Add a top-level Mermaid architecture diagram (architecture/system-architecture.md)
covering all deployable components, external systems, communication protocols,
and security boundaries. Update the arch-doc-writer agent to maintain this
diagram as part of its documentation workflow.
Co-authored-by: John Myers <johntmyers@users.noreply.github.com>
* feat(sandbox): allow egress to private IP space via allowed_ips policy field
Add an allowed_ips field to NetworkEndpoint that accepts CIDR notation and
exact IPs, enabling sandboxes to reach private IP space under policy control
while maintaining SSRF protections for loopback and link-local addresses.
Three modes are supported:
- Default (no allowed_ips): all private IPs blocked (existing behavior)
- Host + allowlist: domain must resolve to an IP in allowed_ips
- Hostless allowlist (no host): any domain allowed if resolved IP matches
Co-authored-by: John Myers <johntmyers@users.noreply.github.com>
TLS certificates are always resolved automatically from cluster metadata,
making the explicit CLI flags unnecessary. The TlsOptions struct and
auto-resolution logic remain intact for programmatic and test use.
* fix(router): use protocol-aware auth headers for inference proxying
The router always sent `Authorization: Bearer` when proxying to backends,
which breaks Anthropic endpoints that require `x-api-key`. Thread the
source_protocol through to proxy_to_backend and set the appropriate auth
header based on protocol family.
* fix(server): remove noisy inference bundle log
The "serving inference bundle" info log fired every 30s per sandbox on
each route refresh poll. Remove it since there is no equivalent log for
policy serving.
* fix(sandbox): keep inference routing active when cluster bundle is initially empty
In cluster mode, if GetSandboxInferenceBundle returned zero routes at
sandbox startup, build_inference_context() returned None, disabling
inference routing for the sandbox lifetime and preventing
spawn_route_refresh() from starting. Routes created after sandbox
startup were never picked up.
Keep inference interception active with an empty cache in cluster mode
so background refresh can activate routes created later. File-mode
semantics are unchanged.
Add tab-completion for positional name arguments across sandbox, cluster,
and provider commands. Completers fetch live data on tab-press: cluster
names from local metadata files (no network), sandbox and provider names
via gRPC to the active cluster's gateway with a 2s timeout.
Key implementation details:
- Move rustls crypto provider init before CompleteEnv::complete() so TLS
works inside completers
- Spawn a dedicated OS thread for async gRPC calls to avoid nested tokio
runtime panic (completers run synchronously inside #[tokio::main]'s
block_on)
- All completers fail silently (return empty vec) on any error
## Summary
- Downgrade `GetSandboxPolicy served from policy history` from INFO to DEBUG — this was firing every 30s per sandbox and filling the 2000-line in-memory log buffer with noise, drowning out real events
- Add `"navigator: log push reconnected (attempt N)"` eprintln on successful reconnect in the sandbox log push loop, so operators can confirm recovery after a gateway restart
## Test Plan
- Deployed to local cluster, created sandbox, verified no policy polling spam in `nav sandbox logs`
- Restarted gateway, confirmed sandbox stderr shows reconnect success message: `navigator: log push reconnected (attempt 5)`
- Verified sandbox log push re-establishes after gateway restart
## Summary
- Add HTTP/2 keep-alive settings (`http2_keep_alive_interval`, `keep_alive_while_idle`, `keep_alive_timeout`, `connect_timeout`) to the sandbox gRPC client channel, matching what the CLI already uses
- Wrap the `PushSandboxLogs` client-streaming RPC in an outer reconnect loop with exponential backoff (1s → 30s) so that server restarts, network blips, or idle connection reaping no longer permanently kill sandbox log delivery
- Buffer up to 200 log lines during reconnect attempts so they are delivered after re-establishing the connection
- Detect stream death proactively via an `rpc_done` channel instead of waiting for the next batch send to fail
## Problem
The `PushSandboxLogs` RPC is a long-lived client-streaming call that stays open for the sandbox's entire lifetime. Without HTTP/2 keep-alive, Kubernetes service proxies reap idle TCP connections, causing a silent `"transport error"`. The push loop would then exit permanently with no retry — all subsequent sandbox-source logs were lost.
