AnotherSkillBundle

Reusable skill belt for Claude Code projects. Pull this repo into any project that wants the same automation patterns — autonomous research, long-term task orchestration, anti-pattern advice — without re-deriving them.

What's in here

Path	Purpose
Skills/	Skill .md files + their owned scripts
Skills/SharedScripts/	Cross-skill helpers
Skills/Index.md	Skill map + script ownership + flow diagrams
Wiki/ClaudeExperience/	Anti-pattern + good-practice knowledge base referenced by ClaudeAdviceSkill
skillSettings.example.json	Template for project-level config (paths to memory, wiki, etc.)

Skills included

• AutoresearchSkill — orchestrates autonomous research runs (BOOTSTRAP only). Dispatcher chains FETCH → INGEST → REVIEW.
• SourceScrapeSkill — finds + fetches web sources (DISCOVER / FETCH sub-protocols).
• IngestionSkill — reads raw sources, extracts facts with verification, integrates into Wiki.
• IngestionReviewSkill — one-shot cross-wiki consistency pass after ingestion.
• LongTermTaskSkill — goals too big for one session. Decomposes into partial tasks + steps, each runs in its own claude -p session.
• YouTubeTranscriptSkill — fetch YouTube transcripts in autoresearch-compatible raw format.
• ClaudeAdviceSkill — always-active anti-pattern nudger (uses Wiki/ClaudeExperience/).

Using these skills in a project

1. Clone AnotherSkillBundle into your project. Either:

   • Copy Skills/ and Wiki/ClaudeExperience/ into your project root, or
   • Add this repo as a git submodule and symlink / merge the folders.

2. Copy skillSettings.example.json → skillSettings.json at your project root. Edit the paths so they match where your project keeps its raw sources, wiki, and long-term task state.

3. Tell Claude where to read settings. In your project's CLAUDE.md, add a line like:

   ## Skill settings
   Read `skillSettings.json` at session start. When skill files reference
   `{{RAW_ROOT}}`, `{{WIKI_ROOT}}`, `{{LONGTERM_ROOT}}`, `{{CLAUDE_EXPERIENCE_ROOT}}`,
   substitute the corresponding value from `skillSettings.json`.
   

4. Permissions. Some skills write outside the standard write zone. Add to .claude/settings.json → permissions.allow:

   "Write(path:MemoryVault/Raw/**)"
   "Edit(path:MemoryVault/Raw/**)"
   "Write(path:MemoryVault/Wiki/**)"
   "Edit(path:MemoryVault/Wiki/**)"
   "Write(path:MemoryVault/LongTermTask/**)"
   "Edit(path:MemoryVault/LongTermTask/**)"
   

   (Adjust paths to match your skillSettings.json.)

Placeholder tokens

Skill .md files use these tokens instead of hardcoded paths. Define each one in skillSettings.json:

Token	skillSettings key
{{RAW_ROOT}}	rawRoot
{{WIKI_ROOT}}	wikiRoot
{{LONGTERM_ROOT}}	longTermRoot
{{CLAUDE_EXPERIENCE_ROOT}}	claudeExperienceRoot
{{SKILLS_ROOT}}	skillsRoot

The PowerShell + Python scripts are path-agnostic — they take the relevant directory as a parameter (-TaskDir, <task_dir> argv). The tokens exist for the skill prose so Claude knows where to look when the user invokes a skill conversationally.

Requirements

• PowerShell 7+ (pwsh) — runners are .ps1, cross-platform on Windows / macOS / Linux. Windows PowerShell 5.1 also works on Windows.
• Python 3.9+ (state CLIs, transcript fetcher)
• pip install youtube-transcript-api (only if YouTubeTranscriptSkill / [youtube] candidates are used)
• Claude Code CLI on PATH as claude (runners spawn claude -p WORKER sessions)

Windows-only feature

Auto-relaunch on usage-limit (ON_LIMIT: relaunch in task.md) uses Windows Task Scheduler via Register-ScheduledTask — Windows-only. On macOS / Linux, set ON_LIMIT: stop and trigger relaunch via cron / launchd / systemd yourself, or just relaunch manually. Everything else (FETCH / INGEST / REVIEW loops, LTT dispatcher, state CLIs) is platform-agnostic.

Conventions

• Skill .md at Skills/<Name>Skill.md (PascalCase + Skill suffix).
• Owned scripts at Skills/<Name>Skill/ — folder matches skill filename.
• Shared scripts at Skills/SharedScripts/.
• Run-*.ps1 = orchestrators / dispatchers. *_state.py = state CLI. _runner-helpers.ps1 = shared PS helpers (underscore prefix = not a runnable script).

See Skills/Index.md for the full map.