* fix(observer): auto-scale max_turns by analysis batch size (#2035)
The hardcoded default of MAX_TURNS=20 is insufficient when
MAX_ANALYSIS_LINES=500 (also the default). Claude exhausts its turn
budget before it can write all discovered instinct files, producing:
Error: Reached max turns (20)
Fix: when ECC_OBSERVER_MAX_TURNS is not explicitly set, compute
max_turns proportionally to the actual analysis batch size:
max_turns = clamp(analysis_count / 10, 20, 100)
This gives:
- 20–199 lines → 20 turns (existing floor, unchanged)
- 500 lines → 50 turns (resolves the reported failure)
- 1000 lines → 100 turns (cap)
Explicitly setting ECC_OBSERVER_MAX_TURNS still overrides the
auto-scaled value, preserving the existing escape hatch.
* test(observer): update max_turns test for auto-scaling; document validation
The max-turns budget test in tests/hooks/hooks.test.js still asserted the removed literal max_turns="${ECC_OBSERVER_MAX_TURNS:-20}", which would fail against the new auto-scaling logic. Assert the auto-scale formula and the 20/100 clamp bounds instead.
Also add the explanatory comment CodeRabbit requested above the max_turns sanitization block, clarifying it guards the explicit ECC_OBSERVER_MAX_TURNS override path.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
Cursor hooks still called `npx block-no-verify@1.1.2`, the broken external
package whose matcher over-matches: it blocks legitimate `git commit`
whenever `--no-verify` (or `no-verify`) appears anywhere in the command
string, including inside the commit message body. The Claude Code surface
already routes through the in-repo `scripts/hooks/block-no-verify.js`,
which performs flag-position-aware tokenisation and passes 25 regression
tests covering every false-positive case from #2107.
Add a thin Cursor wrapper (`before-shell-execution-block-no-verify.js`)
that reads Cursor stdin, transforms to the Claude Code `tool_input.command`
shape, delegates to the local hook's exported `run()`, and forwards exit
code and stderr. Update `.cursor/hooks.json` to call the wrapper instead
of the npx package. New 14-case test file pins the false-positive cases
from the issue plus the still-blocked real bypass attempts.
Fixes#2107
* fix(session-start): support ECC_SESSION_RETENTION_DAYS opt-out + document env var
The retention pass for *-session.tmp files (issue #2151) landed previously,
but the env var that controls it was undocumented in the README and rejected
falsy values (0, off, disabled), silently falling back to the 30-day default.
Users who want to keep all sessions for forensic or research workflows had no
way to opt out.
This patch:
- Extends getSessionRetentionDays() so 0|off|false|disabled|never|none disables
pruning entirely (returns null sentinel; default behavior unchanged).
- Updates the call site in main() to skip pruneExpiredSessions when retention
is null and emits a clear "[SessionStart] Pruning disabled via
ECC_SESSION_RETENTION_DAYS" log line so the operator can tell pruning is off.
- Documents ECC_SESSION_RETENTION_DAYS in the README "Hook Runtime Controls"
section alongside the other ECC_SESSION_* knobs.
- Adds three regression tests in tests/hooks/hooks.test.js covering opt-out
via 0, opt-out via off, and garbage-value fallback to default 30.
Verification:
- node tests/hooks/hooks.test.js — 240/240 green (incl. 3 new retention tests)
- node tests/run-all.js — 2622/2622 green
- npx eslint scripts/hooks/session-start.js tests/hooks/hooks.test.js — clean
- node scripts/ci/validate-no-personal-paths.js — clean
- node scripts/ci/check-unicode-safety.js — clean
- node scripts/ci/validate-hooks.js — 28 matchers validated
- node scripts/ci/validate-rules.js — 115 files validated
Fixes#2151
* docs(readme): list all ECC_SESSION_RETENTION_DAYS opt-out values + add Windows example
Address reviewer feedback on PR #2163:
- CodeRabbit and cubic both flagged that the README docs only listed 3 of 6
opt-out values accepted by getSessionRetentionDays() (0, off, disabled),
while the implementation also accepts false, never, none.