Unary RPCs like `GetSandboxPolicy` were unaffected because they complete in milliseconds and never sit idle.
## Test Plan
- Deployed to local cluster, verified `[sandbox]` logs appear in `nav sandbox logs` output
- Killed `navigator-0` while sandboxes were running; sandbox stderr showed `log push stream lost, reconnecting...` followed by successful reconnection
- Both `[gateway]` and `[sandbox]` log sources confirmed present after server recovery
> **🏗️ build-from-issue-agent**
Closes#71
## Summary
Add dynamic shell completion support using `clap_complete`'s `CompleteEnv`, plus a `navigator completions <shell>` subcommand (rustup-style) that outputs completion scripts to stdout with detailed per-shell setup instructions in `--help`.
Completions are dynamic — the shell calls back into the `navigator` binary on each tab-press, so they are always in sync with the installed version. This also lays the groundwork for future runtime completers (e.g., listing sandbox names on tab, tracked in #83).
## Changes Made
- `Cargo.toml`: Added `clap_complete = { version = "4.5", features = ["unstable-dynamic"] }` to workspace dependencies
- `crates/navigator-cli/Cargo.toml`: Added `clap_complete` dependency
- `crates/navigator-cli/src/main.rs`:
- Added `CompleteEnv::with_factory(Cli::command).complete()` at the top of `main()` (before TLS/argument parsing)
- Added `Completions { shell }` subcommand that re-invokes the binary with `COMPLETE=<shell>` and pipes the output
- Added `CompletionShell` enum (bash, fish, zsh, powershell)
- Added `COMPLETIONS_HELP` with per-shell setup instructions (similar to `rustup completions --help`)
- Added 3 unit tests
- `CONTRIBUTING.md`: Added shell completions section documenting `nav` wrapper setup
## Usage
```bash
# Generate completions
navigator completions fish > ~/.config/fish/completions/navigator.fish
navigator completions bash > ~/.local/share/bash-completion/completions/navigator
navigator completions zsh > ~/.zfunc/_navigator
navigator completions powershell >> $PROFILE
# See full per-shell instructions
navigator completions --help
# For the nav wrapper (rewrite registration target)
navigator completions fish | sed 's/--command navigator/--command nav/' > ~/.config/fish/completions/nav.fish
```
## Design Decision: Why stdout-only (no auto-install)
We researched how 12 popular CLI tools handle completion installation:
| Tool | Command | Writes files? | Modifies RC? |
|------|---------|---------------|-------------|
| gh, glab, kubectl, docker, rustup, mise, starship | `completion <shell>` | stdout only | No |
| atuin | `gen-completions --shell <shell>` | stdout (`--out-dir` option) | No |
| gcloud | `install.sh` | Yes | **Yes** |
| Amazon Q | `q integrations install` | Yes | **Yes** |
**10 of 12 tools** follow the same pattern: output to stdout, never modify RC files, explicit shell selection, clear `--help` instructions. Only gcloud and Amazon Q modify RC files (and both have been criticized for it). Our approach matches the industry consensus.
## Deviations from Plan
- The original issue proposed a static `completions <shell>` subcommand using `clap_complete::generate()`. After discussion, we switched to dynamic `CompleteEnv` which is always up-to-date and enables future runtime completers.
- Added a `navigator completions <shell>` subcommand (not in original plan) for discoverability — it wraps the `COMPLETE` env var mechanism with a familiar CLI UX.
## Tests Added
- **Unit:** `cli_debug_assert` — validates CLI structure; `completions_engine_returns_candidates` — smoke test confirming the completion engine returns subcommand candidates; `completions_subcommand_appears_in_candidates` — verifies `completions` appears in tab-completion results
- **Integration:** N/A
- **E2E:** N/A
## Verification
- [x] All tests passing
- [x] Pre-commit checks passing (clippy warnings are pre-existing, not from this change)
- [x] CONTRIBUTING.md updated with nav wrapper completion setup
Closes#34
## Summary
- Add `nav ssh-proxy` name mode (`--cluster`/`--name`) and `nav sandbox ssh-config` for VS Code Remote-SSH integration
- Fix VS Code Remote-SSH platform misdetection (was defaulting to Windows/PowerShell)
- Fix sandbox network policy for VS Code server connectivity
- Fix sandbox user shell configuration
## Problem
VS Code Remote-SSH was failing to connect to Navigator sandboxes due to three issues:
1. **Platform misdetection**: VS Code sends `uname -rsv` after a `ready:` marker and expects a response within ~1ms. Forced PTY allocation in `shell_request`, interactive bash (`-i`), and leaked supervisor env vars added enough latency that VS Code timed out and defaulted to "Platform: windows", then tried `powershell`.
