Scoped adaptation of PR #2716: keep compact rendering as the default and require ECC_STATUSLINE_USAGE=1. Installer registration, managed aliases and the Codex background bar remain outside this slice.
- questionIdsFor and condensedHintFor take an optional
{ externalContract: true }; without it a code change that handles data
asks data-schema only, and an unknown profile gets the original four
questions and condensed hint.
- The hook reads GATEGUARD_EXTERNAL_CONTRACT_QUESTION with the same
spellings as GATEGUARD_METRICS and passes it to the full denial, the
condensed denial and the metrics record, so all three agree.
- The effectiveness harness's arm patch forwards the option, so an arm
never drops it silently.
Docs: change-profile.md gains an External-contract question section and
qualifies "byte-identical to the fixed four questions"; design-notes.md,
SKILL.md (switch table, data rule, word-for-word set) and CHANGELOG say
the fifth question is opt-in; question-effectiveness.md and the harness
README note that the runner clears GATEGUARD_* variables, so a gate arm
from a current tree measures the gate without the question.
skill-triggers@1.json: registryDigest refreshed for the SKILL.md body
change; the description and triggers are unchanged.
- Read file-type filters as include and exclusion globs: rg -t/--type
and the Grep tool's type include ripgrep 14.1's globs for the type,
rg -T/--type-not exclude them, fd -e/--extension includes *.<ext>.
An unknown type name, --type-add or --type-clear is an unreadable
filter and never credits, like --exclude-from.
- Treat a path pattern as a filter: the Glob tool's pattern must match
the target below its root; a shell glob operand must match the target
or a directory above it; a git pathspec matches the same way with its
* also matching /. One operand that admits the target is enough.
- Extend "a search of the target itself gives nothing" to searches
confined to the target's own file name: when every include, the Glob
pattern, or every path pattern ends in that exact name, the stem never
credits, and the closest-miss note says it was a lookup.
- Tokenize PowerShell searches with backslashes kept as path separators,
as the read-only shell check already does; the old literal-prefix scope
hid that src\core\*.ts reached the parser as srccore*.ts.
- Make an edit non-trivial when either text holds U+2028, U+2029 or
U+0085, which end a JavaScript or C# line comment; the lexer splits
lines only at \n and \r.
Extracting applyLongFilter, grepDetail and stemCredits leaves every
touched function at or below its complexity on main.
Isolate deadline fixtures from inherited dry-run/disabled hook flags. Remove an unapproved emoji from the new routing regression to satisfy Unicode validation.
Greptile P1: sedScriptParts stopped at -- and discarded the remaining
operands, so 'sed -- <script>' skipped script collection entirely. GNU sed
4.9 (verified): with no -e/--expression anywhere, the first operand after --
is the script and an s///e in it still executes.
The operands after -- are now collected as positionals, so the existing
first-positional-becomes-script rule applies. Operands after the script stay
file names (data). 2 regression tests pin both sides. Suite: 812/812.
Greptile P1 findings (both verified by running GNU sed against the hook):
1. Escaped replacements evade the check: GNU sed decodes \xHH, \oOOO, \dNNN,
case operations and unknown escapes like \g before executing an s///e
replacement or e operand. Scanning the raw text let \x67it decode to git.
Decoding that surface faithfully is out of scope, so any backslash in an
exec-bearing fragment now throws UnsupportedSedEscape and the hook blocks
with an explicit reason — mirroring the existing env -S escape precedent
instead of pretending to interpret sed. Escaped-newline continuations of an
e operand are refused by the same check.
2. Continued commands went unchecked: the e operand reader stopped at every
newline, but GNU sed joins escaped newlines into one operand. Now covered
by the same conservative refusal.
Also decomposes the sed readers into sub-50-line helpers per the repository
function-size rule (AGENTS.md): readSedDelimited, readSedCommand,
readSedExecute, readSedLineOperand, assertSedLiteral.
5 regression tests pin the escape paths and the benign boundary (a backslash
in a non-executing script stays data). Suite: 810/810.
CodeRabbit security review (Major, CWE-693) correctly identified that GNU sed
collects -e/--expression scripts from any argv position: a script option after
positional operands is still executed, so stopping the option walk at the first
positional let an e-flagged script hide behind it.
