cli-jaw-skills
Reference skill library for cli-jaw agents.

---

cli-jaw-skills

cli-jaw-skills is a public reference library of agent skills for cli-jaw. Each skill is a directory with a required SKILL.md and optional reference materials, scripts, templates, tests, or domain assets.

The repo is designed as source material for the active cli-jaw skill system: installers and sync jobs can copy selected skills into runtime skill folders, while maintainers can review the full library in one place.

Public Surface

Surface	Status
Skill library	229 top-level SKILL.md files
Reference material	47 skills include reference/ or references/ folders
Helper scripts	28 skills include scripts/ folders
Templates	2 skills include templates/ folders
Office formats	docx, pptx, xlsx, and hwp skill families are present
CI	Focused Python regression tests, skill count validation, known long-skill drift checks, docs drift checks
GitHub Pages	/docs/index.html static landing page, ready for Pages deployment
License	No root license file is currently declared

Remote signal:

• Repository: lidge-jun/cli-jaw-skills, public, 4 stars, 0 forks.
• Existing workflow: Sync OfficeCLI Skills has one recent success and one recent failure from manual dispatch history.
• GitHub Pages is not enabled yet (GET /repos/lidge-jun/cli-jaw-skills/pages returns 404). The added Pages workflow deploys /docs after an authorized push.

Library Shape

The library spans agent operations, browser control, document generation, developer guides, frontend/backend/data/testing/security workflows, search, media, office files, cloud platforms, productivity systems, and language-specific engineering patterns.

Representative families:

Family	Examples
Agent operations	browser, desktop-control, web-ai, memory, telegram-send
	Development guides
Office/document work	docx, pptx, xlsx, hwp, pdf, pdf-vision
Cloud and web	cloudflare-deploy, durable-objects, vercel-deploy, web-perf
Language patterns	python-patterns, rust-patterns, golang-patterns, kotlin-patterns
Media and visuals	imagegen, video, sora, canvas-design, diagram
Business workflows	market-research, inventory-demand-planning, production-scheduling

Skill Directory Contract

Minimum shape:

skill-id/
  SKILL.md

Expanded shape:

skill-id/
  SKILL.md
  references/
  scripts/
  templates/
  tests/

SKILL.md files should stay focused. Large examples, templates, and background material belong in sibling folders so the runtime can load only the context a task needs.

Quickstart

Clone the library:

git clone https://github.com/lidge-jun/cli-jaw-skills.git
cd cli-jaw-skills

Count skills:

find . -mindepth 2 -maxdepth 2 -name SKILL.md | wc -l

Inspect a skill:

sed -n '1,160p' dev-frontend/SKILL.md
find dev-frontend/references -maxdepth 2 -type f | sort

Run local validation for the public surface:

python3 -m pytest tests/test_dev_frontend_refresh.py -q
python3 scripts/validate_public_surface.py
git diff --check

Validation Policy

The public repo checks should prove four things:

1. The stated skill count matches the filesystem.
2. SKILL.md files are present where expected, and any new line-limit drift is caught. The known long Office skills are tracked separately.
3. Focused frontend regression tests continue to pass.
4. README, Pages, and workflows do not drift from the real skill library.

The added CI workflow validates those checks without publishing anything.

The broader OfficeCLI CJK regression suite under tests/test_cjk_regression.py requires a compatible local officecli binary. In this environment that suite currently fails before exercising these documentation changes because the local OfficeCLI install is missing System.Collections.NonGeneric and has command shape drift for PPTX shape props.

OfficeCLI Sync

.github/workflows/sync-officecli-skills.yml receives OfficeCLI sync events and copies OfficeCLI-derived materials into the docx, pptx, and xlsx skill families.

OfficeCLI guidance in skills_ref is consent-based:

• Check install state first with command -v officecli.
• If OfficeCLI is available, recommend it first for high-fidelity Office work, validation, batch/resident flows, CJK/rhwp-aware behavior, and Office-native output.
• If OfficeCLI is missing, do not auto-install from a skill. Ask the user to choose between installing the supported fork, continuing with lightweight fallback tools for the current task, or stopping.
• The supported cli-jaw fork is https://github.com/lidge-jun/OfficeCLI. Do not direct users to download/install upstream iOfficeAI/OfficeCLI unless they explicitly request vanilla upstream behavior.
• Before lightweight fallback, ask again and state likely feature/fidelity loss. If the user chooses lightweight mode, save that preference to memory for future Office work.

That workflow can commit and push from GitHub Actions when triggered. This local work does not push changes; remote CI and Pages deployment require an authorized push.

Security Notes

• Treat skills as executable instructions: review SKILL.md and helper scripts before promoting a skill to active runtime use.
• Keep secrets out of skill files. Use environment-variable placeholders and user-level configuration.
• Prefer small active skill sets. Reference skills should be loaded on demand, not injected into every prompt.
• Keep generated or downloaded assets inspectable and attributable.

Adding or Updating a Skill

1. Create or update a top-level skill-id/SKILL.md.
2. Move long supporting material into references/, scripts/, or templates/.
3. Run the validation commands above.
4. Update README and Pages counts if the public surface changes.
5. For OfficeCLI-derived material, prefer the sync workflow rather than manual copy-paste.

Maintainer Notes

This repository is a public distribution surface, not just a scratch folder. Avoid stale counts, hidden credential assumptions, and broad active-by-default claims. Every public claim should be backed by a local validation command or the current GitHub repository state.