The activation transaction and the release-descriptor probe validate the
full ancestor chain of everything they stage: every path component must
have a trusted owner and grant mutation rights only to trusted identities.
That walk is deliberate TOCTOU defense and stays strict. The staging
directory, however, came from the process temp root, which routinely lives
in a tree that legitimately fails the walk through no fault of ours: under
msys2 TMP resolves inside the msys install tree (C:\msys64\tmp), and on
GitHub runners inside the runner work directory, both of which carry broad
inherited grants such as Authenticated Users modify on an upper component.
The staged pair then failed validation ("launcher descriptor candidate is
unsafe" / the smoke Phase 8 --version probe refusal) even though the staged
files themselves were private.
Add cli_windows_staging_root(): stage install and update transactions under
%LOCALAPPDATA%\codebase-memory-mcp, whose ancestor chain is user-owned end
to end, so the strict walk passes by construction instead of by environment
luck. Falls back to USERPROFILE\AppData\Local and finally the process temp
root when the environment is degraded.
Also make the staging observable where it previously failed blind:
cli_windows_stage_private_file reports which transaction step failed with
the OS error code, and cbm_mkdtemp warns once if the private security
descriptor cannot be applied and it falls back to default directory
security. A blind boolean here cost several blind CI round-trips.
win.sh: scp targets need native C:/ paths on Windows OpenSSH.
Signed-off-by: Martin Vogel <martin.vogel.tech@gmail.com>
The Phase 8 staging diagnostic added two long lines that clang-format-20
(the CI formatter) wraps differently. Reformat them so lint passes; no
behavior change.
Signed-off-by: Martin Vogel <martin.vogel.tech@gmail.com>
cli_windows_prepare_install_pair folded launcher staging, payload staging,
the payload fingerprint match, and the staged-launcher --version probe into
one bool, so the caller could only report "launcher/payload source pair did
not remain runnable after private staging" without saying which sub-step
failed. Windows cbm_exec_no_shell does not inherit the child's stderr, so the
launcher's own refusal reason never reaches the log either. Split the checks
so each failing sub-step prints a distinct error, turning the opaque Phase 8
smoke failure into a precise signal on the next run.
Signed-off-by: Martin Vogel <martin.vogel.tech@gmail.com>
The Windows managed-install path (cli_windows_managed_install_activate)
printed its plan as "Would install managed launcher -> ..." but never
the words "dry-run", unlike the POSIX path's "(dry-run — no files were
modified)" - so `install --dry-run` did not clearly indicate dry-run
mode (smoke Phase 6a, confirmed by echoing the captured output).
Append the same dry-run indicator to the Windows plan line.
Signed-off-by: Martin Vogel <martin.vogel.tech@gmail.com>
A dry-run mutates nothing (every install mutation is guarded by
!dry_run), but on Windows the activation plan's PATH/agent-config probe
can return non-OK, and the caller then returned early WITHOUT the
dry-run summary - so `install --dry-run` neither indicated dry-run mode
nor exited success (smoke Phase 6a). For a dry-run, emit the
"(dry-run — no files were modified)" indicator and return CLI_OK on
that path (the plan simply noted something), and log the underlying
status to stderr for triage. POSIX is unaffected (its dry-run
activation returns OK, so the branch is never taken).
Signed-off-by: Martin Vogel <martin.vogel.tech@gmail.com>
Extensionless bash shims under %USERPROFILE%\.claude\hooks triggered
the Windows 'How do you want to open this file?' dialog whenever an
editor (Cursor) scanned the hooks dir, and could not execute without
bash anyway. All three Claude hook scripts (discovery gate, session
reminder, subagent reminder) now install as .cmd files with cmd syntax
on Windows - @echo off bodies, caret-escaped pipes in the reminder
text, 2>/dev/null redirects - and the install removes the pre-existing
extensionless twin so upgrades stop triggering the dialog. The
settings.json command paths follow automatically since they are built
from the same script-name macros. POSIX behavior is byte-identical and
guarded (no .cmd twin).