Verified against GNU sed 4.9 before fixing:
- 'sed -e s/x/y/ - -e s/y/X/e' runs X (the later -e executes)
- with any script option present, every positional is a file operand
sedScriptParts now walks the full argv, collects scripts from all -e/
--expression positions, and only promotes the first positional to the script
when no script option was given. Unknown options after positionals still fall
back to the opaque default.
4 regression tests pin the reported bypass paths (suite: 805/805).
Ported from the #3051 discussion (test intent, not the old scanner):
- A confidently-parsed sed invocation whose script contains no execution
construct is a pure text transformation: the script operand and every file
operand are data. This removes the specific false positives where file names
or non-executing scripts merely containing 'git ...' were blocked.
- The GNU sed `e` command (bare or with operand) and the `s` command with the
`e` flag run text through the shell. Their literal fragments (the s
replacement, the e operand) and any piped/redirected source are scanned as
real shell source, closing the `printf x | sed 's/x/git commit --no-verify/e'`
gap reported on 2026-10-09.
- Anything the literal parser cannot confidently read (addresses it cannot
consume, -f script files, -i suffix ambiguity, unknown options, expansion
residue) falls back to today's opaque default. No behavior changes for
unmodeled launchers.
- shell-scan.js is untouched and remains the canonical scanner.
Tests: 14 focused cases ported from the #3051 litmus set; full hook suite
801/801.
Validate env -S and --split-string escape syntax before shell scanning or
opaque fallback. Preserve literal single-quoted data and charge each
validation character against the shared input-proportional work budget.
Verified 787 focused cases and 290 real Git/config, harmless native env
and registered-routing controls. Keeps the prior paid-copy protection.
Co-authored-by: Nguyen Thanh Dat <ntdat812.dev@gmail.com>
Co-authored-by: Davemaina1 <dmain7015@gmail.com>
Carry forward bounded literal include/includeIf configuration capabilities
and env -S/--split-string words from PRs #2858 and #2897. Charge array
copy and expansion work before allocation to preserve the shared quota.
Verified 731 hook cases and 60 real Git/config and registered-route
controls. Dynamic remainder and original full source PRs remain held.
Co-authored-by: Nguyen Thanh Dat <ntdat812.dev@gmail.com>
Co-authored-by: Davemaina1 <dmain7015@gmail.com>
Resolve six conflicts:
- hooks/hooks.metadata.json: keep this branch's gateguard description and
fingerprint, add upstream's hookify-runtime entries.
- package.json: keep both gateguard-report.js and github-coordination.js.
- scripts/hooks/gateguard-fact-force.js, tests/hooks/gateguard-fact-force.test.js,
skills/gateguard/SKILL.md: upstream added a simpler session denial cap; this
branch already has the fuller one (cap threaded through
markCheckedAndCountDenial, malformed-value warning, miss and target notes),
so keep it. Upstream's taskset and su wrapper handling, fail-closed
recursion limit and soft-hyphen sanitising merged cleanly and their tests pass.
- manifests/context-packs/skill-triggers@1.json: take upstream's file and
recompute registryDigest for the merged registry.
Match the Windows missing volume-serial exception while retaining strict reported-device, inode, ownership, permission, link and snapshot checks. Add loader warning/block and race regression controls with isolated platform metadata injection.
Open with supported no-follow and nonblocking flags before relying on file metadata. Validate descriptor, pathname and rule-directory consistency before and after reading at most64KiB plus one overflow byte. Reject observed replacement, mutation, premature EOF and non-regular files, and close descriptors on every outcome.
Add12 race and I/O regressions. Keep the Hermes uppercase protected-file fixture real on case-sensitive filesystems and verify that the installed guard still permits first-time creation.
Validation: full6688 reported tests across347 files,12 race tests, full80% coverage gates, scoped Hookify coverage, installed Hookify and Hermes workflows, packed artifact lifecycle, lint, signatures and IOC.
