The one-shot generateClassroom path (skill API) resolved the model once
with `resolveModel({ stage: 'generate-classroom' })` and bound that
single `languageModel` into one `sceneAiCall` closure shared by every
scene-content and scene-actions call (and the classroom `aiCall` shared
by outline + agent profiles). As a result, none of the `MODEL_ROUTES`
entries for `scene-content`, `scene-content:<type>`
(slide/quiz/interactive/pbl), `scene-actions`, or `agent-profiles` had
any effect on this path — the browser UI already routes each stage
independently via the /api/generate/* endpoints, but the one-shot API
did not.
Resolve the model per stage instead, mirroring the browser UI:
- scene-content routes via the composite key `scene-content:<type>`,
falling back to the base `scene-content` route (same resolution as
/api/generate/scene-content). Interactive widget sub-types are not
split, only the four core scene types.
- scene-actions routes via the `scene-actions` stage.
- agent-profiles routes via the `agent-profiles` stage (matches
/api/generate/agent-profiles).
- outline generation keeps using the classroom model, unchanged.
Each stage is resolved lazily and only when a route is actually
configured, so unrouted deployments reuse the classroom model at zero
extra cost. A route that fails to resolve (e.g. an unknown provider)
degrades to the classroom model with a warning, never aborting
generation — the same handling already used for
`web-search-query-rewrite`.
This also fixes a latent bug on this path: PBL scene content generation
needs the resolved `LanguageModel` object (generatePBLSceneContent
drives its own LLM calls through it, not the aiCall closure), but the
call never passed it, so PBL scenes silently returned null and were
skipped. The browser UI already passes it; now the one-shot path does
too — and it also forwards the route's thinking config, so a
`scene-content:pbl` route with a `thinking` block is honored (not just
the model).
The `generate-classroom-scene` telemetry/billing label is preserved on
all scene calls to avoid changing existing logging and accounting;
agent-profiles keeps the `generate-classroom` label.
Co-authored-by: wyuc <wang-yc24@mails.tsinghua.edu.cn>
|
||
|
zhifu gaoandLauraGPT
|
2b3f2eaa37 |
feat(asr): add local FunASR provider (#1044)
Signed-off-by: LauraGPT <199134975+LauraGPT@users.noreply.github.com> Co-authored-by: LauraGPT <199134975+LauraGPT@users.noreply.github.com> |
||
|
littlebullGitandwyuc
|
3d85525665 |
[codex] Add Amazon Bedrock LLM provider support (#538)
* Enable AWS-hosted text models without exposing ambient credentials Rebase the Bedrock provider onto current main and close the review-requested security, credential lifecycle, usage attribution, configuration, and metadata gaps. Constraint: Bedrock may use ambient AWS credentials only when explicitly enabled by the server operator Rejected: Trust client-supplied provider types | permits built-in keyless IDs to reach Bedrock credentials Confidence: high Scope-risk: moderate Directive: Keep provider ID/type validation at both request resolution and model construction boundaries Tested: targeted Bedrock/resolver/config/usage tests; TypeScript; ESLint; i18n alignment; production build Not-tested: Full suite has 27 failures in unchanged quiz/runtime and runtime/chat-storage tests * Give Bedrock a recognizable provider identity Use the official AWS architecture service icon so Bedrock no longer falls back to the generic provider cube in settings and model selection. Constraint: Preserve the AWS-provided artwork without redesigning the service mark Confidence: high Scope-risk: narrow Tested: SVG XML validation; provider unit test; TypeScript; ESLint; production build --------- Co-authored-by: wyuc <wang-yc24@mails.tsinghua.edu.cn> |
||
|
|
12ccd315ea |
Add Atlas Cloud LLM provider (#948)
* Add Atlas Cloud provider * Account for Atlas Cloud non-thinking model metadata * test: verify Atlas Cloud thinking payloads * style: format Atlas Cloud payload test * test: type Atlas Cloud fetch mock --------- Co-authored-by: binyangzhu000-sudo <224954946+binyangzhu000-sudo@users.noreply.github.com> Co-authored-by: wyuc <wang-yc24@mails.tsinghua.edu.cn> |
||
|
wyucandClaude Opus 5
|
97f8585728 |
refactor(pbl-v2): route runtime agents through the shared LLM entry point (#1006)
* refactor(pbl-v2): route runtime agents through the shared LLM entry point The five PBL v2 runtime call sites (instructor x2, evaluator, simulator x2) invoked the AI SDK directly while the other 21 call sites in lib/ and app/ go through callLLM / streamLLM, so they silently opted out of everything the shared entry point applies: - Usage accounting. recordUsage is reached only from the wrappers, so no instructor turn, evaluator pass, simulator line or narrator pass was ever accounted for - the highest-frequency, longest-context LLM traffic in the product. The comment at llm.ts:275, that every server-side call funnels through the wrappers, did not hold. - The LLM_THINKING_DISABLED kill switch, read inside the wrappers only. - One thinking resolution path. runtime-thinking.ts existed precisely because the agents bypassed the wrapper, and it was wired into instructor (x2) and evaluator but neither simulator site. Also fix callLLM recording result.usage, which on a multi-step tool run (stopWhen) is the final step alone; totalUsage is the aggregate, and the instructor turn is exactly such a run. streamLLM already preferred the aggregate. Thinking semantics are preserved per call site: the teaching turns keep force-disabling thinking, now through the wrapper's own `thinking` argument, and keep their existing providerOptions spread, which injectProviderOptions yields to; the simulator keeps honouring a per-request config, as before. The one behavioural difference is on native adapters when no thinking config arrives: the policy now resolves to an explicit lowest/disabled provider option instead of leaving the model's default in place - which is what the policy always claimed to do, and what OpenAI-compatible providers already got. Unifying the wider divergence is a product decision and stays out of scope; runtime-thinking.ts now documents it, along with the fact that the original tool_choice justification no longer matches the tree. A lint guard restricts importing generateText / streamText to lib/ai/llm.ts so the split cannot grow back. It uses the @typescript-eslint variant of no-restricted-imports deliberately: flat config replaces a rule's options per key rather than merging them, and the base rule is already configured for the lib/choreography and lib/video-export module boundaries. The runtime-thinking test moves from asserting the retired helper to asserting what the call site is responsible for, at the provider boundary: an evaluator turn reaches the provider with a disabled config in the thinking store, does not leak it, and is accounted for under its own usage source. Closes #1003 Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> * fix(pbl-v2): address cross-review findings on the shared-entry migration Consensus findings from an independent cross-vendor review of |
||
|
杨慎andClaude Opus 4.8
|
84b1907255 |
feat(video-export): service-backed MP4 render + in-app one-click export (#866) (#937)
* feat(video-export): service-backed MP4 render + in-app one-click export (#866)
Adds the last mile of classroom video export: turning the self-contained
Hyperframes project ZIP (#865) into an MP4 via an isolated render service,
one-click in-app.
- render-service/: standalone Node 22 + Chromium + FFmpeg container wrapping
@hyperframes/producer's library API. Async job model (POST /render -> 202
jobId, GET poll, GET download, DELETE cancel). Swappable JobStore /
ArtifactStore seams (in-memory + local-disk now; Redis/S3 + presigned-302
download later) so it scales horizontally without changing the HTTP contract.
Concurrency + per-user guards are config knobs.
- App integration: thin Next proxy routes under app/api/export-video/* (forward
only, no rendering) + capability probe. use-render-video.ts uploads the ZIP,
polls via runPolledTask, downloads the MP4; shared buildExportZip prefix with
the existing ZIP path. Export menu gains resolution/fps/quality selectors and
a progress bar; degrades to ZIP download when RENDER_SERVICE_URL is unset.
- docker-compose: render-service under an opt-in "video-export" profile.
- Entry is main.ts (not server.ts): the producer auto-starts its own server on
:9847 when the process entry path ends with /src/server.ts.
Verified end-to-end in the container: rendered a real 640s (10.7 min) classroom
ZIP to a valid H.264 720p + AAC MP4 (duration matches source) in ~9.6 min
(~0.9x realtime, 4-worker frame capture). Degrade path, queued-cancel + cleanup,
and per-user 429 guard all exercised. pnpm check / lint / tsc / i18n pass.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* feat(video-export): global render progress store, percent+ETA UI, ring on export button (#866)
Addresses two UX issues found while driving the in-app MP4 export:
1. Progress display was raw and unfriendly (showed producer's English stage
strings like "Capturing frame 5130/19220") and had no time estimate. Now the
menu shows only "<percent>% · about <remaining> left". ETA is computed from a
recent-speed estimate (percent-per-ms over the last sample), EMA-smoothed —
which tracks the render's non-uniform pace (prep -> frame capture with a
4->1 worker drop -> encode) far better than a whole-run average, and never
shows a stale/rising ETA.
2. Switching scenes mid-render unmounted the export menu and lost the progress
(and reset the local "already rendering" ref, allowing a duplicate submit).
The whole render lifecycle now lives in a global store
(lib/store/video-render.ts), so progress survives menu close / scene switch
and duplicate submits are guarded by status. A persistent CircularProgress
ring on the export button shows live progress whether or not the menu is open.
Also fixes the progress scale: the producer reports progress as 0..100, but our
HTTP contract (and success path) is 0..1 — the service now normalizes it, so the
client no longer showed "2000%".
- lib/store/video-render.ts: new global store owning submit->poll->download,
recent-speed ETA, duplicate-submit guard.
- lib/video-export-app/use-render-video.ts: thin facade over the store.
- components/ui/circular-progress.tsx: lightweight SVG progress ring.
- components/stage/{header-controls,video-export-menu}.tsx: ring on the export
button; menu shows percent + ETA, subscribes to the store.
- render-service/src/render-manager.ts: normalize producer progress 0..100 -> 0..1.
- i18n: percent/ETA strings across all 8 locales (drops the stage-based string).
- render-service/package-lock.json: complete integrity hashes (reproducible npm ci).
Verified: ETA logic checked against the real segmented render curve (worker drop
raises ETA, encode speedup drives it to ~0); progress scale fix confirmed live
against the container (0.2 -> 20%). tsc / lint / prettier / i18n pass.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(video-export): persist render options in the store, not the menu component
Selecting 720p/24fps/draft, switching scenes, and reopening the export menu
showed the defaults again (1080p/30/standard). The selections lived in the
VideoExportMenu component's local state, which reset when the menu unmounted on
a scene switch — the running render still used the chosen options, but the UI
misrepresented them.
Move resolution/fps/quality into the global video-render store (with a
setOptions action). The menu now reads/writes the store, so selections survive
menu close / scene switch, and while a render runs the selectors reflect the
options that render is actually using. startRender() reads options from the
store instead of taking them as an argument.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(video-export): deployment correctness + resource/isolation controls (PR #937 review)
Addresses the blocking findings from wyuc's review. Output fidelity was fine;
these harden deployment and production resource/isolation boundaries.
#1 Compose advertised MP4 but couldn't render in prod:
- Capability now probes the service's /health (checkRenderServiceHealth), so a
configured-but-absent service reports disabled and the UI degrades to ZIP
instead of 502-ing.
- RENDER_SERVICE_URL is operator-supplied trusted config, so the proxy no longer
runs it through the SSRF guard — the one-command `docker compose --profile
video-export up` now works without globally weakening SSRF via
ALLOW_LOCAL_NETWORKS. resolveRenderServiceUrl() is now synchronous.
- Client degrades to ZIP on any failed submit (not only 501).
#2 Unbounded upload/queue (ZIP-bomb / DoS):
- unzip.ts bounds the archive via fflate's filter BEFORE decompression: entry
count, per-entry and total expanded size, and compression ratio.
- Proxy rejects oversized uploads (413) by Content-Length before forwarding.
- RenderManager enforces a global queue-depth cap (RENDER_MAX_QUEUE).
- All limits are env-tunable knobs in config.ts.
#3 Per-user guard was ineffective + admission ran after extraction:
- Identity is derived server-side (client IP) and forwarded as x-openmaic-client;
the service ignores any client-supplied userId, and the proxy strips it.
- Admission is split into reserve()/submit()/release(): the slot is reserved
BEFORE extraction, so a rejected caller never triggers a decompression.
Additional risks:
- Per-job wall-clock watchdog (RENDER_JOB_DEADLINE_MS) aborts + fails a hung
render so it can't hold a slot/scratch forever.
- Download proxy bounds only the time-to-headers, not the body stream, so large
MP4s over slow links no longer truncate.