2. **Broken network policy**: The `vscode` entry in `dev-sandbox-policy.yaml` was at the YAML top-level instead of nested under `network_policies:` (indentation bug). The VS Code server binary also needed a glob entry.
3. **No login shell**: The sandbox user had `/usr/sbin/nologin` as its shell.
## Changes
| File | Change |
|------|--------|
| `crates/navigator-sandbox/src/ssh.rs` | Remove forced PTY in `shell_request`, remove `-i` flag from pipe exec, add `env_clear()` + explicit `SHELL`/`PATH` to both shell spawn paths |
| `deploy/docker/sandbox/Dockerfile.base` | Change sandbox user shell to `/bin/bash`, set home to `/sandbox`, add `.bashrc`/`.profile` |
| `dev-sandbox-policy.yaml` | Fix `vscode:` indentation under `network_policies:`, add `/sandbox/.vscode-server/*` binary glob |
| `crates/navigator-cli/src/main.rs` | Add `--cluster`/`--name` to `SshProxy`, add `SshConfig`/`Policy`/`Logs` subcommands (rebase merge) |
| `crates/navigator-cli/src/ssh.rs` | Update `print_ssh_config` to use `--cluster` flag |
| `examples/vscode-remote-sandbox.md` | User-facing example and docs |
## How to use
```bash
# Create a persistent sandbox
nav sandbox create --keep my-sandbox
# Generate SSH config
nav sandbox ssh-config my-sandbox >> ~/.ssh/config
# Open VS Code
code --remote ssh-remote+nav-my-sandbox /sandbox
```
## Testing
- All 307 tests pass
- Clippy clean on modified crates
Closes#78
## What You Can Do
### Live policy updates
Update a running sandbox's network policy without recreating it:
```bash
nav sandbox policy set <sandbox> --policy new-policy.yaml --wait --timeout 60
```
The sandbox hot-reloads the policy within 30s (configurable). On failure, the previous policy stays active (last-known-good).
Idempotent — submitting the same policy twice returns the existing version:
```
✓ Policy version 3 submitted (hash: a1b2c3d4e5f6)
$ nav sandbox policy set test --policy same.yaml
· Policy unchanged (version 3, hash: a1b2c3d4e5f6)
```
### Policy history & inspection
```bash
nav sandbox policy list <sandbox> # version history with status
nav sandbox policy get <sandbox> # current policy metadata
nav sandbox policy get <sandbox> --full # print full policy as YAML
nav sandbox policy get <sandbox> --rev 2 --full # specific revision as YAML
```
### Sandbox logs
Stream logs from both the gateway and sandbox supervisor in one view:
```bash
nav sandbox logs <sandbox> # one-shot, last 2000 lines
nav sandbox logs <sandbox> --tail # live streaming
nav sandbox logs <sandbox> --source sandbox # supervisor logs only
nav sandbox logs <sandbox> --source gateway # gateway logs only
nav sandbox logs <sandbox> --level warn # warnings and errors only
nav sandbox logs <sandbox> --since 5m # last 5 minutes
nav sandbox logs <sandbox> --tail --source sandbox --level info
```
Each log line is tagged with its source and includes structured fields:
```
[1772055394.673] [sandbox] [INFO ] [navigator_sandbox::proxy] CONNECT action=allow dst_host=api.anthropic.com dst_port=443 policy=claude_code
[1772055061.005] [gateway] [INFO ] [navigator_server::grpc] GetSandboxPolicy served from policy history
```
---
## Implementation
### Proto changes
- 7 new RPCs: `UpdateSandboxPolicy`, `GetSandboxPolicyStatus`, `ListSandboxPolicies`, `ReportPolicyStatus`, `GetSandboxLogs`, `PushSandboxLogs` (client-streaming)
- `SandboxLogLine`: added `source` (gateway/sandbox), `fields` (structured key-value map)
- `WatchSandboxRequest`: added `log_since_ms`, `log_sources`, `log_min_level`
- `GetSandboxLogsRequest`: added `sources`, `min_level`
- `PolicyStatus` enum, `SandboxPolicyRevision` message
- `Sandbox.current_policy_version` field