Carry forward PR #3055 at 92f53c4c onto ef648e01, preserving contributor attribution and current main. Focused maintainer fixes reuse shared input ceilings, preserve late root context through outer bootstrap, bound frontmatter before immutable parsing, and assert real oversized input in the limits workflow.
plugin-hook-bootstrap (shell mode) and observe-runner probed a bare
`bash.exe`. On Windows that resolves to the first match on PATH, and
%SystemRoot%\System32 normally sits ahead of Git's bin directory, so the
pick was System32\bash.exe: the WSL launcher. The existing guard (require
a clean probe exit) only catches it when no distro is installed. With one
installed, the probe passes inside Linux, then the hook script is handed
over by a Windows path (C:\...) or a Git Bash path (/c/...), neither of
which exists in WSL. The hook exits 127 instead of running, and
observe-runner's continuous-learning observation fails on every call.
Resolve bash candidates on Windows to absolute paths from PATH, skipping
bash.exe in System32, SysWOW64, Sysnative, and the WindowsApps alias
directory, so a native bash later on PATH (Git's) is used. With none
available, the existing "no bash binary found" skip applies. An explicit
BASH override still wins, and other platforms keep bare names.
The shared resolver lives in scripts/lib/windows-bash.js with a
platform-independent test. A Windows integration test puts System32 ahead
of Git's bash and requires the .sh hook to actually run. Upstream's
bootstrap fails that scenario with exit 127 ("/bin/bash: C:Users...hook.sh:
No such file or directory"); this one prints the hook output.
Under OpenCode the skill-health dashboard had no writer and no honest
fallback, so it reported every discovered skill as healthy (#2463).
Root cause: `post:skill:track` is only wired for Claude Code, where
`posttooluse-dispatcher.js` fans the event out to `skill-run-tracker.js`.
The OpenCode plugin re-implements the harness events instead of reusing that
dispatcher, so nothing ever wrote `skill-runs.jsonl` — and the tracker's
matcher only accepted the exact Claude Code tool name `Skill`, while OpenCode
calls the tool `skill`, so even wiring it would have dropped every event.
Commands that locate ECC through the inlined resolver in
`scripts/lib/resolve-ecc-root.js` also fell back to `~/.claude`, which does
not contain ECC, so they could not find their scripts.
Changes:
- `.opencode/plugins/ecc-hooks.ts`: dispatch `post:skill:track` from
`tool.execute.after`. The tracker module is imported lazily and behind
try/catch, matching the existing `changed-files-store` pattern: this file is
OpenCode's startup entry point, so a static import failure would crash the
session before any hook runs (#2530). The module is reused rather than
reimplemented, keeping its identifier validation, length bounds and
"never persist prompt text" guarantees.
- `scripts/hooks/skill-run-tracker.js`: match the tool name case-insensitively
so `Skill` and `skill` are both accepted.
- `scripts/lib/skill-evolution/health.js`: split the summary into
measured/healthy/declining/unmeasured instead of deriving `healthy = total -
declining`, which turned "no telemetry" into "everything is fine" for every
skill.
- `scripts/lib/skill-evolution/dashboard.js`: show the breakdown and warn
explicitly when skills have no recorded runs, since absence of telemetry is
not a health signal.
Under OpenCode the resolver also needs `CLAUDE_PLUGIN_ROOT`, which only the
plugin can supply: `shell.env` exports it when a root is found, without
replacing a root the user set or one an earlier plugin already put in
`output.env`.
Root discovery is per capability and probes required runtime files — never
optional content. Probing for `skills/continuous-learning-v2` (the
`workflow-quality` module) rejected supported modular installs that ship the
plugin and tracker without that optional module, disabling telemetry and the
root export for exactly those users.
Also fixes an incidental bug: the compiled plugin bundle is ESM, where
`__dirname` is undefined, so the ECC version lookup silently fell back to its
default. The module directory is now derived from `import.meta.url`.
Tests: `tests/opencode-plugin-hooks.test.js` covers the telemetry dispatch
(skill tool recorded, other tools ignored, minimal profile respected), a
modular install without the optional skills module, the root export and both
override paths; `tests/hooks/skill-run-tracker.test.js` covers the
case-insensitive matcher; `tests/lib/skill-evolution.test.js` and
`tests/lib/skill-dashboard.test.js` cover the unmeasured bucket, the warning
and the aged-out-coverage wording.
Refs #2463, #2530
haelyra