- Client cancels the server job (DELETE) when a started render fails/times out.
- Compose puts render-service on an internal:true network (no host/internet
route), sandboxing the Chromium that runs the uploaded HTML; the export ZIP is
self-contained so no outbound is needed. README documents the standalone caveat.
Not closing #866: the smoke/golden-render CI acceptance criterion remains a
follow-up (see PR description).
Verified in-container: legal render 202; ZIP-bomb (entry-count + compression-
ratio) rejected 400 before any decompression; per-identity guard 429 with a
spoofed multipart userId ignored; reserve-before-extract leaves no scratch dir
on rejection; watchdog aborts an overrunning job and frees the slot. tsc / lint /
prettier / i18n pass; render-service tsc passes.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(video-export): real audio durations + burned-in subtitles (PR #937 review)
Two export-fidelity issues found in wyuc's deeper E2E:
A. Narration was scheduled from estimated durations, cutting audio off mid-
sentence and advancing the timeline early. The scheduler trusted
AudioFileRecord.duration (recorded only since #861), so the many existing
classrooms without it fell back to text-length estimates — measured 4.35s
average / 10.23s max underestimate across 47 clips. timeline-deps now probes
the real duration from each narration blob via an off-document <audio>
(symmetric to the existing video probe), preferring it over the stored
duration, then the estimate only when no audio asset exists. Everything
downstream (narration starts, scene/total duration, subtitle cues) re-derives
from the corrected value in the pure compiler — no compiler change needed.
B. The final MP4 had no subtitles (only H.264+AAC), and the ZIP's SRT/VTT used
the same estimated boundaries. The emitter now renders a burned-in subtitle
overlay: one caption box + a hidden div per cue, revealed/hidden by the paused
GSAP timeline at each cue's start/end (corrected timings from A), so Chromium's
frame capture bakes them in. The producer has no subtitle track of its own, so
burn-in is the v1 approach.
Verified: emitter unit tests + snapshot updated (subtitle overlay + toggle
statements, escaped text, hidden-by-default); 82 video-export tests pass incl.
the determinism red-line proxy. Rendered a synthetic subtitle project through the
container and confirmed by pixel analysis that captions appear only within their
cue window (2429 near-white px in the caption band at t=1.5s vs 0 at t=0.05s).
tsc / lint / prettier / i18n pass.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(video-export): subtitle layout + upload/admission hardening (PR #937 review)
Address the three P1 blockers plus actionable P2s from the
|
||
|
Yizuki_Ameandwyuc
|
7c2aaafc91 |
fix(ssrf): harden provider redirects and ISATAP detection (#928)
* fix(ssrf): detect private IPv4 in ISATAP addresses * fix(ssrf): reject redirects in provider probes --------- Co-authored-by: wyuc <wang-yc24@mails.tsinghua.edu.cn> |
||
|
Yanpeng Wangandwyuc
|
9670100363 |
feat(extraction): media (audio/video) extraction + AliDocMind provider (#887)
* feat(extraction): add AliDocMind provider + media extraction abstraction Adds AliDocMind (Aliyun Document Mind LLM version) as a new vendor alongside unpdf/MinerU, and introduces the media (audio/video) extraction layer that mirrors the document extraction one. Document side (file: pdf/docx/pptx/xlsx/images): - PDF_PROVIDERS gains an `alidocmind` entry; parseWithAliDocMind() maps layouts[] -> ParsedPdfContent. Flows through the existing document extractor registry, so AliDocMind is selectable anywhere MinerU is. - PDFParserConfig / DocumentExtractorConfig gain accessKeyId/accessKeySecret (AliDocMind uses AK/SK, not a single apiKey); env fallback via ALIDOCMIND_ACCESS_KEY_ID / ALIDOCMIND_ACCESS_KEY_SECRET. Media side (mp4/mp3/wav/... -> MediaArtifact): - New lib/media-parse/ domain mirroring lib/pdf/ (types/constants/providers). parseMedia() maps AliDocMind segments[]/audio_frames/video_frames -> MediaArtifact (transcript + keyframes). - MediaExtractorProvider interface + media registry + extractMedia() entry, symmetric to DocumentExtractorProvider / extractDocument(). - MediaArtifact and the ExtractionResult/Artifact/Error/Job envelope live in lib/document/types.ts (re-exported from @/lib/document). Shared AliDocMind SDK wrapper (lib/pdf/alidocmind-client.ts) handles the submit -> poll -> get flow for both sides via @alicloud/docmind-api20220711. Tests: env-gated smoke test (tests/document/alidocmind.smoke.test.ts) drives a real PDF and a real video through AliDocMind; media-artifact type test; extractor-registry test updated for the new provider. Part of #621 (MAIC ETL). Media extraction is the sibling to the document extraction landed in #741. * feat(extraction): wire AliDocMind AK/SK through settings UI + routes Surfaces AliDocMind in the (post-#837) Document Parsing settings panel as a peer of unpdf/MinerU — one panel, one credential entry, more supported formats. Threads Aliyun AccessKey ID/Secret from the store through the extraction routes. - Store: pdfProvidersConfig + setPDFProviderConfig gain accessKeyId/accessKeySecret; default alidocmind entry added. - pdf-settings.tsx: AliDocMind branch renders AccessKey ID + Secret inputs (secret masked with show/hide) and a Test Connection button; request-URL preview shows the DocMind endpoint. Provider auto-appears in the panel via PDF_PROVIDERS; supported-format badges come from #837's registry (ALIDOCMIND_MIMES added to lib/document/mime.ts). - verify-pdf-provider route: alidocmind branch verifies AK/SK via a lightweight authenticated probe (verifyAliDocMindCredentials) — auth-level errors fail, anything else passes. - extract-document route + generation flow (app/page.tsx, generation-preview): accessKeyId/accessKeySecret carried through session → FormData → config. - i18n: alidocmindAccessKeyId / alidocmindAccessKeySecret in 8 locales. - Icon: reuse /logos/bailian.svg (Aliyun family) instead of a missing asset. Verified end-to-end against the running app: AliDocMind panel renders, Test Connection returns "连接成功" through the real Aliyun API. Part of #886. * feat(extraction): route audio/video uploads through extractMedia() (reuse document path) Media uploads now flow through the same upload picker, /api/extract-document route, and generation pipeline as documents — no separate upload area. Only the extraction differs: media mimes dispatch to extractMedia() -> MediaArtifact, which is flattened to the text shape the generation pipeline already consumes. - mime.ts: register audio/video formats in DOCUMENT_FORMATS (accept string, extension map, badges resolve for them); add MEDIA_PROVIDER_SUPPORTED_MIME_TYPES + SUPPORTED_MEDIA_MIME_TYPES, kept separate from PROVIDER_SUPPORTED_MIME_TYPES so the document drift-guard stays document-only. mimesForProviders() folds in a provider's media mimes, so the existing upload helpers (getAcceptStringForProviders / isMimeSupportedByProviders / format badges) cover media automatically when the provider supports it. - extract-document route: media mimes dispatch to extractMedia(); MediaArtifact flattened to timestamped text (synopsis + transcript + keyframes). - Tests: 6 media cases in mime.test.ts (accept/validation/badges/normalization). Verified: uploading a real video through the route returns synopsis + timestamped transcript/keyframes as text (curl, real AliDocMind key). Part of #886. * feat(extraction): use Alibaba Cloud icon for AliDocMind provider Replace the placeholder bailian.svg with a dedicated Alibaba Cloud icon mark (the square symbol from the official wordmark, text removed) so the provider reads as Aliyun rather than Bailian. Square aspect matches the other provider icons. Source: Alibaba Cloud / Alibaba Group brand assets. * style: prettier format AliDocMind + media extraction files * fix(extraction): harden AliDocMind — SSRF guard, cred verify, env gate, table text Addresses review findings on the AliDocMind provider: - SSRF (high): the media branch of /api/extract-document now runs the same validateUrlForSSRF check on a client-supplied baseUrl as the document branch, so an audio/video upload can't point the server's Aliyun SDK at an internal host. - Credential verify (high): verifyAliDocMindCredentials now whitelists success signals instead of blacklisting auth errors. Probed against the real API: valid creds + bogus job returns a no-throw "BizIdNotExistOrResultExpired" body; invalid creds throw InvalidAccessKeyId.NotFound. Only a no-throw response or a "biz-not-found" business error counts as valid — an unreachable endpoint, a localized error, or throttling now correctly reports failure instead of a false "connection successful". - Env-fallback gate (med): resolveCredentials no longer reads ALIDOCMIND_ACCESS_KEY_ID/SECRET unconditionally. Env fallback is opt-in via allowEnvFallback, which the route sets only for a server-managed provider, so an unauthenticated client request can't silently run on the server account. - getDocParserResult error body (med): fetchResult throws on a non-200 result envelope instead of returning {} (empty text presented as success). - Media pagination (med): stop after the first page when segments[] are present — layoutNum/layoutStepSize address layout blocks, not media segments, so re-requesting would loop over the same segments to the safety cap. - Table content (med): tables/charts carry content in llmResult, not markdownContent; the layouts→text mapping now prefers llmResult for those types so table content isn't dropped (tables: true was advertised). - Dead code: remove getCurrentMediaParseConfig (referenced non-existent store fields; the single-panel UI reuses pdfProvidersConfig). - Dedup: media MIME list lives only in lib/document/mime.ts (ALIDOCMIND_MEDIA_MIMES); the media registry imports it. - Tests: smoke test asserts durationMs is in ms (>1000 for a ~52s clip) to guard a ms/s unit mismatch; uses allowEnvFallback for the env-cred path. Verified with the real key: PDF + video smoke tests pass; verify classifies valid vs invalid creds correctly. Part of #886. * feat(extraction): extract AliDocMind images to base64 (parity with unpdf/MinerU) AliDocMind embeds figure/picture image URLs inside each layout's markdownContent (markdown ``), not a dedicated field, and the URLs are short-lived OSS signed links. Previously we emitted `images: []` and left the expiring URLs inside the extracted text. Now, for figure/picture layouts we: - parse the OSS image URL out of markdownContent, - fetch it at extraction time (before the signature expires) and re-encode to PNG base64 via sharp — the same base64 `images[]` contract unpdf/MinerU produce, so downstream storeImages → IndexedDB → slide works unchanged, - populate metadata.pdfImages + imageMapping (the generation flow prefers pdfImages), - strip the remote-URL markdown from the emitted text so expiring links don't leak into the prompt. Downloads run concurrently; a failed/`sharp` image is dropped, never failing the whole parse. Also fixes the prior over-broad table handling: only `table` blocks read llmResult; `figure` is treated as an image (chart-figure llmResult still kept in text). Verified with the real key: a sample PDF yields 24 base64 images in both images[] and metadata.pdfImages, and no oss-cn-hangzhou URLs remain in text. Part of #886. * docs(test): note AliDocMind video smoke test is non-deterministic server-side * fix(extraction): correct AliDocMind pageCount (pageNum is 0-based) Verified against a real response: AliDocMind reports pageNum 0..13 and pageCountEstimate 13 for a 14-page document — both are 0-based. The metadata pageCount previously used pageCountEstimate directly, undercounting by one. Use the already-1-based maxPage, falling back to pageCountEstimate+1 only when no blocks were seen. Document the 0-based convention at the normalization site. * fix(extraction): address AliDocMind review — verify/SSRF/config/media (P1+P2) Resolves all P1/P2 findings from the cross-review on #887. P1 (blocking): 1. verifyAliDocMindCredentials now inspects the no-throw response body.code. An OSS-only key returns NoPermission without throwing; previously that was green-lit as "connection successful" then failed at extraction. Only a success/200 or the job-not-found probe code is accepted. Deterministic mocked test added (tests/document/alidocmind-verify.test.ts). 2. verify route trust boundary: managed → server-owned AK/SK + default endpoint only (ignore client values); unmanaged → client creds only, never env fallback, and the client endpoint is SSRF-validated before signing. 3. image fetch hardened: restricted to Aliyun OSS hosts, redirects disallowed, per-image byte cap, image-count cap, bounded concurrency (was unbounded Promise.all over provider-returned URLs). 4. AliDocMind is now selectable in the generation toolbar — availability recognizes the AK/SK pair, not just apiKey. P2 (correctness): 5. Explicit server-config for the AK/SK pair (applyAliDocMindFallback + resolveManagedAliDocMindCredentials); verify and extract now resolve managed/env identically instead of verify-uses-env / extract-rejects. 6. Poll loop checks body.code before status, so a body-level error (e.g. NoPermission) fails fast instead of retrying for the full 15 min. 7. Image page numbers preserved through fetch/filter — no longer hard-coded to page 1, so multi-page image→page association is correct. 8. Empty media extraction (no synopsis/transcript/keyframes) returns 422 PARSE_FAILED instead of HTTP 200 with empty text. 9. Format matrix trimmed to the official contract: images JPG/JPEG/PNG/BMP/GIF (dropped WebP/JP2), media MP4/MKV/AVI/MOV/WMV/MP3/WAV/AAC (dropped M4A). 