Closes#929
Signed-off-by: Martin Vogel <martin.vogel.tech@gmail.com>
Aider has no MCP support - it reads CONVENTIONS.md but can only run
shell commands. The installer wrote the shared MCP-tool-centric
instructions there, telling the model to call search_graph(...) etc.,
tools Aider cannot invoke; a reporter reasonably asked how that was
ever supposed to work (#1032).
Aider now gets a dedicated variant: the same discovery priority
expressed as runnable codebase-memory-mcp cli commands (usable via
/run), including first-index and list_projects bootstrapping, with the
no-MCP constraint stated explicitly.
Closes#1032
Signed-off-by: Martin Vogel <martin.vogel.tech@gmail.com>
cbm_cli_build_args_json typed unknown flags as strings and shipped them
as JSON args the server ignores: 'trace_path --max-depth 1' succeeded
but traced at the default depth 3 — silently-wrong output that inflated
an agent evaluation by thousands of tokens of unwanted hop-2/3 rows.
Unknown flags for a known tool now fail loudly with a Levenshtein
nearest-flag suggestion:
unknown flag --max-depth for this tool (did you mean --depth?) — run
'cli trace_path --help' for the supported flags
Unknown-tool (no schema), raw-JSON, --args-file, stdin, and supervisor
worker argv paths are unaffected — none of them route through the flag
converter.
Reproduce-first: cli_build_args_json_unknown_flag_rejected is RED on
the permissive code (JSON produced) and GREEN with the rejection.
Closes#997
Signed-off-by: Martin Vogel <martin.vogel.tech@gmail.com>
Four refinements to the coverage signal:
- extraction: subtract DEFINITE recovery before flagging. Tree-sitter
error recovery plus the ERROR-descending def walker often re-extract
constructs inside a failed region (verified: `def broken(:` comes
back as a def); a region whose every line is covered by definitions
that START inside it is not a miss, and a fully recovered file is not
flagged at all. Container defs (Module/Package) don't count as
evidence, and partially covered regions stay flagged — the
#ifdef-split case keeps its flag because the first branch's function
is genuinely lost
- naming: the query_graph option is graph="missed" (the graph shows
ONLY misses — "coverage" was misleading); shadow project renamed to
"<name>::missed"; tool descriptions updated so agents discover both
the option and its semantics
- hook: the CLI-installed PreToolUse augmenter now also matches Read
and injects a coverage note when the file being read is listed as not
fully indexed ("line ranges X-Y could not be parsed — the file
content you are reading is ground truth"). Safe against the old
issue-362 hazard: the augmenter is structurally non-blocking (always
exit 0, additionalContext only), mirroring the Gemini matcher that
already includes read_file; matcher upgrade bookkeeping updated
- ui: "Missed files" toggle in the graph sidebar renders the miss
graph as a second graph option — /api/layout gains graph=missed
(same db file, shadow-project scoping; base project name validated
as before)
Tests: recovery-subtraction cases (recovered def unflagged, garbage
region flagged, trailing recovered defs keep the flag), CLI matcher
tests updated, e2e resilience fixtures switched to an unrecovered
miss; verified end-to-end: hook emits the note for a flagged file and
stays silent otherwise, /api/layout?graph=missed serves the miss graph
and the code graph is unchanged, frontend build + tests green.
Refs #963
Signed-off-by: Martin Vogel <martin.vogel.tech@gmail.com>
cbm_cli_sha256_file shelled out to sha256sum/shasum, which is unportable:
the tool may be absent, and under cmd.exe the single-quoted path is not
dequoted so the hash of a downloaded release could never be computed on
Windows. Add a small in-process SHA-256 (FIPS 180-4, src/foundation/sha256.c)
and use it directly — no external tool, no per-OS branching, no popen. The
checksum test now covers three NIST vectors (empty, "abc", and a 56-byte
input that forces padding into a second block).