### Server
- **Policy persistence**: New `sandbox_policies` table (SQLite + Postgres) with per-sandbox monotonic versions, status tracking, and policy hash
- **UpdateSandboxPolicy**: Validates static field immutability (filesystem/landlock/process), network mode consistency (Block↔Proxy), deterministic hash comparison for idempotent updates
- **Lazy backfill**: First `GetSandboxPolicy` call creates version 1 from `spec.policy` for existing sandboxes
- **Log broker**: `TracingLogBus::publish_external()` injects sandbox-pushed logs into the same broadcast channel + tail buffer (2000 lines). Server forces `source="sandbox"` and `sandbox_id` on all pushed logs
- **Source/level filtering**: Applied server-side in both `GetSandboxLogs` and `WatchSandbox` streams
- **Version supersession**: When a new version is loaded, all older pending+loaded versions are marked superseded
### Sandbox
- **`OpaEngine::reload_from_proto()`**: Full `from_proto()` pipeline (L7 validation, access preset expansion) with atomic engine swap. On failure, previous engine untouched (LKG)
- **Policy poll loop**: Background task polls every 30s (configurable via `NAVIGATOR_POLICY_POLL_INTERVAL_SECS`), reports status via `ReportPolicyStatus` RPC
- **`LogPushLayer`**: Tracing layer captures events at INFO+ (configurable via `NAVIGATOR_LOG_PUSH_LEVEL`), sends structured fields via `PushSandboxLogs` client-streaming RPC. Background task batches 50 lines / flushes every 500ms. Best-effort (drops on full channel, never blocks)
- **`CachedNavigatorClient`**: Persistent mTLS channel for both policy polling and log push
### Database migration
- `002_create_sandbox_policies.sql` (SQLite + Postgres)
## Tests
- **Unit**: 8 policy persistence tests (put/get/list/status/supersede/isolation)
- **Integration**: 4 test files updated with new RPC stubs
- **E2E**: `test_live_policy_update_and_logs` — full lifecycle: create → set same (unchanged) → push new → wait for load → verify connectivity → push same (unchanged) → verify history → fetch logs
## Documentation
- `architecture/sandbox.md`: Log streaming architecture, LogPushLayer, push task, server broker, source tagging, structured fields, CLI filtering, failure modes
- `architecture/security-policy.md`: Live update semantics, deterministic hashing, CLI filter flags, policy inspection
- `architecture/plans/issue-78-sandbox-log-streaming.md`: Design plan for log streaming
## Security
- Trust boundary documented: shared mTLS cert model (per-sandbox auth tracked in #80)
- Server forces `source="sandbox"` and `sandbox_id` on pushed logs (can't impersonate gateway or other sandboxes)
- Per-batch line cap (100) prevents flooding
> **🏗️ build-from-issue-agent**
Closes#81
## Summary
The proxy was forwarding the client's original `model` field in the request body unchanged to the backend, causing backends to reject requests for models they don't serve. Now `proxy_to_backend()` always sets the `"model"` field in the JSON body to the route's configured model before forwarding. This covers all 4 supported inference protocols (openai_chat_completions, openai_completions, openai_responses, anthropic_messages).
## Changes Made
- `crates/navigator-router/src/backend.rs`: Added JSON body model override — parses the body and always sets `"model"` to `route.model` (replaces if present, inserts if absent). Non-JSON bodies pass through unchanged.
- `crates/navigator-router/tests/backend_integration.rs`: Added 2 integration tests using wiremock's `body_partial_json` matcher
## Deviations from Plan
Minor improvement: the plan specified only replacing `"model"` when already present. Changed to always set it, since all inference protocols require the field and omitting it would cause backend rejection.