10. A document-only provider (unpdf/mineru) uploaded with a media file now returns a clear 4xx instead of an opaque 500. P3: credential-verify failures return INVALID_CREDENTIALS 4xx (not INTERNAL_ERROR 500); formatTimestamp emits HH:MM:SS past one hour. Verified with the real key: verify classifies valid→ok and invalid→fail; PDF (24 images, correct page numbers) and MP4 extraction pass end-to-end. * fix(extraction): make AliDocMind selectable + correct analysis label for media Two UI/UX fixes found while manually testing the AliDocMind flow end-to-end: - Persisted-state backfill: add ensureBuiltInPDFProviders so a PDF/document provider added after a user's settings were persisted (AliDocMind) is backfilled into pdfProvidersConfig on rehydrate. Without it the provider never appeared in the store, so it couldn't be selected and never picked up its server-configured flag. Wired into both persist migrate() and merge(), mirroring the existing image/video/web-search backfills. - Analysis step label: getGenerationStepText showed "解析 document 文件" for audio/video (the type map fell through to the literal "document"). Documents keep their precise token (PDF/DOCX/PPTX/XLSX/images); audio/video now use a dedicated, locale-correct string (generation.analyzingMediaMaterial, added to all 8 locales) instead of forcing a format token into the "{{type}} 文件" template. Manually verified end-to-end (real key): PDF, MP4, and MP3 (audio extracted from the sample video) each extract and drive full course generation; a DocMind-restricted AK/SK (OSS-only) now correctly fails verification with NoPermission (400) instead of a false "connection successful". * fix(extraction): address 2nd-round AliDocMind review (managed creds, stream cap, empty verify) Resolves the three follow-up findings on #887: 1. [P1] YAML-managed AliDocMind creds now reach extraction. Both extract paths (document + media) previously cleared the managed AK/SK and relied on an env-only fallback, so a YAML-only deployment verified but failed to extract. They now resolve server-owned creds via the shared resolveManagedAliDocMindCredentials() (env OR YAML), matching the verifier. Regression test added for YAML-only creds with no ALIDOCMIND_* env. 2. [P1] Image download is now size-capped while streaming. Instead of buffering the whole response and checking length afterward, the body is read chunk by chunk with a cumulative byte count that aborts the moment it exceeds the cap — a missing/false Content-Length can no longer exhaust memory. Host allowlist tightened to oss-*.aliyuncs.com. Tests: non-OSS host refusal, no-Content-Length overflow abort, declared-oversize rejection. 3. [P2] Empty verification body no longer counts as success. Removed the codeStr === '' branch from the positive-signal whitelist (a working key always returns the job-not-found business code for the bogus probe id). Regression test added for empty and absent bodies. Rebased onto main (package.json: kept @openmaic/storage + AliCloud deps). Verified with the real key: PDF + MP4 extraction still pass end-to-end. * fix(extraction): merge AliDocMind AK/SK into a baseUrl-configured YAML entry Follow-up to the YAML-managed credential fix. When a YAML `pdf.alidocmind` entry also specifies `baseUrl`, the generic loadEnvSection() (pdf requires a baseUrl) creates the pdf.alidocmind entry copying only apiKey/baseUrl/models/ proxy — never AK/SK. applyAliDocMindFallback() then returned early because the entry already existed, so the provider was "managed" but had no usable credentials, and resolveManagedAliDocMindCredentials() returned undefined (verify + extract both silently lost the creds). Merge the AK/SK into the existing entry instead of returning early. Added a regression test with baseUrl + accessKeyId + accessKeySecret together (the previous test omitted baseUrl, so the generic loader skipped the entry and the bug was masked). * fix(extraction): don't mark AliDocMind managed without AK/SK; align poll code check Two edge cases from a follow-up adversarial pass: - [MED] A YAML `pdf.alidocmind` entry with `baseUrl` but no AK/SK (and no env AK/SK) made the generic loader create the entry → isServerConfigured=true (managed) → but resolveManagedAliDocMindCredentials() returned undefined, so the provider was locked out AND client-entered AK/SK were silently dropped. applyAliDocMindFallback now deletes a credential-less entry so the provider stays UNMANAGED (clients supply their own creds). Regression test added. - [LOW] verify accepted a `code: "success"` body but the extraction poll loop threw on any non-"200" code — a success-shaped status would pass verification then fail extraction. The poll loop now treats "200"/"success" as benign, matching verifyAliDocMindCredentials. --------- Co-authored-by: wyuc <wang-yc24@mails.tsinghua.edu.cn> |
||
|
|
c56929510c |
feat(web-search): add SearXNG provider support (#842)
* feat(web-search): add SearXNG provider support Enable self-hosted SearXNG via SEARXNG_BASE_URL, with server-side provider fallback, settings sync, and tests. Co-authored-by: Cursor <cursoragent@cursor.com> * style: fix Prettier formatting in web-search files Restore pnpm check by applying Prettier to imports, headers, and test assertions. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(web-search): keep SearXNG base URLs server-managed only Close the SSRF primitive by ignoring client-supplied SearXNG URLs at the route boundary and rejecting them in resolveSafeClientWebSearchBaseUrl. Add negative route/config tests and an opt-in live smoke test for the JSON API. Co-authored-by: Cursor <cursoragent@cursor.com> * style: fix Prettier formatting for SearXNG web-search changes Co-authored-by: Cursor <cursoragent@cursor.com> --------- Co-authored-by: 徐松 <song.xu@aieshanghai.com> Co-authored-by: Cursor <cursoragent@cursor.com> |
||
|
|
3b63710376 |
feat(ai): add Azure OpenAI provider (#916)
* feat(ai): add Azure OpenAI provider Add deployment-based Azure OpenAI configuration for both client and server-managed setups, normalize Azure portal endpoints, and expose the provider in settings. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: b0d3a94f-089a-470e-9987-608f1506ca6f * style: format Azure provider type Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Copilot-Session: b0d3a94f-089a-470e-9987-608f1506ca6f --------- Co-authored-by: MarshellOnMoon <121332425+MarshellOnMoon@users.noreply.github.com> |
||
|
Yizuki_AmeandClaude Fable 5
|
e136c81c03 |
chore(generation): remove dead legacy pipeline chain (#904) (#905)
Remove the generation-pipeline leftovers orphaned since the old
/api/generate route was deleted in
|
||
|
Yanpeng Wang
|
d54e62f5c4 |
feat(token-plan): one-click token-plan setup + deployment usage dashboard (#784)
* feat(usage): add usage normalization, pricing, model-fetch, balance, storage Foundational layer for token-plan usage tracking (cc-switch-modeled): - lib/usage/normalize.ts: AI SDK v6 usage → four-class token shape - lib/usage/pricing.ts + defaults: per-class USD pricing table - lib/server/model-fetch.ts: /models candidate-URL multi-fallback (ported) - lib/usage/balance-providers.ts: built-in balance queries + detection - lib/server/usage-storage.ts: fire-and-forget jsonl logging to data/usage All pure/storage logic covered by vitest (32 tests). * feat(usage): capture token usage at the callLLM/streamLLM chokepoint - callLLM records result.usage before returning - streamLLM wraps onFinish to record totalUsage on stream completion, preserving any caller-supplied onFinish - provider/model derived from the model instance (no route changes) - fs-backed storage imported dynamically; fire-and-forget, never throws - include_usage is already sent by @ai-sdk/openai, so streaming is covered * feat(usage): add probe-models, balance, and usage API routes - POST /api/provider/probe-models: discover chat models via /models with candidate fallback; SSRF-guarded; filters non-chat ids; 401/404 typed - POST /api/provider/balance: built-in balance detection + billing fallback - GET /api/usage: aggregate jsonl by model/day/source with costIncomplete flag Verified e2e against the live MAIC gateway: 16 chat models (6 filtered), balance detected, and a real callLLM writes a costed usage row. * feat(settings): add token-plan preset picker to provider dialog - lib/config/token-plan-presets.ts: data-driven vendor presets (Huawei/MiniMax/ Xiaomi token plans, OpenRouter/SiliconFlow gateways, DeepSeek/GLM/Qwen/Hunyuan/ Doubao direct) with baseURL, protocol, optional modelsUrl, category - add-provider-dialog: '选择厂商/自定义' tabs; picking a preset auto-fills baseURL+protocol+modelsUrl, custom tab unchanged - ProviderSettings.modelsUrl carries the optional /models override - i18n keys added across all 8 locales Verified in-browser: preset picker renders grouped by category. * feat(settings): add Fetch Models button and balance bar to provider panel - 拉取模型: probes /models, merges discovered ids into the model list (dedupe, keeps manual additions), with success/no-endpoint/auth messages - 查询余额: queries /api/provider/balance, renders a balance bar or a 'check console' hint when unsupported - index.tsx: handleModelsFetched merges probe results into provider config - i18n keys across all 8 locales; removed a now-unused eslint-disable Verified in-browser against MAIC gateway: 16 models fetched, balance shown. * feat(settings): add usage dashboard to System Settings - usage-dashboard.tsx: echarts dual-axis daily trend (tokens + cost), totals cards, by-model table, refresh; reads GET /api/usage - mounted at the top of GeneralSettings (系统设置) - honest disclaimer + costIncomplete marker when a model lacks pricing - i18n keys across all 8 locales Verified in-browser: shows 1 request / 31 tokens / $0.0003 from a prior call. * feat(token-plan): multi-modal one-click setup in System Settings - token-plan-presets.ts: presets now declare per-modality targets (llm/image/video/tts/webSearch); MiniMax is the full-set template, others LLM-only — extend by adding entries (at-our-best adaptation) - apply-token-plan.ts: fills one key into every declared modality via injected store setters, isolating per-modality failures (TDD, 4 tests) - token-plan-settings.tsx: new sidebar page — pick plan, enter key, one-click apply lights up adapted modalities (+LLM model probe), shows 'not adapted yet' for the rest, balance bar reused - reverted add-provider dialog to plain custom form (preset picker moved here) - i18n across all 8 locales Verified in-browser: Token Plan page renders, MiniMax shows all 5 modalities, apply/balance UI wired. 859 tests pass, build clean. * feat(token-plan): add custom token plan entry to Token Plan page - Custom card at the bottom of the provider list: pick it to manually enter name + protocol + baseURL (LLM-only), then key + one-click apply via the same applyTokenPlan flow (mirrors cc-switch's custom provider) - effectivePreset unifies preset vs custom for apply/balance/probe - i18n keys (customGroup/customName/customHint) across all 8 locales Verified in-browser: custom card expands the manual form; preset flow intact. * feat(token-plan): reflect persisted config on the Token Plan page Other settings panels read the store directly, so they survive section switches; the Token Plan page only wrote the store, so it looked blank on return. Now it also reads providersConfig: - selecting a preset prefills its saved API key - configured presets show a '已配置/Configured' badge State persistence was never broken (apply writes zustand→localStorage); this fixes the missing read-back so the page reflects it. * fix(settings): isLLMProviderConfigured crashed on providers without models Root cause of 'token plan config disappears': applyTokenPlan writes a new LLM provider with no models yet (probe fills them later). isLLMProviderConfigured did config.models.length unguarded → threw inside setProviderConfig's resolver → the whole set() aborted → key/baseUrl/type never persisted. - guard models in isLLMProviderConfigured (shared validator; affects any provider written without a models array) - seed models:[] in applyTokenPlan's LLM write for a valid initial shape - regression tests: validator no longer throws; apply+probe write keeps apiKey Verified in-browser: DeepSeek persists key and shows 已配置 after section switch. * feat(token-plan): add remove/teardown for a configured token plan Apply had no inverse. Add removeTokenPlan: clears the API key and disables (enabled:false) every modality the plan declared; a custom LLM provider is deleted entirely (removeProvider), built-ins keep the cleared shape. - removeTokenPlan in apply-token-plan.ts (mirrors applyTokenPlan, isolated per-modality, injected setters; 3 tests) - trash button on configured preset cards in the Token Plan page; resets page state if the removed plan was selected - i18n 'remove' across all 8 locales Verified in-browser: removing DeepSeek empties its key and drops the 已配置 badge. * feat(usage): track multimodal usage, drop cost/pricing Reframe usage stats as pure usage (no cost), per user decision: - usage-storage: drop all cost fields; add kind (llm/image/video/tts/asr) + quantity + unit; LLM keeps token counts, others store quantity - delete lib/usage/pricing.ts + pricing-defaults.json + its test - instrument image/video/tts at the server-only API routes (not in the provider dispatch — those files are in the client graph and importing fs-backed usage-storage broke the client bundle) - /api/usage: aggregate by model/day/modality, no