Agent skills for OpenAgentLock. Each skill teaches an LLM-based assistant (Claude, Codex, Cursor, etc.) how to generate the right agentlock command for a security-relevant intent — most often: turning "I want to block this" into the YAML rule + the install command that lands it in the live policy.
The companion openagentlock/rules repo is the catalog of pre-baked rules. This repo is the toolkit that lets an agent author new ones from natural-language intent and wire them up.
This repo is shaped for the skills CLI from vercel-labs (the open agent-skills ecosystem). Install one or all skills with a single command — no clone, no symlink:
# list what this repo ships
npx skills add openagentlock/skills --list
# install one skill into your default harness (Claude Code by default)
npx skills add openagentlock/skills --skill block-pattern
# install every skill in this repo
npx skills add openagentlock/skills --all
# pick the harness explicitly
npx skills add openagentlock/skills \
--skill block-pattern \
-a claude-code -a cursorThe skills CLI handles the per-harness install paths and conventions; this repo just supplies the SKILL.md files.
skills/<skill-name>/
├── SKILL.md # the prompt the agent loads (frontmatter + instructions)
├── README.md # human-facing description
└── examples/ # input/output pairs that train the agent's expectations
Each SKILL.md carries the YAML frontmatter (name, description, ...) the skills CLI uses for discovery, then the instructions the agent reads at runtime.
| Skill | What it does |
|---|---|
block-pattern |
Convert "block X" intent into a rule.yaml plus the agentlock rules install invocation that registers it. |
More on the way:
audit-deny— given a recent ledger deny, suggest a tighter rule and PR it to a private rules registry.propose-rule— author a fresh community rule and open a PR againstopenagentlock/rules.triage-flag— when a daemon-side gate fires, classify whether the deny was correct and suggest a rule patch.
If a skill you'd like is missing, open an issue or send a PR.
Once installed, ask the agent in natural language: "Block any bash command that pipes a file into nc." The skill kicks in, drafts a rule.yaml, runs agentlock rules install against your daemon, and reports back.
These skills only emit and execute the official agentlock CLI — they don't reach into the daemon or write policy YAML directly. The CLI is the single entry point for installing a rule into the policy, so every skill's output is auditable in your shell history and in the OpenAgentLock ledger entry that records the install.
You should still review the generated rule.yaml before installing in production. The skill optimizes for first-draft quality, not for safe deploys.
Apache-2.0. See LICENSE.