- cubic also flagged the missing Windows PowerShell example for the new
variable, breaking the parallel structure of the existing
ECC_CONTEXT_MONITOR_COST_WARNINGS example block.
Updated the README to:
- Spell out all six opt-out values (0, off, false, disabled, never, none)
and clarify they "keep all sessions (disable pruning)".
- Add an ECC_SESSION_RETENTION_DAYS line to the Windows PowerShell example.
No behavior change. README only.
Verification:
- npx markdownlint README.md — clean
- npx eslint scripts/hooks/session-start.js tests/hooks/hooks.test.js — clean
* feat(gateguard): add env knobs for routine bash gate + extra destructive patterns
The JS port of gateguard-fact-force has two bash gates: a destructive
gate (rm -rf, drop table, git push --force, etc.) that operators want
to keep, and a once-per-session routine gate that fires on the very
first bash invocation regardless of intent. Operators on hosts where
the routine gate is friction without signal (Cursor, OpenCode, etc.)
have been maintaining local patches that get clobbered on every plugin
update; the Python upstream gateguard-ai already exposes equivalent
config via .gateguard.yml.
Adds two env vars, both off-by-default so existing behavior is
preserved:
- GATEGUARD_BASH_ROUTINE_DISABLED — truthy values (1, true, on, yes,
enabled) skip the routine bash gate. Destructive gate is unaffected.
- GATEGUARD_BASH_EXTRA_DESTRUCTIVE — regex source string for additional
destructive patterns. Matches against the same quote-stripped,
subshell-flattened command the built-in DESTRUCTIVE_SQL_DD regex sees,
so a custom phrase inside $(...) or backticks is also caught. A
malformed regex is logged once to stderr and treated as not configured
rather than crashing the hook (hooks must never block tool execution
unexpectedly).
Twelve new tests pin both env vars (truthy aliases, falsy values, unset
baseline, destructive-gate-still-fires, alternation members, malformed
regex degrades safely, custom phrase inside command substitution).
Existing 2619/2619 tests still pass; eslint clean.
Fixes#2078
* fix(gateguard): reset extra-destructive warn-once gate when env value changes
Both reviewers (CodeRabbit + cubic) flagged that
extraDestructiveWarnLogged was never reset when GATEGUARD_BASH_EXTRA_DESTRUCTIVE
flipped from one invalid regex to a different invalid regex. The
sticky boolean meant a long-running process saw bad-pattern-a's
warning then silently swallowed bad-pattern-b's parse failure.
Fix: clear extraDestructiveWarnLogged whenever the cache key changes
(i.e. before the regex compile attempt). The warn-once-per-distinct-
pattern invariant now matches the per-key cache invariant.
Adds a same-process regression test via loadDirectHook() that spies on
process.stderr.write and asserts: same bad pattern warns once across
multiple invocations; switching to a different bad pattern emits a
second warning; switching to a valid regex emits zero warnings.
* fix(suggest-compact): clean up old counter temp files
claude-tool-count-<sessionId> files were written into the OS temp dir
on every hook run and never removed, accumulating one orphan per
session indefinitely.
Sweep stale counter files at the top of main() before opening the
active counter. Retention is env-tunable via COMPACT_STATE_TTL_DAYS
(default 14 days); invalid values fall back to the default. The
active session's counter file is preserved unconditionally even if
its mtime is past the cutoff. Failures during the sweep are swallowed
to preserve the always-exit-0 hook contract.
Adds 7 regression tests covering the sweep, env-var validation, and
the always-exit-0 invariant under a populated temp dir.
Fixes#2156
* fix(suggest-compact): preserve counter files at the TTL cutoff boundary
The cleanup sweep used `mtimeMs > cutoffMs` to short-circuit, which
matched files whose mtime sits exactly on the cutoff boundary and
deleted them. The cleanupOldCounters docstring promises only files
*older than* retentionDays are removed; a file at age == retentionDays
is not older than retentionDays, so it must survive.