Signed-off-by: Martin Vogel <martin.vogel.tech@gmail.com>
The macOS digest path invoked `shasum -a CBM_SZ_256` — a literal string
(not a macro; string literals do not expand) naming an invalid algorithm, so
the digest always failed on macOS, and the self-update path then treated a
could-not-verify result as non-fatal and installed the download regardless.
Use the correct algorithm, and install only a positively-verified download:
a mismatch, a missing checksum entry, or an unavailable hash tool now all
abort. Adds a known-vector digest regression test.
Signed-off-by: Martin Vogel <martin.vogel.tech@gmail.com>
Add the auto_watch config key (default: true) and route both session
watcher-registration sites in mcp.c (autoindex completion and the
already-indexed connect path) through a register_watcher_if_enabled()
helper. When disabled, the skip is logged as a short structured line
(watcher.register.skipped reason=auto_watch_off).
Default is TRUE (opt-out): existing users keep background auto-sync
unchanged; the knob exists for multi-project users who want each
session contained to explicit indexing.
Distilled from the auto_watch half of PR #625. The PR's bundled
project-index dedup commit is excluded here — it belongs to #754's
own review. Unlike the original PR, no watch registration is added
to the index_repository handler (none exists on main; adding one
would be new behavior beyond the gate), and the default is flipped
from off to on to preserve current behavior.
Tests: default-unset registers the watcher on connect (guards the
no-behavior-change promise); auto_watch=false registers nothing —
verified RED against the ungated production code, GREEN with the gate.
Co-authored-by: Andy11-cpu <canada11@duck.com>
Signed-off-by: Martin Vogel <martin.vogel.tech@gmail.com>
On NixOS (and other non-FHS systems) /bin/bash does not exist, so
scripts with an absolute shebang fail to run. Switch the remaining
holdouts to /usr/bin/env bash: eleven scripts/*.sh,
test-infrastructure/run.sh, and the three Claude Code hook scripts
emitted by src/cli/cli.c (gate, session reminder, subagent reminder).
Distilled from PR #674, with parser-test coverage preserved: the
infra_parse_shell* fixtures in tests/test_pipeline.c intentionally keep
#!/bin/bash so absolute-path shebang extraction stays covered, and
tests/repro fixtures are untouched.
Also replace the GitHub-PAT-shaped fixture string flagged in the #674
thread with an obviously fake placeholder (ghp_FAKE...) that still
matches the ghp_ + 36-alnum secret detector.
Co-authored-by: Sandro Jäckel <sandro.jaeckel@gmail.com>
Signed-off-by: Martin Vogel <martin.vogel.tech@gmail.com>
The truncation check `header_end + (int)comp_size > data_len` was
bypassable for comp_size >= 2^31: the int cast turns the size negative,
the sum drops below data_len, and the entry is accepted -- authorizing
inflate to read up to ~4GB past the downloaded archive buffer during
self-update ZIP extraction. Replace it with
`header_end > data_len || comp_size > (uint32_t)(data_len - header_end)`,
which is overflow-safe and strictly tighter.
Regression guard cli_extract_binary_from_zip_rejects_truncated_deflate_size_over_int_max
builds a 52-byte archive whose entry claims comp_size=0xFFFF0000 with a
self-terminating DEFLATE stream: the old check admits it and extraction
returns non-NULL (verified red pre-fix); the new check rejects it.
Secondary cppcheck cleanups from the same review:
- cli.c: extract zip_read_u16le/zip_read_u32le little-endian helpers
(behavior-identical) and widen zip_extract_entry sizes to size_t with
an explicit UINT_MAX guard before the zlib uInt narrowing.
- cypher.c: drop unreachable IS NULL / IS NOT NULL null guards in
eval_condition (resolve_condition_value result is checked earlier).
Distilled from PR #784.