## Tests Added
- **Unit:** N/A
- **Integration:** `proxy_overrides_model_in_request_body` and `proxy_inserts_model_when_absent_from_body` in `crates/navigator-router/tests/backend_integration.rs`
- **E2E:** N/A
## Documentation Updated
- `architecture/sandbox.md`: Updated proxy forwarding description to include model ID rewriting
- `architecture/inference-routing.md`: Updated backend.rs description, sequence diagram, and backend proxying steps
## Verification
- [x] All tests passing
- [x] Pre-commit checks passing (clippy, fmt)
- [x] Architecture documentation updated
## Summary
Fixes "Connection refused" when running `nav sandbox create` or `nav sandbox connect` on local Docker-mapped clusters where the host port differs from the container port.
## Root Cause
`resolve_ssh_gateway()` in `crates/navigator-cli/src/ssh.rs` overrides the gateway host from the cluster URL when the server returns a loopback address, but kept the server's **internal** port (e.g. 8080). For Docker-mapped clusters where the host port (e.g. 29581) differs from the container port (8080), port 8080 is not exposed on the host, causing ECONNREFUSED.
## Fix
When the gateway host is loopback, use the cluster URL's port in both cases (local and remote override). The cluster URL represents the externally reachable Docker-mapped or tunnel endpoint.
## Test plan
- [x] `nav sandbox create --name test` on local cluster with non-default port mapping
- [x] Verified SSH tunnel connects through the Docker-mapped port
> **🏗️ build-from-issue-agent**
Closes#79
## Summary
Replaces the gateway-proxied inference model (`ProxyInference` RPC) with sandbox-local execution. The sandbox now resolves routes from either a standalone YAML route file or a cluster bundle fetched via the new `GetSandboxInferenceBundle` RPC, then forwards requests directly to inference backends using `navigator-router`.
### Benefits of removing the gRPC hop
Moving inference execution from the gateway to the sandbox eliminates the gRPC round-trip for every inference request:
- **No payload size limits**: Direct HTTP forwarding replaces protobuf-over-gRPC serialization.
- **Streaming-ready**: Preserves connection semantics for future SSE support (e.g., `stream: true`).
- **Lower latency**: Direct sandbox → backend path instead of sandbox → gateway → backend round-trip.
- **Reduced gateway load**: Gateway only serves the control-plane bundle delivery RPC.
- **Simpler error model**: HTTP status codes flow through directly without gRPC wrapping.
## Changes Made
- **Proto**: Removed `ProxyInference` RPC. Added `GetSandboxInferenceBundle` RPC for route bundle delivery.
- **Gateway**: Replaced `proxy_inference` with `get_sandbox_inference_bundle`. Removed `Router` and `navigator-router` dependency — gateway is now control-plane only for inference.
- **Router**: Switched config file format from TOML to YAML. Added `Debug` impl that redacts API keys.
- **Sandbox**: New `InferenceContext` with local `Router` + route cache. Routes loaded from `--inference-routes` file (standalone) or cluster bundle via gRPC, with 30s background refresh. New `--inference-routes` / `NAVIGATOR_INFERENCE_ROUTES` CLI arg.
- **Dev sandbox**: Added `inference-routes.yaml` at repo root with default NVIDIA NIM route. `mise run sandbox` now mounts it automatically and supports `-e` flag to forward env vars (e.g., `mise run sandbox`).
- **Example**: Added standalone example (`examples/inference/routes.yaml`) and rewrote README to cover both standalone and cluster workflows.
- **E2E tests**: Updated existing test names/docs. Added tests for Anthropic messages protocol and multi-route policy filtering.
## Tests Added
- **Unit (21 new):** `router_error_to_http` (6 variants), `load_from_file` YAML round-trip (5), `resolve_sandbox_inference_bundle` gRPC handler (6), `build_inference_context` route loading (4)
- **E2E (2 new):** Anthropic messages protocol routing, route filtering by `allowed_routes`
- **Total:** 188 unit tests passing across 3 crates
## Documentation Updated
- `architecture/gateway.md`: Removed router, updated Inference Service
- `architecture/sandbox.md`: Updated orchestration flow, InferenceContext, route loading
- `architecture/inference-routing.md`: Complete rewrite for sandbox-local architecture
## Verification
- [x] All unit tests passing (188 across 3 crates)
- [x] Pre-commit Rust checks passing (fmt, clippy, tests)
- [x] Architecture documentation updated
- [x] E2E tests updated with new test cases
- [x] Standalone example and documentation added
- [x] Default inference-routes.yaml with env passthrough for mise run sandbox
## Summary
- Track last deployed local changes and only redeploy new deltas in auto mode.