cost - dashboard: per-modality usage (tokens/images/seconds/chars), token-only daily trend, no '$'; i18n cost keys replaced with modality/unit keys - backward compatible: legacy rows (no kind, stray cost fields) read as llm 912 tests pass, build clean. * feat(usage): per-modality dashboard layout + softer dark-mode chart - group usage into per-modality sections (LLM/image/video/tts/asr), each table's usage column uses one consistent unit (token/image/sec/char) — no more mixed units in a single column - summary chips per modality with their own unit - trend chart now plots daily REQUESTS (unit-agnostic, works for any modality) instead of LLM-only tokens - theme-aware chart: faint thin line + soft gradient area, muted axis/grid colors via useTheme — fixes the harsh solid stroke in dark mode Verified in-browser (dark): TTS shows '字符', LLM shows 'Token', separate sections; chart no longer has a hard line. * refactor(usage): dedupe model/fetch/usage helpers, parallel balance probe Review cleanups on the token-plan/usage branch: - extract modelInfoFromId() (shared vision heuristic + ModelInfo shape) - extract fetchWithTimeout() shared by model-fetch and balance-providers - extract recordGenerationUsage() to dedupe the image/tts/video routes - queryBalance: fetch billing subscription + usage in parallel - parseOneApiBilling: report quota without remaining when usage endpoint is unavailable, instead of implying zero spend (full balance) - split the two jammed imports in settings/index.tsx * feat(token-plan): drop custom token-plan support Custom token plans (manual baseURL/protocol entry) added complexity for little gain — a one-off provider is better configured directly on the Providers page. Token Plan is now preset-only: - remove custom mode, manual fields, and the custom card from the UI - collapse effectivePreset back to the selected preset - drop the now-dead removeProvider action + custom-id branch in removeTokenPlan - remove orphaned customGroup/customName/customHint i18n keys (8 locales) * feat(token-plan): add Volcengine/Tencent/Bailian plans, drop balance feature Add three vendor token-plan presets (all map to existing built-in LLM providers, so it's data-only — no new adapters): - 火山方舟 Volcengine Ark → doubao, OpenAI /api/v3 - 腾讯 TokenHub Token Plan → tencent-hunyuan, OpenAI /plan/v3 (the plan-specific base; /v1 is the pay-as-you-go gateway) - 阿里百炼 Token Plan → qwen, cross-model plan (Qwen + DeepSeek/Kimi/ GLM/MiniMax) on one key; model list is probed/entered Remove the balance/quota feature entirely — we now track usage, not cost, and every vendor's balance query needs its own cloud AK/SK + signature (Volcengine SigV4 / Tencent TC3 / Aliyun BSS), which the Bearer-key billing-endpoint probe never supported anyway: - delete lib/usage/balance-providers.ts, /api/provider/balance, its test - strip the Check Balance button + balance bar from the token-plan page and the provider config panel - remove the 4 balance i18n keys across all 8 locales - restore an eslint-disable the branch had dropped in provider-config-panel * feat(token-plan): use cloud-brand logos for vendor token plans The three vendor plans are cloud offerings, not single-model products, so icon them with the cloud brand rather than a model logo: - 火山方舟 → volcengine.svg (was doubao.svg) - 腾讯 TokenHub → tencentcloud.svg (was hunyuan.svg) - 阿里百炼 → alibabacloud.svg (was bailian.svg) Logos are the colored brand variants from lobehub/lobe-icons, matching the existing colored-logo style (plain <img>, no dark:invert needed). * feat(token-plan): keep only MiniMax and Volcengine presets Trim the token-plan list to the two we want to ship: MiniMax (full-set template) and 火山方舟 Volcengine Ark. Drop the Tencent/Bailian plans and the OpenRouter/SiliconFlow/DeepSeek/GLM/Qwen entries. - remove the now-unused tencentcloud.svg / alibabacloud.svg logos (volcengine.svg stays; the other logos are still used by the provider registry) - retarget the LLM-only apply test from the deleted deepseek preset to volcengine-ark * feat(token-plan): restore aggregator/third-party presets Previous commit over-trimmed: the intent was to drop only the Tencent and Bailian token plans, not the OpenRouter/SiliconFlow/DeepSeek/GLM/Qwen entries. Bring those back; keep only Tencent/Bailian removed. - token_plan: MiniMax, 火山方舟 Volcengine Ark - aggregator: OpenRouter, SiliconFlow - third_party: DeepSeek, GLM, Qwen Revert the apply test back to the deepseek fixture (restored). tencentcloud.svg / alibabacloud.svg stay deleted (their plans are gone). * fix(token-plan): point Volcengine plan at the Coding Plan endpoint The plan's ark--prefixed API keys authenticate only against /api/coding/v3, not the general /api/v3 endpoint — the latter rejects them with "The API key format is incorrect", so model probing returned nothing. Switch the base URL to https://ark.cn-beijing.volces.com/api/coding/v3. * fix(token-plan): Volcengine is an Agent Plan (Anthropic /api/plan) Per the Ark Agent Plan docs, the ark--prefixed keys authenticate ONLY against the dedicated Anthropic-compatible base https://ark.cn-beijing. volces.com/api/plan ("其他 Base URL 无法在 Agent Plan 中使用"). The general /api/v3 and the Coding Plan /api/coding endpoints both reject the key as "API key format is incorrect", which is why model probing kept returning 0. - baseUrl → https://ark.cn-beijing.volces.com/api/plan/v1 (the /v1 lets the Anthropic SDK land on /api/plan/v1/messages) - apiFormat → anthropic - rename to 火山方舟 Agent Plan Probe still targets /api/plan/v1/models (the path exists); if the Anthropic gateway doesn't return an OpenAI-shaped list, users fall back to typing a model id like ark-code-latest. * fix(token-plan): Volcengine Agent Plan = OpenAI /api/plan/v3 + ark-code-latest Settled after probing the real key and reading cc-switch's approach: - The ark- plan key works on the OpenAI-compatible /api/plan/v3 endpoint (chat/completions returns 200); switch apiFormat back to openai. - The plan exposes NO /models list (every /api/plan/*/models is 404), which is why probing kept returning 0. cc-switch handles this by hardcoding a single ark-code-latest (an auto-routing alias valid on any tier) and does NOT use AK/SK for model listing — so we do the same. - Seed defaultModels: ['ark-code-latest'] only; users add specific ids by hand. Supporting machinery (kept, general-purpose): - applyTokenPlan seeds models from defaultModels instead of wiping to [] - handleApply uses defaultModels and skips the doomed probe when present - drop stray .playwright-mcp/ debug artifacts and gitignore them * style: fix prettier formatting in usage files CI runs prettier on the whole repo (prettier . --check); these four files predate this branch's formatting pass and tripped the check. * feat(token-plan): verify Volcengine Agent Plan's published model set The Agent Plan publishes a fixed model set but exposes no /models endpoint, so carry the documented models as CANDIDATES and verify each on apply: - add verifyModels flag to TokenPlanModalityTarget - new /api/provider/probe-chat-models route: sends a minimal chat request per candidate (OpenAI /chat/completions or Anthropic /messages) in parallel, returns the subset that succeeds; SSRF-guarded, auth-failure short-circuits - handleApply gains a verify branch (before the fixed-defaultModels fast path), falling back to the seeded list if verification fails - Volcengine preset now carries the 12 published Agent Plan text models (doubao-seed-2.0-*/deepseek-v4-*/minimax-m*/glm-5.2/kimi-k2.*) as candidates This auto-prunes retired (docs flag deepseek-v3.2/glm-5.1 as 即将下线) and tier-gated models without code changes. Verified all 12 resolve against a real plan key. * feat(token-plan): wire Volcengine Agent Plan image + video modalities Make the Ark seedream/seedance adapters path-configurable and light up the image/video modalities on the Volcengine plan: - seedream/seedance adapters: resolveArkRoot() uses baseUrl verbatim when it already carries an /api/... path (token plan's /api/plan/v3), else appends the standard /api/v3 — no regression for the pay-as-you-go default host. - applyTokenPlan: image/video branches inject a modality's defaultModels as customModels and set them as the active provider+model, so generation works out of the box. New optional setImageProvider/ModelId + setVideoProvider/ ModelId actions (UI passes the store setters; tests omit them). - Volcengine preset declares image (doubao-seedream-5.0-lite, verified 200 on /api/plan/v3/images/generations) and video (doubao-seedance-2.0/1.5-pro — Medium+ tiers only; lower tiers reject at call time, no code change needed to upgrade). Applying the plan overwrites the shared seedream/seedance slot with the plan config (same overwrite model as LLM); switching back to pay-as-you-go is a manual edit or plan removal. Verified image end-to-end with a real plan key. * feat(token-plan): verify image/video models on apply, disable unsupported tiers The Volcengine plan lit up video optimistically, but lower tiers (Small) don't include video — so using it 404'd with UnsupportedModel. Probe media models on apply and only keep what the tier actually supports: - generalize /api/provider/probe-chat-models with a `kind` (chat|image|video): image hits /images/generations, video hits /contents/generations/tasks with empty content. The model-support check (404 UnsupportedModel) runs before any billable work, so probing never starts a real image/video job; for media, "supported" = any non-404 response. - handleApply: after lighting up image/video, probe each verifyModels modality; prune to the verified model set + re-select a working model, or disable the modality entirely if none pass (no false "available"). - Volcengine preset: image/video targets gain verifyModels: true. - add settings.tokenPlan.tierUnsupported across 8 locales. Verified with a real Small-tier key: image (seedream-5.0-lite) passes and is kept; video (seedance-2.0/1.5-pro) 404s and is disabled. * feat(web-search): add Doubao (豆包搜索) provider Doubao Search (Custom 版) over its REST endpoint POST open.feedcoopapi.com/search_api/web_search with Bearer auth — the same endpoint the askecho-search-infinity MCP server wraps, so the Volcengine Agent Plan key authenticates directly. Mirrors the MiniMax adapter: maps Result.WebResults to WebSearchSource (prefers Summary, the query-relevant excerpt, over Snippet for LLM use) and surfaces errors from ResponseMetadata.Error. - register 'doubao' in WebSearchProviderId + WEB_SEARCH_PROVIDERS - searchWithDoubao adapter, searchWeb dispatch, store default config - SSRF allowlist entry for the search host * feat(audio): support Agent Plan single-key auth for Doubao TTS generateDoubaoTTS now picks auth + endpoint from the key shape, since Volcengine exposes Seed-TTS as two products with separate credentials (verified: a plan key 401s on the normal endpoint, and the plan endpoint rejects appId-style auth): - single key (no colon) -> X-Api-Key, for the Agent Plan /plan endpoint - appId:accessKey -> X-Api-App-Id + X-Api-Access-Key (unchanged) A malformed pair (empty half) fails clearly instead of sending an empty header. Reuses the existing NDJSON/base64-mp3 parsing and voice list. * fix(media): map MiniMax video 720p to its real 768P tier normalizeVideoOptions defaults minimax-video to '720p' (the first supported resolution), but Hailuo 2.3 only accepts 768P/1080P and rejects 720P with '2013 ... does not support resolution 720P'. MiniMax's mid tier is 768P, not 720P (the adapter already falls back to 768P, as does the connectivity test), so map the shared enum's '720p' to 768P. Regression tests lock the mapping. * feat(token-plan): add web search + TTS to Volcengine Agent Plan, widen image tiers Extend the volcengine-ark preset now that the adapters exist: - webSearch -> doubao (own host open.feedcoopapi.com, not the ark endpoint) - tts -> doubao-tts on the /api/plan/tts endpoint (single-key auth) - image defaultModels widened to a best-first Seedream 5.0/4.5/4.0 list so a higher tier keeps the strongest model while verifyModels prunes the rest; video keeps the 2.0 + 1.5-pro candidates Comments record the verified host/auth quirks of each modality. * feat(token-plan): show result panel only after probing, with two clear states Addresses review feedback that a green check implied generation works when it only meant 'configured'. The panel now renders after probing finishes (gated on results && !applying) so it reflects the final set, and uses two states: green when the modality is configured/usable, muted when a live probe proved it unavailable (e.g. video on a tier without it). * feat(token-plan): scope presets to true multi-modal token plans Drop the single-modality LLM presets (OpenRouter, SiliconFlow, DeepSeek, GLM, Qwen) from Token Plan. A token plan's defining trait is one key spanning many modalities; those entries are ordinary LLM API providers already covered by the add-provider flow, and listing them here muddied the 'one key, every modality' promise. Only MiniMax and the Volcengine Ark Agent Plan remain. The UI already hides categories with no entries. apply-token-plan's LLM-only test now uses a local fixture instead of the removed deepseek preset. * feat(token-plan): progressive reveal of probe results on apply The result panel previously rendered all at once after probing finished, reading as dead air during the model probe. Now rows appear immediately on Apply: modalities with a live probe in flight show a spinner ('pending') and resolve to lit/failed independently as each probe returns, while non-probe modalities show lit right away. Probes run in parallel (Promise.all) instead of sequentially. A row only turns green once its own probe confirms, so this reveals structure + live progress without a premature green — complementing the earlier 'render only after probing' intent rather than reverting it. Per review feedback from @wyuc on #784. * fix(token-plan): enrich seeded models with built-in thinking capability Token Plan built ModelInfo objects from probed ids via modelInfoFromId(), filling only streaming/tools/vision — so a model that supports configurable thinking lost capabilities.thinking and InlineThinkingControl was hidden. modelInfoFromId now takes an optional providerId and overlays the catalog thinking capability for that (provider, model) pair; applyTokenPlan does the same for its synchronously-seeded list. Added the Ark Agent Plan's dotted aliases to the metadata table: - native Doubao Seed 2.0 family (doubao-seed-2.0-pro/code/lite/mini) - cross-vendor models the plan serves through its OpenAI-compatible endpoint (deepseek-v4-pro/flash, glm-5.2, kimi-k2.7-code/k2.6, minimax-m3/m2.7, ark-code-latest) All verified against a live plan key: each accepts the gateway's unified reasoning_effort field (low/medium/high) and actually reasons. They share the doubao effort adapter, which disables via 'minimal' (not 'none') — matching what the plan endpoint accepts (it rejects reasoning_effort:'none'). Addresses review point #1 from @wyuc on #784. * Improve token plan capability setup UI * fix token plan setup flow * chore: prettier format tts-providers.ts |