Switch the comparison to `>=` so only strictly older files fall
through to deletion. Add a regression test that pins boundary-aged
files (mtimeMs sitting just past the projected cutoff) are preserved.
Refs #2156
The observe.sh Layer 1 entrypoint guard short-circuits with exit 0 when
CLAUDE_CODE_ENTRYPOINT is not in {cli, sdk-ts, claude-desktop}. Claude
Code's VS Code extension sets CLAUDE_CODE_ENTRYPOINT=claude-vscode, so
VS Code users see no observations recorded — observations.jsonl never
gets created and the instinct pipeline stays empty.
Add claude-vscode to the allowlist, mirroring the precedent in #1522
which added claude-desktop the same way.
Add a regression test that spawns observe.sh under bash -x for each
allowed entrypoint (cli, sdk-ts, claude-desktop, claude-vscode) and
each denied entrypoint (unknown-host, claude-cody, mcp), asserting
that allowed entrypoints reach Layer 2's ECC_HOOK_PROFILE check while
denied entrypoints stop at Layer 1.
Fixes#2102
* feat: add worktree-lifecycle service (ecc.worktree-lifecycle.v1)
The "unowned moat" from the orchestrator landscape research: no existing
tool ships deterministic merge-conflict prediction or a safe worktree GC.
- scripts/lib/worktree-lifecycle/git.js: injectable, hermetic git layer.
Predicts merge conflicts WITHOUT touching the working tree via
`git merge-tree`. Strips inherited GIT_* env so it is safe inside hooks.
- scripts/lib/worktree-lifecycle/lifecycle.js: deterministic state machine
(main/dirty/conflict/merge-ready/merged/stale/idle) + planCleanup that
buckets worktrees into remove / salvage / keep. Only fully-merged trees
are auto-removable; stale (unmerged+inactive) => salvage, never deleted.
- scripts/worktree-lifecycle.js: CLI (--json/--conflicts/--stale/
--cleanup-plan/--base/--stale-days/--repo).
- tests/lib/worktree-lifecycle.test.js: 11 tests (fake-git + real-git).
Safety model mirrors the reference-arch salvage rule, validated by the
2026-06-05 MacBook->Mac Mini consolidation. Tests: 11/0.
* fix: hermetic git env in session adapters + mcp-inventory lint
- session adapters (codex-worktree, opencode): resolveGitBranch stripped
no git env, so the "outside a repo" path returned the host branch when
run inside a git hook (GIT_DIR set). Strip GIT_* before rev-parse.
- mcp-inventory: fix eslint no-unused-vars (signatures) and a stale
eslint-disable directive in the merged code.
* test: run each test with inherited git env stripped (hermetic runner)
When the suite runs inside a git hook (pre-push), git sets GIT_DIR/
GIT_WORK_TREE, which hijack 'git -C <dir>' calls in tests that exercise
real git, making them operate on the host repo. Strip GIT_* before
spawning each test so the suite is isolated from ambient git state.
---------
Co-authored-by: ECC Test <ecc@example.test>
* feat: add MCP inventory (ecc.mcp.v1) across harnesses
Read-only MCP-gateway groundwork: discover MCP server configs across
every installed harness, normalize to a canonical ecc.mcp.v1 inventory,
redact secrets, and report which servers are configured in 2+ harnesses
(the configure-N-times pain). The read+dedup side of a unified gateway,
mirroring how the session-adapter layer started read-only.
