72 lines (49 loc) · 2.67 KB

Claude Project Memory Template

Primary Directive: "Evidence > assumptions | Code > documentation | Efficiency > verbosity"

Project Context

Current Branch: [BRANCH_NAME] (if applicable)
Main Branch: [DEFAULT_BRANCH] (for PRs)
Active Work: [CURRENT_FOCUS] (if applicable)
Task Documents: [TASK_DOC_PATH] (if applicable)

Universal Execution Rules

1. Task Document Compliance

Read task documents COMPLETELY before starting (if provided)
Follow specifications EXACTLY when they exist
Do NOT deviate from the specified plan without explicit approval
Ask for clarification when requirements are unclear

2. Phase-by-Phase Execution

Work sequentially - Complete current phase before moving to next
Respect dependencies - Tasks often depend on previous completions
Use checkpoints between phases (git commits/stash as appropriate)
Document progress before moving to next phase

3. Testing & Validation

Test after changes - Run project's test suite to verify
Zero tolerance for errors - Code must pass all checks
Run specified tests - Use project's testing framework
Validate before commit - Ensure all checks pass

3. Quality Assurance

Validate deliverables - Verify outputs meet specified requirements
Check completeness - Ensure all requirements are addressed
Review before handoff - Validate quality standards are met

4. Git Commit Strategy

Atomic commits - One logical change per commit
Clear messages - Follow project's commit format
Immediate commits - Don't batch unless specified
No broken commits - Each commit should leave code in working state

5. Documentation Updates

Update as you go - Don't leave documentation for later
Record decisions - Document why, not just what
Track issues - Note any problems or deviations
Mark completions - Update task status when done

Subagent Guidance

a. Failure Protocol

STOP → REVERT → ANALYZE → ADJUST → RETRY → ESCALATE (after 2 failures)

b. Execution Strategy

Auto-delegate first - Let Claude Code route tasks intelligently based on context
Explicit invoke - Format: "Use the [agent-name] subagent to [task description]"
Parallel execution - Max 3 concurrent agents for independent tasks
Sequential chains - For dependent outputs: architect → developer → tester → reviewer

c. Coordination Rules

Clear handoffs - Document state changes between agents
Summarize results - Present concise summary of subagent outputs to user
Resource management - Terminate agents when expertise no longer needed