- Add per-component fingerprints (server/sandbox/helm) persisted to .cache/cluster-deploy-fast.state.
- Update docs for the new incremental deploy behavior.
```
mise run cluster
....
________________________________________________________
Executed in 75.21 secs fish external
usr time 578.03 millis 0.25 millis 577.78 millis
sys time 364.45 millis 1.22 millis 363.22 millis
```
with a change to `navigator-sandbox` (triggers rust recompile)
```
⟩ time mise run cluster:deploy
[cluster:deploy] $ build/scripts/cluster-deploy-fast.sh
Change detection took 0s
Fast deploy plan:
build server: 0
build sandbox: 1
helm upgrade: 1
...
________________________________________________________
Executed in 25.88 secs fish external
usr time 618.91 millis 0.24 millis 618.67 millis
sys time 557.71 millis 1.54 millis 556.18 millis
```
(most time is spent in docker build)
## Test Plan
- mise run pre-commit
Closes#61
## Summary
- harden provider discovery home expansion to reject path-escape components like dot-dot
- reuse the same safe expansion logic in MockDiscoveryContext
- add regression tests for escaped paths and valid home expansions
- document the new config-path expansion safety rule
## Test Plan
- mise run pre-commit
## Summary
- Rewrites the README quickstart with accurate, tested setup steps (Docker prerequisite, venv/uv usage)
- Moves the "developing from source" instructions into CONTRIBUTING.md
- Removes the manual cluster deploy step from quickstart (cluster bootstraps automatically on sandbox create)
- Fixes empty bash array expansion in Docker build scripts that broke on macOS default bash 3.2 (`set -u` treats `"${EMPTY_ARRAY[@]}"` as unbound)
## Changes
- **README.md**: Replace placeholder quickstart with working install and sandbox create steps; remove manual cluster deploy step since it happens automatically
- **CONTRIBUTING.md**: Add "developing from source" section with mise setup instructions
- **build/scripts/docker-build-cluster.sh**: Use `${arr[@]+"${arr[@]}"}` pattern for safe empty array expansion
- **build/scripts/docker-build-component.sh**: Same fix
Closes#48, #52
## Summary
- Replace the envoy-gateway-based TLS setup with inline PKI generation during cluster bootstrap, generating CA, server, and client certificates directly in the `navigator-bootstrap` crate
- Remove all envoy gateway Helm templates (`gateway.yaml`, `gatewayclass.yaml`, `grpcroute.yaml`, PKI job, traffic policies) and the `Dockerfile.pki-job`
- Add native mTLS support to the navigator server with `tokio-rustls`, mounting client TLS certs as volumes into sandbox pods
- Update cluster entrypoint, healthcheck, and deploy scripts to work with the new direct-TLS architecture
- Add TLS security e2e test and fix formatting/clippy warnings
## Test Plan
- All unit tests pass (`cargo test --workspace`)
- Clippy clean (`cargo clippy --workspace --all-targets`)
- Format clean (`cargo fmt --all -- --check`)
- Python tests pass (`uv run pytest python/`)
- Full `mise run pre-commit` passes
## Why
Sandbox proxy policy evaluation was re-hashing unchanged binaries on every CONNECT request. In practice, this made the warm path much slower than necessary.
## What changed
- `BinaryIdentityCache` now stores a file fingerprint with the cached TOFU hash.
- `verify_or_cache()` returns the cached hash immediately when fingerprint is unchanged.
- If fingerprint changes, it re-hashes and compares to the cached hash to preserve tamper detection.
## Scope (fix only)
This MR intentionally contains only the identity-cache fix and docs for that behavior.
It does **not** include benchmark harness or perf-instrumentation setup changes.