Readers (per-harness config formats):
- claude-code: ~/.claude.json mcpServers + project .mcp.json
- codex: ~/.codex/config.toml [mcp_servers.*] TOML via @iarna/toml
- opencode: ~/.config/opencode/opencode.json mcp block (command ARRAY)
canonical-mcp.js:
- normalize transport labels (local=>stdio, remote=>http) to stdio/http/sse
- merge servers by name across harnesses; flag DRIFT when signatures differ
- fragmentation report + aggregates
- SECRET REDACTION: env values stripped to key names; secrets in args
(--modelApiKey sk-ant-...), inline --flag=secret, and URL userinfo/token
query params all redacted before storage AND before the dedup signature.
scripts/mcp-inventory.js: CLI (--json, --fragmented, --help).
tests/lib/mcp-inventory.test.js: 12 tests incl. a regression for the
real arg-carried-secret leak found while smoke-testing on live configs.
Tests: 12/0. Real-data smoke: 33 servers across 3 harnesses, 21
configured in 2+ harnesses (7 drift); secret-leak audit clean.
* test: cover reader error paths, collect skip-logic, and CLI main() for mcp-inventory
Lift global branch coverage past the 80% gate (was 79.86%). Adds 6
tests exercising: missing-file/malformed-JSON/missing-block reader
fallbacks, codex no-parser path, collect skipping non-function readers
and swallowing reader errors, CLI usage()/main() help+json+human paths,
and formatHumanReport no-fragmentation + fragmented-only branches.
Also scrub a real API-key fragment that had leaked into a test fixture;
all secret-like fixtures are now obviously-fake FAKE... tokens.
mcp-inventory.js branch 30%->93%, collect.js ->100%. Global branch 80.33%.
* feat: add codex-worktree session adapter
Adds the third session adapter (after dmux-tmux and claude-history),
normalizing Codex rollout sessions into the harness-neutral
ecc.session.v1 snapshot. Reads ~/.codex/sessions rollout JSONL,
derives objective (skipping the AGENTS.md preamble + leading message
UUID), model, originator, worktree cwd, and best-effort git branch.
This is step 1 of ECC-2.0-SESSION-ADAPTER-DISCOVERY (move the
abstraction beyond tmux + Claude-history) and supports the
wrap/adapt control-pane strategy: ECC reads sessions from any
harness rather than owning one UX.
- scripts/lib/session-adapters/codex-worktree.js: adapter + rollout parser
- canonical-session.js: normalizeCodexWorktreeSession
- registry.js: register adapter, codex/codex-worktree target types
- tests/lib/session-adapters-codex.test.js: 4 tests (unit + registry routing)
* feat: add opencode session adapter + allow empty intent objective
Adds the fourth session adapter (after dmux-tmux, claude-history,
codex-worktree), normalizing OpenCode sessions into ecc.session.v1.
Reads ~/.local/share/opencode/storage: session/<project>/ses_*.json
for metadata (id, directory, title, version, projectID, time) and
message/<session>/msg_*.json to extract the model (modelID/providerID
from the first assistant message). Derives objective from the session
title, treating the auto-generated "New session - <date>" title as no
objective. Recency-based active/recorded state.
Schema: relax intent.objective from non-empty to allow empty string
(ensureStringAllowEmpty). Sessions legitimately have no objective yet
(fresh/auto-titled), and claude-history already emitted "" via
metadata.title fallback. This fixes a latent over-strict validation.
- scripts/lib/session-adapters/opencode.js: adapter + storage parser
- canonical-session.js: normalizeOpencodeSession + ensureStringAllowEmpty
- registry.js: register adapter + opencode target type
- tests/lib/session-adapters-opencode.test.js: 5 tests
Tests: opencode 5/0, codex 4/0, session-adapters 14/0,
control-pane-state 10/0, session-inspect 8/0, control-pane 12/0.
Smoke-tested on a real OpenCode session (140 messages, gpt-5.3-codex).
* test: cover error/fallback branches for codex-worktree + opencode adapters
Lift global branch coverage past the 80% gate (was 79.53%). Adds error
and fallback path tests: missing-session/unknown-id throws, findRolloutById/
findSessionInfoById, direct file targets, objective truncation, model
fallbacks, corrupt-line skip, mtime activity fallback, and the real
resolveGitBranch path outside a repo.
codex-worktree.js branch 52.8%->78.3%; global branch 80.04%.
