Rule Author Agent Skill

Teach any AI coding agent to turn natural-language business rules into compiled WASM governance policies enforced by the local Intutic proxy.

Where the Kitkat skill teaches an agent to operate Intutic governance, the Rule Author skill teaches it to extend it: the agent converts a plain-English rule ("block any agent call that touches the prod database") into an AssemblyScript policy, compiles it with the open-core AssemblyScript Rules SDK template (packages/wasm-sdk/, copied into your project — it is not an npm package), dry-runs it against mock contexts, and installs it into ~/.intutic/wasm/ — where the proxy hot-loads it within ~5 seconds. The full loop runs locally with no control plane.

---

How it works

1. You state a rule in the agent chat: "never allow shell tool calls containing rm -rf" — or /fix tells you one is missing. When the proxy scores your governance posture and finds no custom policy, it recommends "add a WASM rule under ~/.intutic/wasm to codify a custom guardrail"; that recommendation is this workflow's other entry point. Run /fix again after installing to confirm the score moved — if it did not, the rule is not loading, and intutic policy list-local plus the sandbox limits below are where to look.
2. The agent authors an AssemblyScript rule against the WASM SDK's RequestContext contract.
3. Compile: intutic policy compile --src assembly/index.ts --out build/rule.wasm
4. Dry-run: intutic policy test against both a should-block and a should-allow mock — the skill requires both to pass before install.
5. Install: intutic policy install --wasm build/rule.wasm --name <name> --priority NN — validated by instantiation first, then copied to the local rules directory the proxy rescans every ~5 s.

Local rules merge with any centrally synced rules most-restrictive-wins: a BLOCK from either source ends the request, so a local rule can add restrictions but can never neutralize a central one.

---

Setup

Manual

Create the skill directory and download the skill file into it:

mkdir -p .agents/skills/intutic-rule-author

Download SKILL.md

---

Guardrails built into the skill

• The agent must produce both a should-block and a should-allow mock and verify them with intutic policy test before installing.
• intutic policy install refuses rules that fail instantiation — the proxy would refuse to load them, so they would enforce nothing.
• Secrets (API keys, tokens, passwords) stay out of rule sources and mock contexts. Nothing scans those files for them: intutic skill audit reads instruction files and skill directories, not rule sources.
• Allow-overrides are structurally impossible: the verdict lattice only adds blocks.

See the WASM rules guide for the full authoring contract and sandbox limits.