Installable guard skill for agent skills.
deskill is meant to be used as a skill first: install it, invoke it from your agent's skill or slash entry, and ask it to inspect another skill before that skill runs. The Node CLI is only a fallback/runtime helper, not the first user experience.
For Codex:
git clone https://github.com/XiaoLuoLYG/deskill.git /tmp/deskill
mkdir -p ~/.codex/skills/deskill
cp -R /tmp/deskill/SKILL.md /tmp/deskill/agents ~/.codex/skills/deskill/Restart or refresh the agent session, then invoke deskill from the skill/slash picker or say:
Use $deskill to inspect this skill before running it: <path-or-url>
Use the examples in this repo or paste any SKILL.md package path/URL:
Use $deskill to inspect examples/safe-skill before running it.
Use $deskill to inspect examples/malicious-skill before running it.
Use $deskill to inspect examples/irreversible-skill before running it.
Expected outcomes:
safe-skill: allow and save a no-state-change deskill task.malicious-skill: block on prompt injection, credential access, or exfiltration findings.irreversible-skill: save a no-deskill blocker for payment-like side effects.
What you should visibly see from the agent:
deskill process
1. Target: ...
2. Evidence read: ...
3. Risk signals: ...
4. Reversibility: ...
5. Generated: ...
6. Verdict: ...
deskill writes small local receipts under .deskill/ so future agents can see what was checked, what was allowed or blocked, and whether rollback is possible. You do not need to care about this folder for the first experience.
The npm package is for hosts that want MCP tools or a deterministic CLI fallback:
npm install -g deskill
deskill mcp-serve
deskill guard <skill-path-or-url>
deskill listLocal development:
npm install
npm run release:checkApache-2.0. See LICENSE.