Co-authored-by: SS-42 <noreply@incogni.to>
Signed-off-by: Martin Vogel <martin.vogel.tech@gmail.com>
The CLI only accepted a single inline JSON argument (cbm cli <tool> '<json>'),
which PowerShell mangles on Windows and which made `cli <tool> --help` return
"repo_path is required" -- the "--help" token was parsed as the JSON args, and
there was no per-tool help at all.
run_cli now resolves tool arguments by precedence: --args-file <path>, a raw
JSON object (back-compat), --flag / --flag=value / bare-boolean flags, piped
stdin, then empty {}. Flags are converted to a JSON object via the tool's
input_schema: kebab-case keys map to snake_case, values are typed
(string/integer/boolean), and repeated flags accumulate into array-typed
properties (--semantic-query a --semantic-query b -> ["a","b"]). stdin and
--args-file are UTF-8-clean and bypass shell quoting entirely (the robust path
on Windows). `cli <tool> --help` prints usage + the tool's flags, generated
from the same input_schema the MCP tools/list advertises -- no hand-maintained
table.
cbm_mcp_tool_input_schema() exposes a tool's schema; cbm_cli_build_args_json()
and cbm_cli_print_tool_help() do the conversion and help.
Reproduce-first: 8 cli tests (string / integer-not-string / bare-boolean /
repeated-array / kebab->snake / --key=value with spaces / bad-positional-errors
/ help) -- RED on a stub, GREEN after. Full suite 5730/0.
Note: true CJK-path support on Windows still needs a UTF-8/wmain entry point
(argv is ANSI-codepage-mangled upstream of parsing); stdin/--args-file mitigate
it and are tracked separately.
Signed-off-by: Martin Vogel <martin.vogel.tech@gmail.com>
QA round 1 (PR #632) flagged that the SubagentStart hook uses matcher "*",
the same matcher a user is most likely to pick for their own catch-all
SubagentStart hook. Because is_cmm_hook_entry keyed ownership on the matcher
string alone, install would remove the user's "*" entry and replace it with
ours, and uninstall could remove the wrong one.
Add an optional `match_command_substr` to the upsert/remove args, threaded into
is_cmm_hook_entry: when set, an entry must ALSO carry a hooks[].command
containing that substring to be claimed as ours. The Claude SubagentStart hook
passes "cbm-subagent-reminder"; all existing callers leave it NULL, preserving
their matcher-only behavior unchanged. Install/uninstall now only ever touch
CMM's own entry and never clobber a foreign "*" hook.
Add cli_claude_subagent_hook_preserves_user_entry covering the case: a
pre-existing user "*" SubagentStart hook survives both install and uninstall.
Signed-off-by: Shane McCarron <shane.mccarron@corvexconnect.com>
Subagents spawned via the Agent tool do not fire SessionStart, so the
existing SessionStart reminder never reaches them — they start without the
codebase-memory-mcp code-discovery guidance and fall back to grep/file-read.
Register a Claude Code SubagentStart hook (matcher "*") that injects a leaner
variant of the protocol via JSON additionalContext, omitting the
index_repository step since the parent session has already indexed the
project. SubagentStart injects context only via a JSON object on stdout, not
plain text, so the generated script emits a static JSON literal — no runtime
escaping and no python3/jq dependency. Advisory only; installed and removed
alongside the SessionStart hook in install_claude_code_config.
Signed-off-by: Shane McCarron <shane.mccarron@corvexconnect.com>
Under -D_GNU_SOURCE with optimization, glibc's string.h macros make
strstr() on a `const char *` (with a constant needle) yield a
`const char *`. gcc 15 then rejects assigning that to `char *` with
-Werror=discarded-qualifiers, breaking the build:
src/cli/cli.c:1503: error: initialization discards 'const'
qualifier from pointer target type [-Werror=discarded-qualifiers]
Minimal repro (gcc 15.2):
const char *c = "hello";
char *p = strstr(c, "ell"); // gcc -D_GNU_SOURCE -O2 -Werror=discarded-qualifiers
codex_hook_strip only reads through begin/end/cut (offset math and
byte comparisons, never writes), so making them `const char *` is the
correct fix and preserves the function's existing const contract
(`content` is already const). The malloc'd output buffer stays char *.
