🛡️ Skill Arsenal

Pro-grade agent skills — battle-tested procedures for autonomous agents, orchestration, debugging, ops, and the craft of skills themselves.

A curated collection of SKILL.md files written to a strict authoring contract, validated by CI, and designed to change agent behavior — not just document it. Every skill here follows the same discipline: trigger discipline, checkable completion criteria, no-op-free prose, and a verification checklist you can run.

What is a skill? In agent frameworks like Hermes Agent, a skill is a markdown procedure an agent loads to make its process predictable. This repo is a library of the good ones.

📦 The Skills

Skill	Category	What it does
bounded-autonomy-loop	Autonomous Agents	Run unattended work with checkpointed state, heartbeat reporting, and a hard-stop contract. Survives crashes, resumes where it left off, never spins forever.
multi-agent-orchestration	Autonomous Agents	Fan-out/fan-in parallel subagents done right — self-contained briefs, verification-handle rules, and merging that catches the lies children tell.
systematic-debugging	Software Dev	4-phase root-cause method: reproduce → isolate → instrument → fix with evidence. No guessing, no "fixed" without proof.
cron-ops	DevOps	Resilient scheduled jobs: idempotency, loud failure, watchdogs, and retry taxonomy. Turns cron from a liability into infrastructure.
service-hardening	DevOps	Production hardening: graceful shutdown, honest health endpoints, classified retries, circuit breakers, bounded resources.
skillcraft	Meta	The discipline of writing skills that change behavior — trigger discipline, completion criteria, anti-sediment rules. The skill that governs all the others.

🚀 Install

Skills install into any agent framework that reads SKILL.md (Hermes Agent, Claude-style agents, etc.):

# Hermes Agent — install from this repo
hermes skills install github:hermesxclaw-ctrl/skill-arsenal

# Or just clone and point your skill loader at skills/
git clone https://github.com/hermesxclaw-ctrl/skill-arsenal.git

✅ The Authoring Contract

Every skill in this repo passes the validator in scripts/validate_skills.py, which enforces:

• Frontmatter at byte 0, valid YAML, non-empty body
• name ≤ 64 chars (lowercase+hyphens), description ≤ 1024 chars
• Description starts with "Use when..." (trigger discipline — the first ~57 chars are what agents see in indexes)
• version / author / license / metadata.hermes.{tags, related_skills} present
• Required sections: ## Overview, ## Common Pitfalls, ## Verification Checklist
• Total file ≤ 100,000 chars (target 5–15K: dense, not sprawling)

Run it locally:

python scripts/validate_skills.py          # required checks
python scripts/validate_skills.py --strict # + recommended fields

CI: .github/workflows/validate.yml runs the validator on every push/PR. A skill that doesn't pass doesn't merge.

🧭 Contributing

Want to add a skill? Follow skillcraft — it's the meta-skill that defines the standard:

1. Name the behavior change first (if the agent would do it anyway, don't write it)
2. Write trigger-first description, numbered steps with completion criteria, pitfalls, verification checklist
3. Run python scripts/validate_skills.py --strict until green
4. Open a PR — CI will check it again

📄 License

MIT — use it, fork it, teach your agents from it. See LICENSE.

---

Never finished, always evolving — same as the agents it serves.