Imported from earthly/skills (skills/lunar-cataloger/references/guardrail-specs/AGENTS.md). Install upstream with npx skills add earthly/skills --skill guardrail-specs. Copyright stays with the author.

Guardrail Specifications

This directory contains detailed specifications for 500+ Lunar guardrails organized by category.

Purpose

These specs are rough guidelines only — they provide ideas for what guardrails to build and a starting point for implementation. They are not maintained over time and will fall out of date. Do not treat them as the source of truth for Component JSON paths, policy logic, or collector behavior. When a spec conflicts with an existing implementation, the implementation wins.

Each entry includes:

• Summary and detailed description
• Collector requirements (what data to collect)
• Component JSON paths (where data lives)
• Policy logic (what to check)
• Configuration parameters (tunable thresholds)

Files

Usage

Important: Do not read all specs at once! Read only select files, and only in small chunks, as needed. Avoid reading all of them as that is too much information to process at once.

If you need to find a specific spec, you can grep these files for the guardrail identifier.

How to implement a guardrail:

1. Pick a guardrail from any spec file (or grep for one based on the guardrail identifier)
2. Read the ai-context docs for implementation details:
   • about-lunar.md — what Lunar is and how it works
   • core-concepts.md — core concepts
   • collector-reference.md — how to write collectors
   • policy-reference.md — how to write policies
   • strategies.md — implementation strategies
   • component-json-conventions.md — schema design
   • component-json-reference.md — schema reference
3. Implement the collector(s) and policy following the spec