agentic-coding-playbook: Skill for Claude Code

Federal AGENTS.md Configuration

This skill generates a project-specific AGENTS.md by walking the user through compliance-relevant questions and producing a customized behavioral contract.

The generated file is a thin, project-specific layer: it references the universal Federal AI Agent Behavioral Best Practices (the universal AGENTS.md) as a prerequisite and does NOT restate the universal rules. This keeps a single source of truth for the universal contract and avoids copy-paste drift. The generated file instructs the agent to STOP if it cannot confirm the universal contract is available, and to require the user's affirmative permission before proceeding without it.

When to Use

• User needs to create an AGENTS.md for a new federal project
• User wants to update their existing AGENTS.md for compliance
• User asks "how do I configure my AI agent for federal requirements"
• User mentions FIPS, ATO, or federal compliance in the context of agent setup

How It Works

1. Ask the user compliance-relevant questions (elicitation sequence below)
2. Accumulate answers into a JSON configuration object
3. Run scripts/generate-agents-md.py with the config to produce AGENTS.md
4. Present the output for human review
5. Iterate if the user wants changes

Configuration Object

Track this JSON structure as you collect answers. Fields marked required must be filled before generation. Fields marked optional have sensible defaults.

{
  "system_name": null,
  "agency_name": null,
  "system_description": null,
  "impact_level": "moderate",
  "language": null,
  "framework": null,
  "data_classification": "internal",
  "ato_status": "pre-ato",
  "agent_names": [],
  "prohibited_actions": [],
  "permitted_actions": [],
  "approval_required_actions": [],
  "sensitive_data_types": [],
  "approved_storage": [],
  "secrets_backend": null,
  "approved_registries": [],
  "license_restrictions": [],
  "network_allowlist": [],
  "test_command": null,
  "test_coverage_target": "80",
  "ci_checks": ["lint", "test", "sast", "sca", "secrets-scan"],
  "branch_protection": "1 review, no force push",
  "project_lead": null,
  "security_contact": null,
  "isso_contact": null,
  "reviewed_by": null
}