error-occurred.sh reads its two python helpers' stdin with `<<<`, a bash
here-string. dash has no such operator, so the script is a parse error, not a
runtime error: `sh error-occurred.sh` dies at line 32 with "Syntax error:
redirection unexpected" before any line runs.
tests/test_planning_disabled_optout.py runs every Copilot hook through
`["sh", str(script)]`, which ignores the `#!/bin/bash` shebang. On Ubuntu
/bin/sh is dash, so the hook exits 2 and the enabled-run assertion at line 200
fails. That is the whole of the current master CI failure; macOS and Windows
stay green because their /bin/sh parses `<<<`.
The four sibling hooks already pipe stdin instead, so this is the last `<<<`
under .github/. They pipe with `echo`, which is not safe to copy here: dash's
builtin echo expands backslash escapes, and $INPUT is raw JSON from Copilot,
so an escaped \n inside an error message becomes a real control character and
json.load rejects the payload. The hook then reports {} and the error is
dropped. `printf '%s\n'` never touches the escapes, which is the same reason
3c78c3c moved the output side of these hooks off echo.
Scoped to .github/hooks/scripts/. The four .gemini/hooks copies use `<<<`
too, but nothing invokes them through sh and settings.json runs them by path,
so their shebang applies and they are not part of this failure.
Seven shell hooks build their JSON by interpolating a python json.dumps
result and emitting it with echo. bash's echo leaves the escapes alone, so
the hooks are correct when run through their own #!/bin/bash shebang. Run
them with sh on macOS, where /bin/sh is bash in POSIX mode with xpg_echo
set, and echo turns the escaped \n inside the string back into a real
newline, producing JSON with a raw control character that every parser
rejects.
printf '%s\n' never interprets backslashes in its argument, so the emitted
bytes are the same under every shell.
Reported by @dylanpulver while rebasing #226: the macOS failure of
test_copilot_shell_hooks_skip_context_when_disabled is this, not his
change. Diagnosis and the one-line repro were his.
The #195 opt-out lives in scripts/inject-plan.sh, and the Cursor hooks read
task_plan.md directly instead of dispatching to it, so PLANNING_DISABLED=1
never reached any of the eight. Each guard reproduces its own hook's
no-plan-file output, so the Cursor protocol shape is unchanged: PreToolUse
still answers {"decision": "allow"} because that is what it emits
unconditionally today, and the other three stay silent.
Three defects found while auditing #222, #223 and #224:
- The Copilot PreToolUse guard answered permissionDecision "allow" on the
disabled path, so setting the variable to make planning inert also handed
every tool call a blanket approval. Both copies now emit {}, matching
their own no-plan-file path, which returns the decision to Copilot.
- .cursor/hooks/stop.ps1 was the last copy the #191 zero-phase guard never
reached. It answered "0/0 phases done" and auto-continued on a plan that
was never phase-structured.
- .github/hooks/scripts/error-occurred.ps1 read stdin into $input, which is
PowerShell's automatic pipeline variable. Under -File the assignment does
not stick, so the JSON parse always saw an empty string and the hook has
never logged an error on Windows. Reproduced under pwsh 7 and Windows
PowerShell 5.1 before the rename.
The opt-out tests now run every hook twice, unset and set, and fail if the
unset run is already inert. The merged versions asserted only the disabled
run, which a fleet gutted to unconditional {} also passed. That baseline is
what caught the $input bug.
Windows PowerShell 5.1, still the default shell on Windows and the interpreter
every dispatcher reaches for first, reads a .ps1 as ANSI unless the file
carries a UTF-8 BOM. The UTF-8 encoding of an em dash is E2 80 94, and CP1252
maps 0x94 to a closing curly quote, so the character opens a string literal
that never closes. An arrow does the same through 0x92. The result is a parse
error before the first statement runs, and every dispatcher wraps its
PowerShell call in fallbacks that swallow it, so the script silently does
nothing.
Eight shipped scripts failed to parse: the Cursor plan injection hook, both
.kiro asset scripts, the Arabic check-complete, and check-complete plus
init-session for both Chinese variants. The last of those means a Windows user
of either Chinese variant could not create a plan at all.
Every tracked .ps1 holding a byte above 0x7F now carries a BOM. Pure ASCII
files deliberately do not, because the dual-shipped script parity tests
compare them byte for byte between locations.
The guard asserts the invariant in plain Python so it runs everywhere, and
where a real powershell.exe exists it additionally asks the interpreter to
parse all 81 tracked scripts through ParseFile, which catches any encoding
trap rather than only this one. Both halves were verified to fail against the
pre-fix files.
Every Stop-hook style adapter (Codex, Cursor, GitHub Copilot's AgentStop, and
the canonical Claude Code check-complete.sh/.ps1 used by the gate and
advisory paths) checked only that task_plan.md exists, never that it contains
any "### Phase" headings. A task_plan.md that exists but isn't a structured
plan gives TOTAL=0, which fell through to the same branch as an in-progress
plan, producing "Task in progress (0/0 phases complete)" or, worse, Cursor and
GitHub Copilot's variants auto-continuing with "0/0 phases done ... continue
working on the remaining phases."