No behavior change; build succeeds on gcc 15.2.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Signed-off-by: Jeremy Maserang <gothuncc@gothaunty.com>
* fix(cli): copy and sign the built binary on install instead of re-signing the stale target
`install` (especially with --force) only ad-hoc-signed whatever already sat at
~/.local/bin/codebase-memory-mcp and never copied the binary the operator ran
install from. An operator who built from source and ran `install -y --force`
was silently left with the OLD binary at the target — running stale code while
believing they had upgraded (#472).
Install now mirrors `update`: detect the running binary, copy it to the
canonical target (skipping the copy when we already ARE the target, which would
otherwise truncate the running file), make it executable, ad-hoc sign the
TARGET, and point all agent configs at the target. A different pre-existing
binary is only replaced with --force or interactive confirmation; --dry-run
reports the intended swap without touching anything.
The copy+sign also resolves the secondary report: a hand-copied clang build is
linker-signed (flags=0x20002) and gets Killed:9 when spawned by an MCP host —
re-signing ad-hoc (flags=0x2) makes it launchable, which install now does for
the placed binary.
Adds cbm_copy_binary_to_target() (with a same-file guard) plus regression tests
covering the binary swap and the self-copy guard.
Closes#472
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Signed-off-by: Shane McCarron <shane.mccarron@corvexconnect.com>
* fix(cli): address QA round 1
Extend cli_install_same_file_guard_issue472 with a symlink case: two distinct
path strings resolving to the same inode (which a non-canonical
cbm_detect_self_path vs the hardcoded target can produce) must also be detected
as same-file and skipped, not truncated. POSIX-only (#ifndef _WIN32).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Signed-off-by: Shane McCarron <shane.mccarron@corvexconnect.com>
* test(cli): guard install copy exec-bit assertion to POSIX
cli_install_copies_binary_to_target_issue472 asserted the executable bit on the
copied target, but cbm_copy_binary_to_target only chmods on POSIX (the exec bit
is not meaningful on Windows, and MinGW stat() derives it from the file
extension). Wrap the stat/S_IXUSR assertion in #ifndef _WIN32 so test-windows
passes; the content and creation assertions still run on all platforms.
Refs #472
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Signed-off-by: Shane McCarron <shane.mccarron@corvexconnect.com>
---------
Signed-off-by: Shane McCarron <shane.mccarron@corvexconnect.com>
Co-authored-by: Shane McCarron <shane.mccarron@corvexconnect.com>
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
The standard linux release binary dynamically links glibc 2.38+ and
GLIBCXX_3.4.32, so it fails to start on Debian 11, Ubuntu 20.04/22.04,
RHEL/Rocky 8/9, Amazon Linux 2, etc. — yet install.sh, the npm and PyPI
wrappers, and the binary's own self-update all fetched it by default,
contradicting the "single static binary" promise.
Point every linux install + self-update path at the fully-static
"-portable" asset (gcc -static), which has no glibc floor. macOS/Windows
are unaffected and unchanged.
- install.sh, pkg/npm/install.js, pkg/pypi _cli.py: select -portable on linux
- src/cli/cli.c: self-update download URL AND checksum archive name both use
-portable on linux (they must match or the update fails checksum verify)
- scripts/smoke-test.sh: assert linux self-update targets the -portable asset
- scripts/ci/check-glibc-compat.sh: new guard — runs the binary inside
debian:bullseye (glibc 2.31) and asserts it starts
- _smoke.yml: run the guard on the portable binary in smoke-linux-portable
Reproduced: standard binary -> "GLIBC_2.38 not found" on glibc 2.31;
portable binary runs cleanly.
