Cover 30-day month validation (Sep/Nov 31 rejection), getSessionStats
path heuristic with multiline content, combined date+search+pagination
in getAllSessions, ambiguous prefix matching in getSessionById, unclosed
code fence in parseSessionMetadata, empty checklist item behavior,
reserved name case sensitivity (LIST/Help/Set), negative limit in
listAliases, and undefined title in setAlias.
Replace console.log(data) with process.stdout.write(data) in both
pass-through paths to prevent appending a trailing newline that
corrupts the hook output. Add 7 tests covering exact byte fidelity,
malformed JSON, missing file_path, non-existent files, exclusion
patterns in check-console-log, non-git repo handling, and empty stdin.
Validate args parameter in getExecCommand() against SAFE_ARGS_REGEX to
prevent command injection when returned string is passed to a shell.
Escape regex metacharacters in getCommandPattern() generic action branch
to prevent malformed patterns and unintended matching. Clean up stdin
listeners in readStdinJson() timeout path to prevent process hanging.
Replace shell: true with npx.cmd on Windows in post-edit-format.js and
post-edit-typecheck.js to prevent command injection via crafted file paths.
Replace console.log(data) with process.stdout.write(data) in
check-console-log.js to avoid appending extra newlines to pass-through data.
- session-manager: clamp offset/limit to safe non-negative integers to
prevent negative offset counting from end and NaN returning empty results
- session-aliases: add success field to cleanupAliases return value for
API contract consistency with setAlias/deleteAlias/renameAlias
progressBar() in skill-create-output.js could crash with RangeError when
percent > 100 because repeat() received a negative count. Fixed by
clamping filled to [0, width].
New tests:
- progressBar edge cases: 0%, 100%, and >100% confidence
- Empty patterns/instincts arrays
- post-edit-format: null tool_input, missing file_path, prettier failure
- setup-package-manager: --detect output completeness, current marker
The command cross-reference regex /^.*`\/(...)`.*$/gm only captured the
LAST command ref per line due to greedy .* consuming earlier refs.
Replaced with line-by-line processing using non-anchored regex to
capture ALL command references.
New tests:
- 4 validate-commands multi-ref-per-line tests (regression)
- 8 evaluate-session threshold boundary tests (new file)
- 6 session-aliases edge case tests (cleanup, rename, path matching)
The OpenCode sub-package had stale 1.0.0 versions in package.json,
index.ts VERSION export, and package-lock.json while the main package
is at 1.4.1. Updated all three to match.
- Test valid project-config detection (.claude/package-manager.json)
- Test priority order: project-config > package.json > lock-file
- Test package.json > lock-file priority
- Test default fallback to npm
- Test setPreferredPackageManager success case
- Test getCommandPattern for test and build actions
The observer agent creates instinct files as .md with YAML frontmatter,
but load_all_instincts() only globbed *.yaml and *.yml. Add *.md to the
glob so instinct-cli status discovers all instinct files.
The replaceInFile function in utils.js accepts an optional `options`
parameter with `{ all?: boolean }` for replacing all occurrences, but
the .d.ts type declaration was missing this parameter entirely.
Fix grepFile() silently skipping matches when called with /g flag regex.
The global flag makes .test() stateful, causing alternating match/miss
on consecutive matching lines. Strip g flag since per-line testing
doesn't need global state.
Add first-ever tests for evaluate-session.js (5 tests: short session,
long session, missing transcript, malformed stdin, env var fallback)
and suggest-compact.js (5 tests: counter increment, threshold trigger,
periodic suggestions, below-threshold silence, invalid threshold).
- post-edit-typecheck: verify 25-level-deep directory completes without
hanging (tests the max depth=20 walk-up guard)
- cleanupAliases: document behavior when sessionExists callback throws
(propagates to caller, which is acceptable)
- Load both .yaml and .yml files in load_all_instincts() (#216)
The *.yaml-only glob missed .yml files, causing 'No instincts found'
- Implement evolve --generate to create skill/command/agent files (#217)
Previously printed a stub message. Now generates SKILL.md, command .md,
and agent .md files from the clustering analysis into ~/.claude/homunculus/evolved/
Tests the new transcript parsing from PR #215:
- entry.message.content format (string and array content)
- tool_use blocks nested in assistant message content arrays
- Verifies file paths and tool names extracted from both formats
* docs(zh-CN): sync Chinese docs with latest upstream changes
* docs: improve Chinese translation consistency in go-test.md
* docs(zh-CN): update image paths to use shared assets directory
- Update image references from ./assets/ to ../../assets/
- Remove zh-CN/assets directory to use shared assets
---------
Co-authored-by: neo <neo.dowithless@gmail.com>
* fix: Windows compatibility for hook scripts
- post-edit-format.js: add `shell: process.platform === 'win32'` to
execFileSync options so npx.cmd is resolved via cmd.exe on Windows
- post-edit-typecheck.js: same fix for tsc invocation via npx
- hooks.json: skip tmux-dependent hooks on Windows where tmux is
unavailable (dev-server blocker and long-running command reminder)
On Windows, execFileSync('npx', ...) without shell:true fails with
ENOENT because Node.js cannot directly execute .cmd files. These
hooks silently fail on all Windows installations.
The tmux hooks unconditionally block dev server commands (exit 2) or
warn about tmux on Windows where tmux is not available.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: parse Claude Code JSONL transcript format correctly
The session-end hook expected user messages at entry.content, but
Claude Code's actual JSONL format nests them at entry.message.content.
This caused all session files to be blank templates (0 user messages
despite 136+ actual entries).
- Check entry.message?.content in addition to entry.content
- Extract tool_use blocks from assistant message.content arrays
Verified with Claude Code v2.1.41 JSONL transcripts.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: ddungan <sckim@mococo.co.kr>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
The test was using 'claude-sonnet-4-5-20250929' which isn't in VALID_MODELS
(haiku/sonnet/opus). Use 'sonnet' with a description field containing
colons to properly test colon handling in frontmatter values.
Windows cmd.exe treats single quotes literally, so `echo '...' | node -e '...'`
fails. Switched to execFileSync with the `input` option to pipe stdin data
directly without shell quoting issues.
- hooks.json: add \b word boundary anchors to dev server blocking regex
to prevent false positives matching "npm run develop", "npm run devtools" etc.
- skill-create-output.js: guard box() horizontal repeat with Math.max(0, ...)
to prevent RangeError when title exceeds container width
- Add 13 tests for setup-package-manager.js CLI argument parsing
- Add 14 tests for skill-create-output.js SkillCreateOutput class
- All 333 tests passing
On Windows, os.homedir() uses USERPROFILE env var instead of HOME.
Tests that override HOME to a temp dir must also set USERPROFILE for
the session-manager, session-aliases, and session-start hook tests
to find files in the correct directory.