||
|
jackefnandYanpeng Wang
|
3b7a0ca0c0 |
feat(document): add multi-format course material upload (#741)
* feat(document): add multi-format course material upload * fix(document): support office uploads with mineru cloud * fix(document): handle mineru cloud office edge cases * fix(document): align mineru cloud artifact metadata --------- Co-authored-by: Yanpeng Wang <yanpg.wang@gmail.com> |
||
|
|
cec3c6aaea |
Refactor widget actions into the scene action pipeline (#796)
* refactor: align widget actions with scene pipeline * feat(edit): migrate legacy teacherActions off interactive content on load The widget-actions pipeline no longer persists a teacherActions authoring layer; playback reads only scene.actions. Add migrateInteractiveContent to strip the dead field from existing documents as they load through migrateScene (wired into all store/classroom load paths). Satisfies the #794 forward-migration acceptance item. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * fix(generation): default widget_setState.state to {} when the LLM omits it The removed TeacherAction->Action converter always defaulted state to {}. Restore that guard in the structured-output parser so an action missing state can't forward state: undefined into the widget iframe, where SET_WIDGET_STATE handlers dereference it. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * fix(edit): also strip nested widgetConfig.teacherActions in migration Legacy WidgetConfig variants each declared their own teacherActions, so existing documents can carry the dead field nested inside content.widgetConfig as well as at the top level. Extend migrateInteractiveContent to drop both, preserving all other widgetConfig fields. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * feat(prompts): document stable widget-action selectors for diagram/game/viz3d The interactive-actions prompt documented stable target selectors only for procedural-skill and simulation widgets. Add the canonical selectors the diagram / game / visualization3d content prompts actually emit (node IDs via revealOrder; #zoom-in-btn / #canvas-container / #speed-slider / ...; #start-btn), plus a rule to use widget_setState or a speech-only beat instead of guessing a target when no stable selector is known (e.g. free-form code widgets). Hardens against widget actions silently no-oping on non-existent selectors. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com> Co-authored-by: 杨慎 <117187635+cosarah@users.noreply.github.com> |
||
|
Yizuki_Ame
|
243e9f3a48 |
feat: auto-retry transient failures during scene generation (#788)
Adds automatic retry with exponential backoff for scene generation (content/actions/TTS): shared withGenerationRetry helper at the fetch boundary, smart error classification (retries 429/5xx/timeouts/network, fails fast on 401/403/404), AI SDK RetryError unwrapping, abort-aware backoff, per-route thinking config preserved, and upstream HTTP status surfaced for client retry classification. Fixes #755. |
||
|
杨慎andwyuc
|
0f44e9a46f |
Add PBL v2 runtime APIs and classroom UI (#799)
* feat: add PBL v2 runtime UI * fix: ship PBL v2 instructor avatar * fix: ship PBL v2 workspace logo * fix: prevent scenario PBL legacy fallback * chore: restore package lint comments * test: cover ordinary PBL legacy fallback * fix: front-load PBL single-call validation * fix: preserve PBL v2 content when assembling scenes --------- Co-authored-by: wyuc <wang-yc24@mails.tsinghua.edu.cn> |
||
|
wyuc
|
1d1ce80e04 |
feat(maic-agent): editor-agent line (v0) — Pro-mode "Edit with AI" (#777)
Promote the editor-agent line to main: read-then-act editor agent (read_scene_content, regenerate_scene, regenerate_scene_actions, edit_interactive_html), a routable maic-agent model stage, the agent sidebar UI, interactive-scene runtime-error capture, and the timeline/script editor. Cross-reviewed and smoke-verified; see #777. |
||
|
wyucandwyuc
|
267965d6ed |
feat(generation): infer concise courseTitle from outlines for readable course names (#756)
* feat(generation): infer concise courseTitle from outlines for readable course names
Generated courses were named after the raw user prompt (truncated to 500 chars
client-side, or 50 chars / first outline title server-side), which is long and
unreadable. The outline-generation step already understands the full course
semantics, so have it emit a concise `courseTitle` alongside `languageDirective`
and `outlines`, and use it as the stage name.
- Prompts (requirements-to-outlines, interactive-outlines): top-level JSON gains
a required `courseTitle` (≤30 chars, in the teaching language, noun phrase).
- Streaming route: extract `courseTitle` head-bound like `languageDirective`,
emit a `courseTitle` SSE event, and include it in the `done` event.
- Non-streaming outline-generator: parse and return optional `courseTitle`
(defensive trim + 120-char cap).
- Naming: stage.name = courseTitle || outlines[0]?.title || requirement fallback
(client preview path backfills stage.name after outlines resolve; server
classroom-generation uses it directly).
- Session: persist `courseTitle` on GenerationSessionState for reload safety.
- Tests: add courseTitle parsing coverage; e2e mock done event carries it.
Fully optional in the pipeline — every consumer falls back to the previous
behavior when the field is absent, so legacy/LLM-skipped cases are unaffected.
* fix(generation): make courseTitle propagation robust across all outline paths
Address review feedback on the courseTitle threading.
Prompts — align every outline-prompt schema statement to the 3-key shape.
The user-prompt "final reminder" / top-level shape blocks for
requirements-to-outlines, interactive-outlines, and task-engine-outlines still
demanded the old two-key {languageDirective, outlines} object; as the last
instruction the model reads, that caused it to omit courseTitle and the stage
name to fall back to the raw requirement. Task Engine prompts now emit
courseTitle as well.
Streaming route — normalize the streamed courseTitle (trim, ignore
whitespace-only, cap length) to match the non-streaming parser, and add a
full-buffer fallback so a title emitted after the outlines array or beyond the
head-scan window is still recovered before the done event.
Client — reset latched languageDirective/title on outline retry so a
succeeding attempt that omits them falls back instead of inheriting a failed
attempt's stale value; also carry courseTitle in the stream-end resolve path
that fires when the stream closes without an explicit done event.
courseTitle remains optional on every consumer path.
---------
Co-authored-by: wyuc <zdq1204@gmail.com>
|
||
|
wyucandClaude Opus 4.8
|
25cf58d19f |
feat(model): optional per-stage LLM model routing (#745) (#747)
* feat(model): optional per-stage LLM model routing (#745) Add a config-only stage -> model map consulted during model resolution, falling back to DEFAULT_MODEL when unset (zero behavior change unless opted in). - lib/server/model-routes.ts: parse/validate/cache MODEL_ROUTES env JSON; LLM_STAGES registry of routable stages; getStageModel(stage). - resolveModel: resolution order x-model > stage route > DEFAULT_MODEL > builtin; thread optional `stage` through resolveModelFromHeaders/FromRequest. - Wire each route's resolveModel call site to its stage; classroom-generation resolves generate-classroom and web-search-query-rewrite independently. - .env.example: document MODEL_ROUTES; unit tests for both modules. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * fix(model): address cross-review findings for per-stage routing (#745) - classroom-generation: resolve the web-search-query-rewrite model lazily, only when explicitly routed and inside the web-search branch, with try/catch fallback to the classroom model. Fixes a regression where a misconfigured optional route (keyless/invalid provider) threw and aborted ALL classroom generation even with web search disabled; also avoids wasted resolution on the common no-search path and preserves classroom-model inheritance when unrouted. - resolve-model: type the stage param as LlmStage across resolveModel/ resolveModelFromHeaders/resolveModelFromRequest so a mistyped stage literal is a compile error instead of silently falling through to DEFAULT_MODEL. - model-routes: simplify the cache to a process singleton (matching provider-config), dropping the raw-string-keyed cache and redundant STAGE_SET. - .env.example: clarify x-model precedence and that a route to an unconfigured provider fails at request time (no startup validation), like a bad DEFAULT_MODEL. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * fix(model): stage route takes precedence over client x-model (#745) Cross-review (codex P2 + Claude tracer) found that the browser UI always sends its saved model as x-model, so the previous x-model > stage order made MODEL_ROUTES inert for normal UI traffic on exactly the heavy stages the RFC targets (scene-content, quiz, pbl-chat, chat). Flip resolution to stage route > x-model > DEFAULT_MODEL: a configured route is the operator's deliberate per-stage choice and wins, while unrouted stages still honor the client's x-model. Update docs and tests accordingly. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * style(model): prettier-format changed files for #745 CI Run the repo formatter (pnpm check / prettier) on the changed routes, the new model-routes module, and the new tests so `prettier . --check` (CI) passes; also correct a stale precedence comment in the resolve-model test. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * fix(model): isolate routed model from client connection params (#745) When a stage route overrides the client's x-model and points at a different provider, the client-sent apiKey/baseUrl/providerType (for the client's model) must not bleed onto the routed provider — otherwise a routed Anthropic model would be constructed with the client's OpenAI providerType/key and fail. A routed model now resolves its connection params purely from server config, as if no x-model was sent. Unrouted stages still honor the client params. Adds tests covering both the drop (routed) and keep (unrouted) cases. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * docs(model): correct graceful-degradation comment in classroom (#745) resolveModel does not throw on a missing key; clarify that a keyless route surfaces later in callLLM (degraded by the outer try/catch), while only a resolution-time failure is caught at the rewrite re-resolve. Comment-only. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat(model): route scene-content per scene type (#745) Extend MODEL_ROUTES with composite keys scene-content:<type> for the four core scene types (slide/quiz/interactive/pbl). getStageModel now resolves composite `a:b` stages most-specific-first, trimming `:` segments, so scene-content:<type> falls back to the base scene-content route, then x-model, then DEFAULT_MODEL. The scene-content route derives its stage from outline.type. Backward compatible: with no composite key configured, behavior is unchanged. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * feat(model): per-stage thinking effort + drop hardcoded gpt fallback (#745) MODEL_ROUTES values may now be {model, effort} (not just a model string) to pin the thinking effort per stage. Arbitration mirrors model routing: a routed stage with effort set wins over the client's thinking (effort "none" disables); routed with no effort uses the model's default and drops client thinking; unrouted stages keep the client thinking. resolveModel is the single arbiter — body thinking is threaded in via resolveModelFromRequest. classroom-generation passes the resolved thinking into its callLLM calls so generate-classroom and web-search-query-rewrite honor route effort too. Also: removed the hardcoded `|| 'gpt-5.4-mini'` fallback in resolveModel — if no model resolves (no route / x-model / DEFAULT_MODEL) it now throws instead of silently picking a vendor default. model-routes: getStageRoute returns {model, effort?}; getStageModel delegates to it. Docs + tests updated. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * refactor(model): route value carries full ThinkingConfig (#745) Per-stage routing accepts the full unified ThinkingConfig in the route value ({model, thinking:{mode,effort,level,enabled,budgetTokens,excludeReasoningOutput}}) instead of just an effort string. The route's thinking is passed through resolveModel and normalized per the model's capability by callLLM, so budgetTokens (qwen), level (Gemini), enabled/mode, etc. all work. (qwen3.7-plus/max thinking capability now comes from upstream #753, so this no longer registers them.) Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * fix(model): honor routed thinking for chat-adapter; doc fixes (#745) Cross-review (codex P2 + Claude) found chat-adapter was the only routable stage that ignored the route's thinking: it took the routed model but built its own thinkingConfig from the request body, so operator-pinned thinking never applied and client thinking still leaked onto a routed model. Now pass the client thinking into resolveModel and use the resolved thinkingConfig (route-pinned for a routed stage, client otherwise; defaults to disabled for low-latency chat). Also: clarify DEFAULT_MODEL is now required for server-side stages (the hardcoded gpt-5.4-mini fallback was removed → resolveModel throws if nothing resolves), and fix a stale {model,effort}→{model,thinking} doc comment. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com> |
