How to operate effectively within this repository as an AI agent.
This guide provides the file reading order, extraction strategy, and execution guidelines for AI agents working on this repository.
| Order | File | Extract |
|---|---|---|
| 1 | README.md |
Repository purpose, structure overview |
| 2 | REPO_MAP.md |
Navigation paths, key concepts |
| 3 | core/README.md |
What agentic-ready means |
| Order | File | Extract |
|---|---|---|
| 4 | core/agent-operating-model/AGENT_OPERATING_MODEL.md |
Roles, workflow loop, stop conditions |
| 5 | core/spec/CHECKPOINTS.md |
C0-C4 gates, required proofs |
| 6 | core/agent-operating-model/ESCALATION_RULES.md |
When to pause and ask |
| File | When to Read |
|---|---|
core/agent-operating-model/HANDOFF_TEMPLATE.md |
Transitioning between roles |
core/agent-operating-model/RISK_MODEL.md |
Assessing change impact |
core/templates/ADR_TEMPLATE.md |
Making architectural decisions |
runtime/quality-gates/definition-of-done.md |
Validating completeness |
STOP and escalate if ANY of these apply:
- Architecture or design decisions needed
- Security implications
- Database schema changes
- New external dependencies
- Multiple valid approaches (need human choice)
- Unclear or ambiguous requirements
- Changes affecting > 5 files
See ESCALATION_RULES.md for details.
| Size | Criteria | Workflow |
|---|---|---|
| Trivial | < 10 lines, 1 file, no triggers | Fast Path |
| Small | 10-50 lines, 1-2 files | Lite (C0+C2+C3) |
| Medium | 50-200 lines, 2-5 files | Full (C0-C3) |
| Large | > 200 lines or > 5 files | Full + Extra Review |
When in doubt, use Full workflow.
If ALL criteria met: < 10 lines, 1 file, no escalation triggers, doc/bugfix only.
- Quick C0: State objective in one line
- Implement: Make the change
- Quick C3: Minimal PR with Fast Path template
See CHECKPOINTS.md for template.
- Read the task description completely
- Identify affected files and components
- State the objective in your own words
- List what's in scope and out of scope
- Raise questions if anything is unclear
- Analyze existing code in affected areas
- Design your approach
- List files to create or modify
- Identify risks using RISK_MODEL categories
- Document your plan
- Execute changes per plan
- Follow existing code patterns
- Run available tests locally
- Fix issues iteratively (max 5 iterations)
- Document any deviations
- Create PR with required sections
- Attach all proof artifacts
- Self-review changes
- Request human review
- Read relevant files before modifying
- Work in small, atomic changes
- Provide proofs at each checkpoint
- Escalate when uncertain
- Document your reasoning
- Skip checkpoints
- Exceed iteration limits without escalating
- Make architecture changes without ADR
- Ignore test failures
- Assume when you should ask
When you must escalate:
- Stop current execution
- Document what you attempted
- Format escalation per ESCALATION_RULES.md
- Wait for human response
- Resume only after receiving guidance
## Plan: [Task Name]
### Objective
[Clear statement]
### Files to Modify
- path/to/file1.md — [change description]
- path/to/file2.md — [change description]
### Risks
- [Risk 1]: [Mitigation]
### Checkpoints
- C0: [How verified]
- C1: [How verified]Use the template in CHECKPOINTS.md.
| Action | Guidance |
|---|---|
| Reading codebase | Start with README, follow imports |
| Making changes | Small commits, clear messages |
| Running tests | Execute all relevant test suites |
| Creating docs | Follow existing style |
| Asking questions | Use escalation format |
- System prompt: CLAUDE_CODE_SYSTEM.md
- Role prompts: runtime/prompts/
- Quality gates: runtime/quality-gates/
Follow this guide to operate effectively within this framework.