## Benchmark results (same environment)
Pre-fix vs post-fix p50:
- `sandbox_connect_local`: `1784.8 ms` -> `1696.4 ms`
- `sandbox_connect_network_via_proxy`: `5692.9 ms` -> `1860.3 ms`
- Derived proxy-path network overhead: `3908.1 ms` -> `163.9 ms` (**95.8% reduction**)
Additional timing signal from perf logs:
- `PERF_SANDBOX_PROXY_CONNECT decision_ms` p50: `3742 ms` -> `1 ms`
## Validation
- `cargo fmt`
- `cargo test -p navigator-sandbox identity::tests`
## Summary
- Upgrade the `github` network policy from L4 pass-through to L7 TLS-terminated with Git Smart HTTP rules
- Allow read operations (clone/fetch/pull) for all repos via `git-upload-pack`
- Allow write operations (push) only to `johntmyers/alpha-claw` via scoped `git-receive-pack` rule
- All other repos (e.g. `bravo-claw`) are implicitly denied push access
- Update OPA tests to reference the `github` policy (replacing removed `gitlab` policy)
## Test plan
- [x] `mise run pre-commit` passes (119 sandbox tests, clippy, fmt, python)
- [x] Verified live: `git clone` works for bravo-claw
- [x] Verified live: `git push` blocked for bravo-claw
- [x] Verified live: `git push` allowed for alpha-claw
## Summary
- Fix L7 REST relay response framing so chunked HTTP responses are parsed by chunk-size + trailer termination, instead of waiting for socket close semantics.
- Prevent ~30s stalls observed in `gh auth status` when talking to `api.github.com` through the sandbox proxy.
- Keep detailed debug timing around response framing/chunk completion to make future networking investigations easier.
## Additional changes in this branch
- Adds sandbox policy generation skill docs under `.agent/skills/generate-sandbox-policy/`.
- Includes related sandbox config/image updates in `dev-sandbox-policy.yaml` and `deploy/docker/sandbox/Dockerfile.base`.
## Test plan
- `cargo test -p navigator-sandbox l7::rest::tests:: -- --nocapture`
- `cargo test -p navigator-sandbox -- --nocapture`
- `mise run pre-commit`
- Manual: rebuilt `navigator-sandbox:dev`, ran `gh auth status` in sandbox, and confirmed the previous long proxy delay no longer reproduces.
Closes#67
## Summary
Implements transparent inference interception and routing for sandboxed AI agents. The sandbox proxy intercepts outbound AI SDK calls (OpenAI, Anthropic) and reroutes them through the gateway to policy-controlled backends — enabling organizations to redirect inference traffic to local or self-hosted models without modifying agent code.
**Decision model** — a tri-state OPA evaluation for every CONNECT request:
1. Binary + endpoint explicitly allowed in `network_policies` → **allow** (pass through)
2. Not explicitly allowed + `inference.allowed_routes` configured → **inspect for inference** (TLS intercept, detect API patterns, route through gateway)
3. Otherwise → **deny**
No endpoint declarations or binary lists needed for inference routing. Just configure `inference.allowed_routes`.
## Key Changes
### Sandbox (interception)
- **OPA policy**: New `network_action` Rego rule with three outcomes (`allow`, `inspect_for_inference`, `deny`). New `NetworkAction` enum replaces `PolicyDecision.allowed` bool for the proxy's main decision path.
- **Proxy**: New `InspectForInference` path — TLS-terminates client, parses HTTP, detects inference API patterns (`POST /v1/chat/completions`, `/v1/completions`, `/v1/messages`), strips auth headers, forwards via gRPC.
- **New module**: `l7/inference.rs` — `InferenceApiPattern`, `detect_inference_pattern()`, HTTP request/response parsing.
- **gRPC client**: New `proxy_inference()` for sandbox→gateway forwarding.
- **Sandbox init**: Creates OPA engine when inference is configured, even without `network_policies`.
### Gateway (dispatch)
- **InferenceService**: `ProxyInference` RPC loads sandbox policy, resolves allowed routes, dispatches to router. Full CRUD for inference routes.
- **Proto**: `InferenceRoute`, `InferenceRouteSpec`, `ProxyInferenceRequest/Response`, Inference gRPC service.
### Router (backend proxying)
- **New crate**: `navigator-router` with `Router`, `proxy_with_candidates()`, protocol-based route selection, backend HTTP proxying with auth header rewriting.