||
|
|
0989bd5d15 |
fix(proxy): bypass proxy for loopback hosts and honor NO_PROXY (#718)
proxyFetch routed every request through the configured HTTP(S) proxy, including loopback targets. A deployment with HTTP_PROXY set would send a http://localhost:8888 SearXNG (or any self-hosted service) request to the external proxy, where localhost resolves to the proxy machine and the request fails. - Always bypass loopback hosts (localhost, *.localhost, 127.0.0.0/8, ::1). - Honor the standard no_proxy / NO_PROXY env var: comma-separated hosts, `*` wildcard, optional `:port` (with protocol-default ports), and curl-style domain-suffix matching (`example.com` matches `api.example.com`; a leading dot is equivalent). Co-authored-by: Claude Fable 5 <noreply@anthropic.com> Co-authored-by: wyuc <wang-yc24@mails.tsinghua.edu.cn> |
||
|
RongHuiQiang
|
50c0aefd13 | fix(vocational): gate procedural content generation | ||
|
|
3ccd5da6da |
feat(generation): opt-in parallel scene-content generation (#660)
* feat(generation): opt-in parallel scene-content generation Scene generation was strictly serial — content -> actions -> TTS, one outline at a time — so an N-scene classroom paid N x (content + actions + TTS) of wall-clock latency, which dominates the post-outline wait. Add an opt-in hybrid two-phase path (#572): - Phase 1 fetches scene *content* concurrently (bounded). Content is the only per-scene step independent of cross-scene state, so it is safe to parallelise; a content failure marks just that outline and does not pause the batch. - Phase 2 keeps the existing in-order actions + TTS loop, so previousSpeeches threading, ordered addScene, the abort/epoch guards, and the pause-on-failure UX are all unchanged. Gated by a server-side PARALLEL_SCENE_CONCURRENCY (default 0 = off, clamped to 10), surfaced to the client through the existing /api/server-providers response. With it unset the content map is null and the loop is byte-for-byte the original serial path, so out-of-box behaviour is unchanged. The bounded worker pool is extracted to lib/utils/concurrency.ts (mapWithConcurrency, with an early-stop hook for abort/epoch) and unit-tested; getParallelSceneConcurrency env parsing/clamping is tested too. Closes #572 Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * refactor(generation): pipeline content fetches instead of a barrier Address @wyuc's review on #660. The previous version awaited the whole bounded content pool before Phase 2, so for N > C scenes the user saw the first scene, then a stall on a single page for the entire content phase, then a burst — perceived as worse than serial despite the wall-clock win (the gap scales with ceil((N-1)/C)). Pipeline instead: lazyBoundedMap starts the content fetches (<= C in flight) and returns one promise per outline immediately; the serial loop awaits them in order, so each resolves as soon as its own content is ready (usually already done, hidden behind the previous scene's actions/TTS). The first scene now paints after content(1)+actions(1)+TTS(1) — same as serial — with the full concurrency benefit and no stall. Also from the review: - a content fetch can no longer take its siblings down: an unexpected throw is caught and returned as a failure result, routed through the same mark-failed path as the serial loop (symmetric); - drop the cosmetic Phase-1 setCurrentGeneratingOrder — the loop already sets it per outline; - note the intentional belt-and-suspenders clamp; - concurrency.test.ts asserts <= limit (not == limit), and gains lazyBoundedMap tests including the no-barrier property. mapWithConcurrency is kept as a thin await-all wrapper over lazyBoundedMap. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> --------- Co-authored-by: ly-wang19 <ly-wang19@users.noreply.github.com> Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Co-authored-by: wyuc <wang-yc24@mails.tsinghua.edu.cn> |
||
|
|
d7c0d80afe |
feat(tts): per-agent auto-voice quality + register-once timbre stability (#670) (#672)
* feat(tts): voxcpm voice-design types + deterministic voice id helpers (#670) Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * feat(tts): emit + persist per-agent voiceDesign descriptor (#670) Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * feat(tts): voxcpm voice registration backend client (#670) Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * feat(tts): voxcpm-voice ensure/register endpoint (#670) Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * feat(tts): client auto-voice registration + reference-clip cache (#670) Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * feat(tts): reference registered voice id in vLLM-Omni speech (#670) Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * feat(tts): thread voiceDesign + backend through tts call sites (#670) Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * fix(tts): match vLLM-Omni voice registration contract (#670) Live e2e against the real backend revealed the assumed multipart contract was wrong: POST /v1/audio/voices needs name + consent + audio_sample (not voice_id/file), and there is no per-name GET (405) — existence must list /v1/audio/voices and check membership. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * refactor(tts): make auto-voice register-once pattern provider-neutral (#670) The descriptor + register-once/reference-by-id pattern is not VoxCPM-specific. De-couple it into a provider-neutral seam so other registration-capable providers (ElevenLabs/MiniMax/Doubao voice cloning, …) can plug in: - lib/audio/voice-design.ts: VoiceDesign + buildVoiceDesignPrompt / normalizeVoiceDesign / getDeterministicVoiceId (neutral 'auto-<hash>' id, namespaced by providerId). - lib/audio/voice-registration.ts: VoiceRegistrationAdapter interface + providerId->adapter registry + supportsVoiceRegistration. - lib/audio/voice-registration-client.ts: neutral ensureRegisteredVoice() + IndexedDB clip cache. - app/api/generate/voice (replaces .../voxcpm-voice): dispatches by providerId. - voxcpm-registration.ts becomes the VoxCPM adapter (sole registered provider). - AgentConfig.voiceDesign + DB cache table renamed neutral. Behavior-preserving; VoxCPM-specific bits (inline (prompt)text, backend kinds, capability gate) stay in the voxcpm modules. Full suite green (674). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * fix(tts): build the voice clip Blob from an inline Uint8Array (#670) A factored-out helper returning Uint8Array widened to Uint8Array<ArrayBufferLike>, which next build (stricter than bare tsc) rejects as a BlobPart. Inline the buffer like the rest of the repo does. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * feat(tts): managed TTS providers resolve model server-side (#670) A server-managed TTS provider's model was still client-driven, but the managed-provider settings UI hides the model field — so a backend whose model id isn't the client default (e.g. VoxCPM/vLLM-Omni serving the full model path) always 500'd. resolveTTSModel() makes the model authoritative from server config (${PREFIX}_MODELS, first entry) for managed providers, like key/baseUrl; unmanaged/unconfigured providers keep the client model unchanged. Used by the tts and voice routes. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * fix(tts): single agent-voice resolver + stable teacher narration voice (#670) Route all TTS voice resolution (narration, discussion, preview) through one resolveAgentVoiceOptions(agent, ...) that reads the agent profile and registers + references the voice by id. Eagerly warm up generated agents' voices on save. Critically fixes the teacher narration drifting (male/female jumps): the registry is always seeded with DEFAULT_AGENTS, so the old narration lookup find(role==='teacher') returned the default teacher (no voiceDesign) instead of the generated one — so narration never registered a voice and fell back to the inline prompt. pickNarratorAgent() now prefers the teacher carrying a voiceDesign. Also: drop language from the deterministic voice id (descriptor already encodes it) so narration (directive) and discussion (locale) resolve the same id; log the effective registeredVoiceId. Regression tests for pickNarratorAgent. Verified e2e: agent-profiles -> 1 voice registration -> 16 narration TTS all referencing the same registeredVoiceId; voice present on the backend. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * feat(tts): fall back to persona as the voice seed for agents without a voiceDesign (#670) Preset/default agents carry no LLM voiceDesign. Rather than add a static field, derive the bootstrap descriptor from the agent's persona when voiceDesign is absent, so they still register one stable reference voice (stable-but-generic; persona is not a vocal spec). Generated agents keep their LLM voiceDesign. Also hardens replay: stage snapshots drop voiceDesign but keep persona. Unit-tested: resolveAgentVoiceOptions uses real voiceDesign when present, persona otherwise. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * fix(tts): match Auto Voice by its localized label in the agent voice picker (#670) The picker filtered on voice.name ('Auto Voice'), but Auto Voice is shown via its localized label (自动音色). Searching '自动' returned '没有匹配音色'. Match the localized label for the Auto Voice option so it's findable in any language. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * fix(tts): address code-review round 1 (#670) - strip parentheses in the voice-design prompt so a paren in the descriptor/ persona can't corrupt the (prompt)text bootstrap delimiter - gate POST /api/generate/voice on isServerTTSProviderDisabled (#665): a force-disabled provider was off for the tts route but not this sibling - check voiceExists before re-registering a client-cached clip, so a cached voice that's still live isn't needlessly re-uploaded every session - dedup concurrent ensureRegisteredVoice calls via an in-flight promise map (eager warm-up + first utterance no longer double bootstrap/register) - warm up only the narrator (teacher), not every generated agent, to avoid synthesizing voices at save time for agents that may never speak - collapse the duplicated vLLM-Omni speech payload into one object Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * chore(tts): prettier formatting (#670) Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * fix(tts): key auto-voice memo by (voiceId, backend) so a base-URL change re-registers (#670) Addresses review (cosarah): registeredThisSession/inFlight were keyed by voiceId alone, so switching the VoxCPM base URL mid-session made ensureRegisteredVoice short-circuit and return an id registered only on the old backend — TTS then sent that stale registeredVoiceId and skipped the inline fallback, failing with voice-not-found on the new backend. Memo key now includes the base URL; the IndexedDB clip cache stays keyed by voiceId (the reference clip is backend-independent and reused to re-register elsewhere). Regression test added. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * fix(tts): include API key in the auto-voice memo key (#670) Addresses review (cosarah, round 2): memoKeyFor was keyed by (voiceId, baseUrl) but not the API key. Registration/existence checks and speech calls are auth-scoped, so switching account/key on the same base URL could reuse a registeredVoiceId from the old credentials and skip re-validation. Memo key now includes the API key (in-memory only, never persisted/logged). Regression test covers the key-switch case. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com> Co-authored-by: 杨慎 <117187635+cosarah@users.noreply.github.com> |
||
|
|
3e1e9f083b |
feat(tts): unify provider-enablement model, default browser-native off (#665) (#666)
A single TTS provider-enablement model honored everywhere (picker,
auto-assignment, fallback, lecture/playback, server).
Model
- lib/audio/provider-enablement.ts: isTTSProviderConfigured /
isTTSProviderEnabled / listEnabledTTSProviderIds. Availability drops the
registry defaultBaseUrl fallback (fixes Lemonade's always-shown voices),
honors the per-provider `enabled` flag, server-disable overrides (server
precedence).
