Agent Task Bootstrap

Read project documentation AND the current PROMPT.md task instructions. Use this when starting work in an agent worktree.

Instructions

1. Read CLAUDE.md: Find and read CLAUDE.md in the current working directory or repository root

2. Follow references: If CLAUDE.md references other CLAUDE.md files (e.g., in submodules), read those as well

3. Read PROMPT.md: Find and read PROMPT.md in the current directory to understand:

   • The specific task you need to complete
   • Acceptance criteria
   • Files to create or modify
   • Inventory checks to perform
   • Completion criteria

4. Acknowledge and Begin: After reading, confirm:

   • What task you're working on
   • Key deliverables
   • The completion promise format you'll use

Expected PROMPT.md Structure

PROMPT.md files contain:

• Task description and user stories
• Files to create/modify
• Inventory checks (prerequisites to verify)
• Completion criteria
• Promise format: <promise>TASK_COMPLETE</promise> or <promise>TASK_BLOCKED: [reason]</promise>

Example Output

Context loaded:
- Project: [name]
- Task: [task number and title]

Assignment:
- [Key deliverable 1]
- [Key deliverable 2]
- ...

Inventory checks to perform:
- [ ] [Check 1]
- [ ] [Check 2]

Completion promise: TASK_COMPLETE

Beginning work...