- **Mock support** for testing (`mock://` scheme).
### CLI
- `nav inference create/update/delete/list` commands for route management.
### Python SDK
- Updated protobuf bindings. Removed old `inference.py` client (replaced by transparent interception).
### Documentation
- New `architecture/inference-routing.md` — end-to-end system documentation.
- Updated `architecture/sandbox.md` — proxy, OPA, and source index sections.
- Updated `architecture/README.md` — new subsystem overview and diagram.
## Addendum: Chunked Transfer Compatibility
This branch now also fixes intercepted SDK requests that send chunked request bodies:
- `inspect_for_inference` now accepts `Transfer-Encoding: chunked` and decodes chunked request bodies before forwarding to the gateway
- Removed the prior `411 Length Required` response for chunked intercepted requests
- Added request/response header sanitization for framing and hop-by-hop headers (`content-length`, `transfer-encoding`, `connection`, etc.) to keep forwarded requests and returned responses valid
- Added unit tests for chunked parsing and header sanitization
Note: this improves compatibility for streaming-style SDK request patterns; true token-by-token passthrough response streaming is still a separate follow-up.
## Minimal Policy for Inference Routing
```yaml
inference:
allowed_routes:
- local
```
Any outgoing connection from a binary not explicitly allowed in `network_policies` will be intercepted and checked for inference API patterns.
## Test Plan
- [x] `cargo test --workspace` — all tests pass
- [x] `mise run pre-commit` — all checks pass
- [x] E2E: OpenAI chat completions routed through gateway
- [x] E2E: Anthropic messages routed through gateway
- [x] E2E: Python OpenAI SDK from sandbox (`examples/inference/inference.py`)
- [x] E2E test: `e2e/python/test_inference_routing.py`
Closes#20, #36, #46
## Summary
- Replace the external `rsync` dependency with built-in tar-over-SSH for file synchronization, using the `tar` crate to stream archives through the existing SSH proxy tunnel
- Add a new `nav sandbox sync` subcommand supporting bidirectional file transfer (`--up` to push local files, `--down` to pull sandbox files)
- Add sync workflow example in `examples/sync-files.md`
## Changes
| File | Change |
|---|---|
| `crates/navigator-cli/Cargo.toml` | Add `tar = "0.4"` dependency |
| `crates/navigator-cli/src/ssh.rs` | Remove `sandbox_rsync()`, add `sandbox_sync_up()`, `sandbox_sync_down()`, `sandbox_sync_up_files()` |
| `crates/navigator-cli/src/run.rs` | Add `sandbox_sync_command()` dispatcher, update re-exports, update `--sync` call site |
| `crates/navigator-cli/src/main.rs` | Add `Sync` variant to `SandboxCommands` with `--up`/`--down` flags |
| `architecture/sandbox-connect.md` | Rewrite File Sync section for tar-over-SSH and new command |
| `examples/sync-files.md` | New example walkthrough for sync workflows |
## New UX
```bash
# Push local files up to sandbox
nav sandbox sync my-sandbox --up ./src /sandbox/src
# Pull sandbox files down to local
nav sandbox sync my-sandbox --down /sandbox/output ./output
# Existing --sync flag on create still works (now uses tar internally)
nav sandbox create --sync -- python main.py
```
## Design
- **No new host dependencies**: `tar` crate handles archive creation/extraction in Rust; `ssh` is already required. The sandbox side uses the `tar` binary from the base image.
- **No new server-side infrastructure**: reuses the existing SSH tunnel, no new gRPC RPCs or gateway changes.
- **Bidirectional**: the same tar-pipe pattern works for both push and pull by reversing which side produces and which consumes the archive.
- **No compression** for v1 — the SSH tunnel is local-network; can add gzip via `flate2` later.
## Test Plan
- All existing tests pass (`mise run pre-commit` green: fmt, clippy, 206 Rust tests, 6 Python tests)
- Manual e2e testing required for actual sandbox sync operations
The sandbox Dockerfile was moved from deploy/docker/Dockerfile.sandbox to
deploy/docker/sandbox/Dockerfile.base but the multiarch publish script
was not updated, causing publish_main_artifacts to fail.
Drew Newberry