Reported against Codex specifically in #191 by @mixian939, with a correct
diagnosis and suggested fix. Auditing the same pattern found it in the
canonical scripts and two other IDE adapters too, so the guard lands
everywhere at once: a TOTAL=0 check that exits silently (or emits empty JSON
for Copilot's schema) before any status message is built.
Propagated to every check-complete.sh/.ps1 copy that shares this logic
(canonical, all sync-ide-folders.py-managed IDE mirrors via a sync run, the
five language-variant skills, and the standalone .kiro copy) so the fix is not
Claude-Code-only. inject-plan.sh, gate-stop.sh and the ledger scripts already
build on check-complete.sh's TOTAL and needed no separate change.
Resolving the Python interpreter solely from user PATH allows a malicious
PATH entry to redirect execution. Prefer well-known system interpreter
paths (/usr/bin/python3, /usr/local/bin/python3, /opt/homebrew/bin/python3)
before falling back to PATH-based resolution. Applies to session-start.sh,
pre-tool-use.sh, and error-occurred.sh.
Co-authored-by: Claude Code <noreply@anthropic.com>
Unbounded injection of session-catchup output into model context is a
medium-severity security concern: prior session content (which may include
web/search results or user messages) was injected verbatim with no size
limit. Add a 100-line cap via head -100 and prefix the context with a
label so the model knows this content is from a previous session.
Co-authored-by: Claude Code <noreply@anthropic.com>
v2.24.0 — Full IDE support audit and cleanup:
New features:
- Gemini CLI: add hooks (settings.json + 5 scripts: SessionStart, BeforeTool,
AfterTool, BeforeModel, SessionEnd). BeforeModel is unique to Gemini — injects
plan awareness before every model call.
- FactoryAI Droid: add hooks to SKILL.md frontmatter (was skills-only, now has
all 4 lifecycle events) + add scripts.
- GitHub Copilot: add errorOccurred hook for automatic error logging to plan.
Bug fixes:
- Continue: fix .prompt file — was entirely in Chinese, now English
- Pi Agent: fix package.json repo URL (was ttttmr/, now OthmanAdi/)
- Mastra Code: sync drifted check-complete.sh and session-catchup.py to canonical
- CodeBuddy/Codex: remove duplicate assets/templates/ directories
Consolidation:
- Remove .adal/, .agent/ (Antigravity), .kilocode/, .openclaw/ directories.
These IDEs implement the standard Agent Skills spec with no hooks — install
via `npx skills add` which handles IDE-specific placement automatically.
- README: restructure IDE table into Enhanced (hooks) vs Standard (Agent Skills)
- sync-ide-folders.py: remove entries for consolidated IDEs
Net: -4,609 lines (removed duplication, added real value)
Fixes#106. Three root causes identified and fixed:
1. Session catchup not working after /clear: Added UserPromptSubmit hook
to all 7 IDE SKILL.md files. When task_plan.md exists, injects a
directive to read all planning files before proceeding. Fires on every
user message, ensuring context recovery after /clear or compaction.
2. Progress not updating consistently: PostToolUse hook message now leads
with "Update progress.md" instead of only mentioning task_plan.md.
Added plan existence guard to avoid noise when no plan is active.
3. Post-plan additions not tracked: check-complete scripts now output
re-entry guidance when all phases are complete. Copilot agent-stop
scripts now return continuation context instead of empty JSON.
Added Critical Rule #7 (Continue After Completion) to SKILL.md body.
Applied across: 7 IDE SKILL.md files, 16 synced check-complete scripts,
4 Copilot hook scripts, CHANGELOG, plugin.json, marketplace.json.
v2.16.1 fixed the output pipe encoding but missed the read side.
Get-Content without -Encoding in PS5 reads files with the system ANSI
code page (Windows-1252), corrupting non-ASCII characters in
task_plan.md and SKILL.md before they reach the output pipe.
Changes to pre-tool-use.ps1, session-start.ps1, agent-stop.ps1:
- All Get-Content calls: added -Encoding UTF8
- [System.Text.Encoding]::UTF8 replaced with
[System.Text.UTF8Encoding]::new($false) — UTF-8 without BOM,
preventing stray 0xEF 0xBB 0xBF preamble in JSON output
Closes#82
PowerShell scripts (all 4):
- Add $OutputEncoding and [Console]::OutputEncoding = UTF8 before any output
- Fixes diamond replacement characters (◆) caused by PowerShell 5.x
defaulting to UTF-16LE stdout — Copilot reads stdout as UTF-8, causing
the byte mismatch that produces garbled output on Windows systems
Bash scripts (session-start.sh, pre-tool-use.sh):
- Use json.dumps(..., ensure_ascii=False) instead of default ensure_ascii=True
- Preserves UTF-8 characters (CJK, emojis, accented letters) in task_plan.md
context instead of converting them to raw \uXXXX escape sequences
README:
- Move Contributors from ugly bottom section into collapsible near top
- Add full contributors table with all PR authors since v2.8.0
- Add @Hexiaopi for Issue #82 bug report
- Update version badge and releases table to v2.16.1
CHANGELOG: add [2.16.1] entry
Fixes#82 — reported by @Hexiaopi