- voice-resolver: getEnabledProvidersWithVoices (enabled server/custom
providers) and getSelectableProvidersWithVoices (the ONE client source of
truth = enabled providers + opt-in browser-native when the browser exposes
voices). resolveAgentVoice picks deterministically among the given providers,
honors a voiceConfig only while its provider is present.
- Picker + discussion build their list from getSelectableProvidersWithVoices, so
the AgentBar and the live discussion agree on selectable voices (fixes
student agents going silent in auto mode). Students resolve via
resolveAgentVoice; the teacher uses the global lecture voice verbatim (same
voice the pre-generated lecture sends), so lecture and discussion never
diverge.
- Voice assignment vs fallback: an agent's voice is bound at generation (the
agent-profile LLM picks from availableVoices; preset uses the user's pick).
The fallback for a missing/invalid/unavailable voice is applied only at the
live TTS calls, not baked into the saved agent — so it reflects current
provider availability.
Defaults / opt-in
- Browser-native: first-class, default OFF, opt-in. Built-in providers default
enabled:true (only surface once configured).
- TTS master toggle defaults OFF; first server-sync auto-enables it when a
provider is configured (mirrors image/video).
- Persist v3->v4 migration normalizes the never-user-editable flags.
Server/admin disable (server precedence): server-providers.yml `enabled:false`
and env `TTS_<P>_ENABLED=false` (incl. browser-native). Surfaced as a per-provider
`disabled` flag on /api/server-providers; merged as serverDisabled; enforced at
the /api/generate/tts trust boundary, classroom media generation, and /api/health.
UI
- Settings TTS panel: per-provider enable toggle, disabled (with a hint) when the
provider is unavailable, locked off (with a note) when server-disabled.
- AgentBar voice pills render disabled with a "No voice" placeholder when no
provider is enabled. The toolbar media-popover TTS tab keeps its existing
on/off toggle (no picker/CTA).
Tests: predicates, resolver determinism + unified selectable list, server
force-disable, store merge + migration + TTS auto-enable, i18n parity. Verified
in a real browser (Playwright): settings toggle states + a full Gemini classroom
generation whose TTS requests used the picked voice.
Found on main @
|
||
|
|
d2b6db60a4 |
fix: point SSRF local-network rejection at ALLOW_LOCAL_NETWORKS escape hatch (#667)
Co-authored-by: Matt Van Horn <455140+mvanhorn@users.noreply.github.com> Co-authored-by: wyuc <wang-yc24@mails.tsinghua.edu.cn> |
||
|
|
2586cb5e05 |
fix(tts): surface provider rate-limit errors as HTTP 429 (#644)
TTS providers throw a plain `Error` on any non-OK response, and the `/api/generate/tts` route catches everything as `GENERATION_FAILED` (500), so an upstream 429 (quota/concurrency) was indistinguishable from a real failure — clients had no signal to retry/back off. `TTSRateLimitError` already existed for exactly this (with a TODO noting the route collapses everything uniformly), but only Doubao threw it and the route never mapped it. - Add a `RATE_LIMITED` API error code. - Map upstream HTTP 429 to `TTSRateLimitError` at every provider's `!response.ok` site via a shared `throwIfTtsRateLimited` helper. - Return `429 RATE_LIMITED` from the TTS route when a `TTSRateLimitError` is caught (all other errors unchanged). Purely additive: non-429 responses keep their existing error messages. Co-authored-by: ly-wang19 <ly-wang19@users.noreply.github.com> Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Co-authored-by: wyuc <wang-yc24@mails.tsinghua.edu.cn> |
||
|
wyuc
|
2165e55bac | feat: add MiniMax web search provider (#634) | ||
|
|
35d36905d1 |
Make server-configured providers admin-managed, not client-overridable (#620) (#624)
* refactor(providers): make server-configured providers admin-managed, not client-overridable A server-configured provider is now authoritative: the operator owns its API key and base URL, and the client cannot override either. This collapses the tri-state (server key / client base URL / client key) that was the root of a recurring "missing API key" bug class — when the settings UI echoed a base URL with an empty key, routes that treated "any base URL ⇒ fully client-configured" dropped the server key. #533 patched that per-route via canUseServerApiKeyForBaseUrl; this removes the guard entirely by fixing the design. Server: - resolve*ApiKey/BaseUrl ignore client values when the provider is server- configured; add isServerConfiguredProvider(section, id); drop canUseServerApiKeyForBaseUrl + normalizeProviderBaseUrl. - getServer*Providers() no longer return base URLs — only allowed models (LLM/image) and the managed flag (presence) — so an internal gateway/proxy endpoint is never exposed to the client. - routes (chat/image/video/tts/asr/pdf + verify-*) resolve managed providers purely server-side; SSRF validation now applies only to unmanaged (client) base URLs. Client: - drop serverBaseUrl from ProviderSettings, the settings store, request-building (TTS base URL no longer echoes the server URL) and voice resolution. - settings UI renders managed providers read-only: the key/base-URL override inputs and region toggles are hidden, leaving the "managed by server" notice. Tradeoff: overriding a server provider's key with your own for the same provider is removed; add a separate custom provider instead. Closes #620. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * chore(providers): refresh managed-provider copy, drop dead audio-settings, add #620 e2e - serverConfiguredNotice now says the provider is managed and not editable here, pointing users to add a separate provider for their own credentials (7 locales). - delete components/settings/audio-settings.tsx — unreferenced dead code. - add e2e/tests/managed-provider-620.spec.ts: managed provider hides the key/base-URL inputs; unmanaged provider keeps them. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * feat(providers): lock the model catalog for managed providers with a pinned list When the operator pins an allowed model list (MODELS env/yaml ⇒ serverModels), the model catalog is admin-managed too: the LLM provider panel shows the pinned models read-only with a "Server" badge and no add / reset / edit / delete affordances. Without a pinned list the server manages only credentials and the user keeps curating models. This removes the prior inconsistency where the settings panel let you add models to a pinned managed provider that the generation toolbar then filtered out. e2e: the managed-openai case now also asserts the catalog is locked. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> * fix(providers): ignore (not reject) client overrides for managed web-search; clear legacy serverBaseUrl Addresses review feedback on #624: 1. Managed web-search providers now ignore a client-sent key/baseUrl instead of validating and rejecting it. A server-configured provider is admin-owned, so the server config is authoritative and a stale client base URL is dropped rather than 400-ing the request. The allowlist validation still runs for unmanaged providers, where the client base URL is actually used. 2. Drop the removed `serverBaseUrl` field from persisted localStorage on rehydrate (settings version 2 -> 3, stripped in both migrate and merge), so old clients don't keep a stale server URL in client state. Tests: web-search route test split into unmanaged-rejects-allowlist and managed-ignores-client-baseUrl; settings rehydrate test asserts a legacy serverBaseUrl is stripped. Full suite 381 pass; e2e 2 pass. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com> Co-authored-by: 杨慎 <117187635+cosarah@users.noreply.github.com> |
||
|
|
86c8e0c543 |
feat: inline external assets at export for offline classrooms (#596) (#613)
* feat(export): add external-asset reference scanner for HTML inlining * feat(export): add cached asset fetcher and data-uri encoder Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * feat(export): recursively inline url() refs inside stylesheets Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * feat(export): orchestrate inlining of links, scripts, images, and inline css Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * refactor(export): drop dead escapeAttr, harden importmap skip in scanner * feat(export): inline importmap modules incl three/addons graph closure Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * refactor(export): extract shared module to break inline-assets import cycle Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * feat(export): inline interactive-scene assets into .maic.zip at export Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * feat(export): inline interactive assets in Resource Pack export too * docs: document offline export asset inlining * style: prettier formatting * fix(test): resolve tsc type errors in export-classroom-inline test * fix(export): keep importmap prefix fallback on partial failure, chunk base64, preserve media attr, log pptx inline failures Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * perf(export): parallelize asset fetching, share fetcher across scenes, promise-cache Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * fix(export): retry transient asset fetches, bound concurrency, warn user on partial inline Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * perf(export): inline only woff2 fonts, drop woff/ttf siblings to halve size Within @font-face blocks that have a woff2 url(), rewrite sibling woff/ttf/otf/eot refs to url(about:invalid) instead of inlining them. Falls back to full inline for ttf-only blocks. Drops katex output from ~1.79MB to ~0.71MB (~3x fewer font fetches). Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * i18n(export): translate partial-inline warning and surface failed asset hosts Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * feat(proxy-media): follow redirects with per-hop SSRF revalidation (#398) Replace blanket redirect rejection with a bounded 5-hop follow loop that re-validates every Location target through validateUrlForSSRF before fetching, preventing redirect-to-internal SSRF while allowing CDN redirects (e.g. cdn.tailwindcss.com 302→/3.4.17). * feat(export): route asset inlining through proxy-media to bypass CORS Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * fix(export): keep importmap prefix fallback, forward proxy 4xx, cap proxy size, re-export FetchAsset * style: prettier formatting for proxy-media + proxied-fetch * fix(export): report CSS-nested asset failures in inline report Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> --------- Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com> Co-authored-by: 杨慎 <117187635+cosarah@users.noreply.github.com> |
||
|
LooThaoandwyuc
|
6d29bbe9af |
Fix server API key fallback when client echoes provider base URL (#533)
* Fix server API key fallback for provider base URLs * Format verify PDF provider route --------- Co-authored-by: wyuc <wang-yc24@mails.tsinghua.edu.cn> |
||
|
Ismail Ariyan
|
07115dff17 |
feat(asr): add Azure STT (Fast Transcription) provider (#175)
* add Azure ASR config and localization * add transcribeAzureASR func * fix(settings): prevent Azure STT crash in ASR modal * feat(i18n): add Azure STT provider support in language locales |
||
|
Yizuki_Ame
|
47cc2a510c |
feat: add Brave Search and Baidu Search integration
Add Brave Search and Baidu Search providers with Baidu sub-source support, provider configuration updates, tests, and docs links. |
||
|
wyuc
|
2dff6d10be |
feat: add generated video manifest refs (#540)
* feat: add generated video manifest refs * style: format video manifest changes * fix: address video manifest review feedback |
||
|
Helckandxuruiray
|
46b61dedfd |
feat: add HappyHorse video adapter (#509)
Add Alibaba Cloud DashScope HappyHorse text-to-video support as an OpenMAIC video generation provider. Co-authored-by: xuruiray <15962843+xuruiray@users.noreply.github.com> |
||
|
|
be759a8252 |
feat: add lemonade to provider (#508)
* feat: add lemonade to provider * fix(lemonade): correct TTS/ASR model IDs and expand Kokoro voices - TTS model: kokoro → kokoro-v1 - ASR model: single 'whisper' → 6 Whisper variants (Base/Large-v3/Large-v3-Turbo/Medium/Small/Tiny), default Whisper-Base - Expand lemonade-tts voices to full Kokoro v1 set (54 voices across en-US/en-GB/zh-CN/ja-JP/es/fr-FR/hi/it-IT/pt-BR) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(lemonade): zh-TW i18n parity, real Kokoro default voice, language code consistency - zh-TW.json: add missing providerLemonadeTTS/ASR/Image keys - TTS default voice: 'default' → 'af_heart' (Kokoro expects real voice ID, not placeholder) - Voice language codes: 'es'/'hi' → 'es-ES'/'hi-IN' for consistency Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * test(lemonade): add unit tests for image/TTS/ASR adapters and wav-utils Covers: - lib/media/adapters/lemonade-image-adapter: request shape, base URL fallback, custom model, optional Bearer auth, error paths, connectivity probe - lib/audio/tts-providers (lemonade-tts case): request body (model/voice/format), af_heart fallback, optional auth, error path - lib/audio/asr-providers (lemonade-asr case): WAV-only validation, model/language form fields, "audio is empty" graceful empty result, error propagation, default model fallback - lib/audio/wav-utils: isWavBlob detection by MIME and filename, normalizeASRUploadAudio pass-through and wav handling Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(lemonade): remove text path fork and harden fallbacks * style: format lemonade tts files * i18n: add lemonade voice language labels * fix(lemonade): align image default model and harden wav detection * test(lemonade): fix File BlobPart typing * fix(lemonade): simplify tts voice selection and clear asr test state * docs: highlight Lemonade support --------- Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com> Co-authored-by: wyuc <wang-yc24@mails.tsinghua.edu.cn> |
||
|
HelckandCodex
|
58acad55d8 |
feat: support OpenAI image env fallback (#510)