Antigravity support targeted the pre-unification layout, so it never actually
configured the CLI:
- detection probed ~/.gemini/antigravity/ — the CLI now installs under
~/.gemini/antigravity-cli/ (brain/, mcp/, settings.json)
- MCP config was written to ~/.gemini/antigravity/mcp_config.json — Antigravity
reads the SHARED ~/.gemini/config/mcp_config.json (mcpServers, command/args)
- the SessionStart hook + AGENTS.md targeted the stale dir
Point detection at ~/.gemini/antigravity-cli/, write the MCP server to the
shared ~/.gemini/config/mcp_config.json (creating ~/.gemini/config if needed),
and place AGENTS.md + the SessionStart reminder under ~/.gemini/antigravity-cli/.
Install, uninstall, install --plan, the detection test, and the README table
updated.
Codex, Gemini CLI, and Antigravity all gained lifecycle-hook support, so give
them the same non-blocking SessionStart reminder Claude Code has — stdout is
injected as session context, nudging the agent to use codebase-memory-mcp
graph tools before grep.
- Codex: append a sentinel-delimited [[hooks.SessionStart]] block to
~/.codex/config.toml (idempotent upsert, preserves other content, removable).
- Gemini CLI: SessionStart hook in ~/.gemini/settings.json, reusing the shared
JSON hook upsert (alongside the existing BeforeTool reminder).
- Antigravity: same JSON hook in ~/.gemini/antigravity/settings.json (it shares
Gemini hook semantics).
The reminder command is written to be valid both as a TOML single-quoted
literal and a JSON string (no single quotes, no newlines). Wired into install,
uninstall, and the install --plan receipt. remove_hooks_json now prunes an
emptied event key so removal leaves no stale "<Event>": [] cruft. README
Multi-Agent table updated. Tests cover Codex upsert/idempotency/removal and
Gemini parity.
An installer that configures MCP entries, instruction files, Skills, and hooks
across many agents should let callers see what it will mutate before it does.
`--dry-run` already lists planned writes in human-readable form; this adds the
machine-readable `agent.install.plan.v1` JSON receipt the issue asked for.
`install --plan` runs the real install dispatch in record-only mode — a global
g_install_plan recorder that each write-site function appends to (at the same
point it would write) while all mutations are disabled — then emits JSON with
agents_detected, config_files_planned, instruction_files_planned,
hooks_planned, writes_started:false, network_after_install:false, and
next_safe_command. Because the plan is recorded on the actual install code
path, it cannot drift from real behavior. No config is written, no index
deleted, no network used.
cbm_build_install_plan_json is exposed for testing; the test asserts the
receipt content AND that building it creates no config files.
cbm_ensure_path appended `export PATH="<dir>:$PATH"` to every shell rc,
including config.fish — which is a syntax error in fish and broke users
fish config. Detect a .fish target and emit `fish_add_path <dir>` instead
(idempotent, prepends only if absent). POSIX shells are unchanged.
The symlink-attack vector (predictable /tmp/cbm-code-discovery-gate-$PPID)
was removed in c29e6d5 when the blocking bash gate became a stateless
compiled augmenter writing to stdout only. Add a regression guard so a
predictable temp/state file can never be reintroduced: assert the emitted
gate shim contains no /tmp path and no $PPID, and delegates to hook-augment.
Export cbm_install_hook_gate_script (was static) so the security property
is directly testable.
Cursor was never in the editor-detection table, so install/update only
ever wrote the VS Code config and silently skipped ~/.cursor/mcp.json,
even when Cursor was present. Uninstall removed the Cursor entry but a
subsequent install never re-created it.
Detect ~/.cursor/ as the Cursor agent, register the MCP server in
~/.cursor/mcp.json (mcpServers format, same as Windsurf/Gemini) on
install, remove it on uninstall, and list Cursor in detected agents.
Agent CLIs on Windows are installed as .cmd/.ps1/.exe shims (e.g. opencode
via mise/npm), but find_in_path only probed the bare name, so they weren't
detected. Try .exe/.cmd/.bat/.ps1 per PATH entry on Windows.
Relates to #221 (opencode not detected). The install-time taskkill 'eq'
error from the same report is addressed separately — cbm_kill_other_instances
now uses _spawnvp with an argv array rather than a shell string.
Martin Vogel