Adds dynamic workflow/team orchestration skills, the content pack, and control-pane work-item/Kanban state DB support. Includes reviewer hardening for state-db CLI validation, optional state DB failure handling, and mergeStateStatus projection.
Adds de-DE docs, installer wiring, and locale tests. Pre-validated on current main with install manifest checks, markdownlint, locale-install tests, and ECC 2.0 release-surface tests.
Fail fast when the OpenCode home install is attempted from a source checkout without the compiled .opencode/dist payload. PR had the full CI matrix green.
Addresses CodeRabbit review: the negative-only assertions could have
passed on an empty plan. Add a positive assertion that the non-foreign
'rules' path is still planned under .claude/rules/ecc so regression to
zero ops would fail loudly.
Completes the install-target matrix for Claude Code. Until now, ECC's
Claude support was home-scope only (~/.claude/) via the `claude` target.
This adds a project-scope counterpart (./.claude/) via a new
`claude-project` target so teams can install ECC per-repo without
contaminating ~/.claude/ — matching the existing project-scope adapters
for Cursor, Antigravity, Gemini, CodeBuddy, Joycode, and Zed.
Symmetric with `claude`:
- Same namespace under rules/ecc and skills/ecc
- Same docs/<locale> handling for --locale
- Same hooks placeholder substitution for hooks.json
- Reuses claude-home's destination-mapping logic 1:1
Use cases:
- Monorepos with multiple Flow-managed projects
- Teams that want ECC scoped per-project without touching ~/.claude/
- Per-project skill/rule isolation when global install isn't desirable
No breaking change: existing --target claude continues to route to
claude-home (user-scope) unchanged. New target is opt-in.
Tests
-----
- 4 new tests in tests/lib/install-targets.test.js
(root resolution, lookup-by-id, plan parity with claude, foreign-path filtering)
- All install-target regression guards (schema enum / SUPPORTED_INSTALL_TARGETS)
still pass
- End-to-end smoke: `--target claude-project --profile minimal --dry-run`
emits 359 ops with destinations rooted at <projectRoot>/.claude/ (parity
with --target claude which emits 359 ops rooted at ~/.claude/)
Salvages the useful harness-audit scoring work from #1989 while preserving the current hook registry and newer plugin install detection. Adds GitHub integration checks, conditional deploy-provider categories, dynamic applicable category metadata, and CODEOWNERS coverage.
Refresh the active 2.0 release surface for the affaan-m/ECC repo identity, update package/plugin/workflow launch metadata, and add an operator command center for release video, partner, sponsor, consulting, and social launch execution.
Generate the inline hook root resolver with single-quoted JavaScript literals so Windows Git Bash does not choke on nested escaped double quotes before Node starts. Refresh hooks.json and add regression coverage for parsed hook commands and installed hook manifests.
Two round-1 review findings in `tests/lib/session-bridge.test.js`,
both about test correctness rather than the underlying fix:
1. **greptile P1 + coderabbitai Major + cubic P2 (all three): concurrent-write test ran sequentially.**
The test spawned two child processes with two consecutive
`spawnSync` calls. Because `spawnSync` blocks until the child
exits, the second writer started *after* the first finished —
the two writers never overlapped, so the rename race the fix
targets was never actually exercised. The test would have passed
with the old broken `${target}.tmp` suffix.
Fix: introduce a one-off "race runner" helper that runs inside
its own subprocess and uses async `spawn` to start both writers
simultaneously. The runner waits for both to exit (the event
loop is local to the runner subprocess, so this stays compatible
with the synchronous test harness used elsewhere in this file)
and reports both exit codes plus stderrs on stdout. The test
then calls the runner via `spawnSync` and parses the result.
Both writer children now overlap for the duration of their 200
`writeBridgeAtomic` calls each, which is enough wall time to
reliably trigger the rename race against the pre-fix code.