Summary: - Enable the OpenAI image provider from standard OpenAI server env vars when image-specific env vars are absent. - Add provider-config coverage for resolving the fallback API key and base URL. Rationale: - Allows gpt-image-2 image generation to work with common OpenAI proxy configuration. Tests: - pnpm vitest run tests/media/openai-image-adapter.test.ts tests/server/provider-config.test.ts - pnpm prettier lib/server/provider-config.ts tests/server/provider-config.test.ts --check - pnpm lint -- lib/server/provider-config.ts tests/server/provider-config.test.ts Co-authored-by: Codex <codex@openai.com> |
||
|
wyuc
|
dbe1d2ec76 |
[codex] Add Bocha web search provider (#524)
* Add Bocha web search provider * Address Bocha web search review feedback |
||
|
wyuc
|
5e12ad892c |
[codex] add VoxCPM2 TTS provider (#496)
* add voxcpm2 tts provider * address voxcpm review feedback * translate voxcpm locale strings * narrow toolbar model pill * refine toolbar media controls * address voxcpm review followups |
||
|
wyucand杨慎
|
e613b7578f |
[codex] add per-model thinking config (#494)
* add per-model thinking config * Refine thinking model controls --------- Co-authored-by: 杨慎 <117187635+cosarah@users.noreply.github.com> |
||
|
wyuc
|
a95aa87bf1 | support recent model providers (#481) | ||
|
wyucandClaude Opus 4.7
|
550c43a3f3 |
fix(generation): unify languageDirective across outline + scene pipeline (#472) (#474)
* fix(generation): unify languageDirective across outline + scene pipeline (#472) Root issue: languageDirective was dropped, renamed, or hardcoded at several points in the scene-generation pipeline, silently degrading prompt-level language control. Slide paths survived because their fallback happened to hint requirement-language; widget paths degraded to zh-CN default. Fix plan (all code-only threading + prompt hygiene, no schema changes): 1. Thread languageDirective through every buildPrompt in scene-generator.ts: - Add param to generateSlideContent / generateQuizContent / generatePBLSceneContent - Forward from generateSceneContent's slide / quiz / pbl branches - Include key in the four generateSceneActions buildPrompt calls - Demote generatePBLSceneContent's hardcoded English to a fallback - Drop the dishonest `as 'zh-CN' | 'en-US'` cast on the widget path 2. Unify variable naming across widget templates + callers: - Rename {{language}} → {{languageDirective}} in 6 widget templates (simulation / diagram / code / game / visualization3d / widget-teacher-actions) - generateWidgetContent and generateWidgetTeacherActions now pass `languageDirective: languageDirective || ''` instead of a locale-code shim 3. Make the outline LLM reliably emit the wrapper object: - requirements-to-outlines/system.md + user.md restructured so the "top-level JSON object with languageDirective + outlines" contract is surfaced up-front, short, and unambiguous — drowning-by-scene-examples removed. Important Reminders reordered: top-level shape first. - interactive-outlines/system.md + user.md aligned with main outliner: adds full Language Inference section, requires wrapper output, drops the orphan {{language}} placeholder that never had a value source. 4. Remove lib/prompts/loader.ts prompt/snippet caches: - Disk reads for prompt templates are not on any hot path (LLM call dominates by 5 orders of magnitude). The cache's only observable effect was blocking prompt edits from taking effect without restart — which masked the wrapper bug during this investigation. Verification: - 7 new regression tests in tests/generation cover directive threading at each buildPrompt site plus the PBL caller-forwarding guarantee. - outline-language eval (50 scenarios, gemini-3-flash-preview inference, qwen-3.5-flash judge): 48/50 (96%). The 2 persistent failures are in the proxy-request category (teacher / tutor preparing lessons for students) where the system prompt's rule 5 has genuine ambiguity — orthogonal to this fix. - Curl burst tests: 8/8 runs each for regular-mode and interactive-mode outline endpoints emit the wrapper object and extract a detailed inferred directive. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * style: apply prettier formatting to loader.ts and new test file Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix: thread languageDirective in scene-builder and classroom-generation Code review caught two additional sites missing the directive: - lib/server/classroom-generation.ts:371 — generateSceneActions call was missing languageDirective, reproducing the issue #472 symptom in the shareable-classroom server generation path. - lib/generation/scene-builder.ts:114 — buildSceneFromOutline's generateSceneActions call had the same miss; latent (no current live caller) but wrong for consistency. Also renumber duplicate `8.` in requirements-to-outlines/system.md scene-level rules. Extend regression tests with a buildSceneFromOutline end-to-end case that captures both the content and actions prompts via a stub aiCall and asserts the directive reaches both phases — this would have caught the scene-builder miss. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * review: widget-path test + docs + fallback constant (PR #474) Address @cosarah's review feedback: - Add widget-path regression test (tests/generation/...): interactive scene with widgetType=simulation captures both widget content and widget-teacher-actions prompts, asserts the directive reaches each and no `{{language}}` / `{{languageDirective}}` placeholder leaks. This is exactly the path issue #472 was filed for. - Rewrite lib/prompts/README.md "Caching" section: the cache was removed in this PR; README still instructed callers to use a clearPromptCache() function that no longer exists. - Sync interactive-outlines/user.md with the main outliner's "## Language Context" reminder block (line 13-19) so both outline paths give the LLM the same language-inference reminders. - Hoist 'Teach in the language that matches the user requirement.' into DEFAULT_LANGUAGE_DIRECTIVE, exported from outline-generator. Replaces 4 copies across outline-generator, scene-generator, and the scene-outlines-stream route. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
wyucandClaude Opus 4.7
|
f40c92f1aa |
refactor(orchestration): migrate prompt builders to file-based templates (#459)
* refactor(prompts): move loader to lib/prompts/ Move file-based prompt loader from lib/generation/prompts/ to project-level lib/prompts/ so it can be shared by orchestration and PBL prompts in subsequent commits. - Templates and snippets moved alongside the loader - getPromptsDir() now points at lib/prompts/ - Generation pipeline imports updated to @/lib/prompts (outline-generator, scene-generator, search-query-builder, scene-outlines-stream route) - Add tests/prompts/loader.test.ts as sanity gate Pure file move — no behavior change. * test(orchestration): add baseline snapshots for prompt builders Snapshot tests cover variant matrix: - buildStructuredPrompt: role × scene-type × peer/ledger/discussion/profile - convertMessagesToOpenAI: mixed message kinds - summarizeConversation: truncation - buildDirectorPrompt: Q&A vs discussion, ledger, profile - buildPBLSystemPrompt: default config These lock current output so the body refactor in subsequent commits can be verified byte-equal. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * test(orchestration): plug snapshot coverage gaps from review - Split convertMessagesToOpenAI test: same-agent (existing) + cross-agent (new). The original test passed currentAgentId matching the message's agentId, so the cross-agent role-conversion branch was never exercised. - Add 'teacher / quiz scene' snapshot to lock the spotlight/laser strip path in getEffectiveActions. The previous variant-7 test (whiteboard-open) only triggered the mutual-exclusion warning, not the strip — renamed it to reflect what it actually tests. - Add assistant-role variant (was uncovered: ROLE_GUIDELINES, buildLengthGuidelines, buildWhiteboardGuidelines all branch on role). * refactor(orchestration): extract summarizers from prompt-builder.ts Move state-context, virtual whiteboard ledger replay, peer context, message converter, and conversation summary out of the 890-line prompt-builder.ts into focused modules under lib/orchestration/summarizers/. prompt-builder.ts now only owns prompt assembly. director-graph imports updated to the new locations. Snapshot tests pass byte-equal — no behavior change. * refactor(orchestration): move agent-system prompt to template file The buildStructuredPrompt template literal is now lib/prompts/templates/agent-system/system.md, assembled by a thin variable-pass through the shared prompt loader. Per-variant ternaries (format example, ordering, spotlight examples, mutual-exclusion note) are kept as module-level constants in prompt-builder.ts. ROLE_GUIDELINES, buildLengthGuidelines, and buildWhiteboardGuidelines stay too (they may further migrate to template partials in Phase 2). Also: PROMPT_IDS gains a 'satisfies Record<string, PromptId>' clause to prevent constant/union drift as more IDs are added. Snapshot tests (12 prompt-builder + 4 director + 1 pbl) pass byte-equal. * refactor(prompts): align agent-system template with project conventions - Rename agent-system placeholders snake_case → camelCase to match the convention already used by slide-actions/user.md and other generation templates. Settles convention before Task 5 lands two more templates that would otherwise inherit drift. - Extract '## Speech Guidelines (CRITICAL)' into lib/prompts/snippets/speech-guidelines.md for reuse by director and PBL prompts in Task 5. Snapshot tests pass byte-equal — placeholder name changes are invisible in rendered output, and snippet inclusion happens at load time. * refactor(orchestration,pbl): move director + PBL prompts to templates director-prompt.ts and pbl-system-prompt.ts now use the shared template loader. Bodies collapse to thin variable assembly. Three orchestration-tier prompts now share one infrastructure: agent-system, director, pbl-design. Snapshot tests pass byte-equal. * refactor(orchestration): extract WhiteboardActionRecord + AgentTurnSummary to types module These two interfaces are imported by 6+ modules including summarizers/ — having them live in director-prompt.ts created an awkward upward import direction (summarizers reaching back to a sibling prompt builder for types). Move them to lib/orchestration/types.ts and update all callers. director-prompt.ts now imports them too. Pure type-location refactor — snapshots pass byte-equal. * test: remove snapshot tests in favor of eval-as-integration-test Per review feedback: byte-equal snapshots were the wrong invariant for this refactor. The goal was a stable substrate for editing prompts, not a frozen contract. Snapshots would have generated maintenance churn (~100-200 line diffs per intentional Phase 2 prompt tweak) without providing value the eval doesn't already. Test posture going forward: - tests/prompts/loader.test.ts (3 tests) — bounds the loader infrastructure (template load, snippet inclusion, variable interpolation, missing-id behavior) - pnpm eval:whiteboard — bounds end-to-end agent loop, template composition, and integration with chat/director/state-manager Removed: 1879 lines of .snap + 320 lines of test scaffolding + fixtures. Net PR diff drops by ~2200 lines. * docs(prompts): add README + structural assertion tests (review feedback) Addresses two items flagged by PR #459 final review: 1. lib/prompts/README.md — conventions, syntax, gotchas, and local-testing recipe for template authors. Previously there was no doc for the stated goal of "non-engineers can edit prompts." 2. tests/prompts/templates.test.ts — 8 structural assertions covering: - no surviving {{...}} placeholders in any rendered template - role dispatch (teacher → LEAD TEACHER, student → not) - scene-type action stripping (quiz scene has no spotlight/laser) - director prompt output spec mentions next_agent These replace the removed byte-equal snapshot suite at much lower maintenance cost — they assert behaviors the refactor must preserve, not exact bytes. * chore(prompts): relocate templates added on main during rebase PR #461 (interactive mode clean) added 7 new templates under lib/generation/prompts/templates/ (the pre-refactor location). Rebase onto origin/main landed them at the old path; git's rename detection didn't forward them through the directory move in db56c40. Manually move them to lib/prompts/templates/ to match the new convention, and fix scene-generator.ts's relative import ('./prompts/types' → '@/lib/prompts/types') since the dir no longer exists as a sibling. No behavior change — templates read from PROMPT_IDS values via getPromptsDir() which already points at lib/prompts/. * refactor(prompts): address review feedback from @cosarah - loadSnippet now throws on missing file instead of silently returning a literal {{snippet:id}} string. A typo like {{snippet:speach-guidelines}} now fails at load time instead of reaching the LLM. - README gains a "Still in TypeScript" section listing the role-conditional content that still lives in prompt-builder.ts (ROLE_GUIDELINES, buildLengthGuidelines, buildWhiteboardGuidelines) so contributors expecting a pure-markdown workflow know where to look. - Expand tests/prompts/templates.test.ts to cover the conditional branches Phase 2 is most likely to touch (9 new assertions): - assistant role dispatch - peer-context section toggles on agentResponses presence - language constraint toggles on stage.languageDirective presence - director Q&A vs discussion mode branching - pbl-design {{issueCount}} substituted at all 3 occurrences - placeholder-naming-convention lint scans every template for non-camelCase placeholders (slide-content grandfathered) - Comment above interpolateVariables regex documents why kebab-case placeholders pass through silently (the lint test now catches them). - New test in loader.test.ts locks the throw-on-missing-snippet behavior. --------- Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |