/acreadiness-policy — AgentRC policies

Use this skill when the user asks about policies, strict mode, custom scoring, disabling checks, org standards, or CI gating of readiness.

A policy is a small JSON file with three optional sections — criteria, extras, thresholds — that customise how AgentRC scores readiness.

Built-in examples

AgentRC ships with three example policies in examples/policies/:

Recommend these as starting points before writing a custom policy.

Policy schema

{
  "name": "my-policy",
  "criteria": {
    "disable":  ["env-example", "observability", "dependabot"],
    "override": {
      "readme":      { "impact": "high", "level": 2 },
      "lint-config": { "title": "Linter required" }
    }
  },
  "extras": {
    "disable": ["pre-commit"]
  },
  "thresholds": {
    "passRate": 0.9
  }
}

Impact weights

Score = 1 − (deductions / max possible weight). Grades: A ≥ 0.9, B ≥ 0.8, C ≥ 0.7, D ≥ 0.6, F < 0.6.

Sub-commands

show

List policies currently in effect (from agentrc.config.json policies array, or none).

new <name>

Scaffold policies/<name>.json with sensible defaults. Walk the user through:

1. What to disable — irrelevant pillars or extras for their stack (e.g. disable observability for a static site).
2. What to raise — override impact to high or critical for must-haves (e.g. readme, codeowners).
3. Pass-rate threshold — typical org baselines: 0.7 (lenient), 0.85 (standard), 1.0 (strict).
4. Reference the policy from agentrc.config.json:

   { "policies": ["./policies/<name>.json"] }
   

apply <path-or-pkg>

Run agentrc readiness --json --policy <source> and re-render the report by handing off to the assess skill / ai-readiness-reporter agent. Supports chaining:

npx -y github:microsoft/agentrc readiness --json --policy ./org-baseline.json,./team-frontend.json

CI gating

Combine policies with --fail-level to enforce a minimum maturity level in CI:

- run: npx -y github:microsoft/agentrc readiness --policy ./policies/strict.json --fail-level 3

Advanced

JSON policies can disable, override, and set thresholds — but cannot add new criteria. For new detection logic, point users at AgentRC's TypeScript plugin system (docs/dev/plugins.md).

Operating rules

• Never silently disable a pillar. If the user wants to disable observability, confirm and explain the trade-off.
• Prefer overriding impact over disabling. Disabling hides the gap entirely; overriding lets it still appear in the report.
• Recommend extras stay enabled. They cost nothing — they don't affect the score.
• Suggest layering — most orgs want a baseline policy + per-team overrides chained with --policy a.json,b.json.