Verified: with the fixed `${target}.${pid}.${nonce}.tmp` suffix,
the test passes; with the old fixed `${target}.tmp` suffix
reintroduced, it fails as expected (one writer hits ENOENT on
roughly half its rename calls).
2. **greptile P2 + cubic P3: `assert.throws` used a string as the second argument.**
Node deprecated passing a string as the second argument to
`assert.throws` years ago: the string is silently treated as
the assertion failure message (what to print when the function
does *not* throw) rather than as an error matcher. The check
passed for any thrown error, not just the rename failure.
Fix: pass a regex matcher as the second arg and keep the
explanatory text as the third. The regex matches `EISDIR`,
`EPERM`, `ENOTDIR`, or `ENOENT` because `renameSync` of a
regular tmp file onto an existing directory raises different
codes on Linux / macOS / BSD — making the matcher portable
across CI runners.
Test count unchanged at 14; `npm test` green; `npm run lint` clean.
The two helper files (`tests/__tmp_bridge_writer.js`,
`tests/__tmp_bridge_race_runner.js`) are written and unlinked
inside the test's try/finally so they never persist beyond the
test run.
Two regression tests pin down the previous two commits' atomic-rename
fixes:
1. **concurrent writes don't throw ENOENT or corrupt the file** —
spawns two child Node processes (`tests/__tmp_bridge_writer.js`
created in-test, cleaned up in finally) that each call
`writeBridgeAtomic(sid, …)` 200 times against the same session
ID with independent payloads. Asserts both subprocesses exit 0
(the previous implementation produced ENOENT on roughly 50% of
rename calls, all swallowed by the in-test catch) and the final
bridge file is parseable JSON belonging to one of the two writers
(last-writer-wins is fine; the contract is *no corruption* and
*no rename ENOENT*, not data preservation).
2. **tmp file cleanup on rename failure** — pre-creates a directory
at the target bridge path so `renameSync(tmp, target)` fails,
calls `writeBridgeAtomic`, asserts the call throws AND that no
tmp file with the writer's `pid.<nonce>.tmp` prefix is left
behind in `os.tmpdir()`. The previous code had no cleanup; the
fix's `try/catch + unlinkSync` keeps tmpdir from accumulating
orphan files across repeated rename failures.
The first test deliberately writes independent payloads from each
subprocess so this regression doesn't try to claim a property the
fix doesn't actually deliver (read-modify-write race in the caller
is a separate issue and out of scope per PR body).
Test count: 12 → 14 in `tests/lib/session-bridge.test.js`;
`npm test` green; `npm run lint` clean.
9 new test cases pin down the two previous commits' denylist
extensions. Each verifies both detection (validator exit non-zero +
the expected `dangerous-invisible U+<HEX>` line on stderr) and,
where applicable, `--write` sanitization.
Coverage:
Tag block (commit 1):
- U+E0041 TAG LATIN CAPITAL LETTER A — the range's printable ASCII
shadow; this is the byte sequence demonstrated in published ASCII
smuggling proofs of concept.
- U+E007F CANCEL TAG — the range end.
Other invisibles (commit 2):
- U+180E MONGOLIAN VOWEL SEPARATOR
- U+115F HANGUL CHOSEONG FILLER
- U+1160 HANGUL JUNGSEONG FILLER
- U+2061 FUNCTION APPLICATION (range start)
- U+2064 INVISIBLE PLUS (range end)
- U+3164 HANGUL FILLER
Detection table is data-driven (one loop, one assertion per row) so
adding the next invisible to the denylist also gets a paired
regression test by simply appending to NEWLY_COVERED_RANGES.
Plus a `--write` integration test:
- writes a markdown file containing both Tag block (5 chars) and
U+180E, runs `--write`, asserts both removed and surrounding text
preserved character-for-character ('# Title\n\nBenigntext.\n').
- re-runs the validator without `--write` and asserts exit 0,
confirming the sanitizer's output is idempotent under the
extended denylist.
Test count: 5 → 14 in this file; full `yarn test` green; `yarn lint`
clean.
