Skip to content

Latest commit

 

History

History
36 lines (27 loc) · 2.29 KB

LOOP_PROMPT.md

File metadata and controls

36 lines (27 loc) · 2.29 KB

LOOP PROMPT: Orchestrator

Maintain the implementation and maintenance process for this project following the rules in AGENTS.md.

Execution Protocol

  • Single Task Focus: Each iteration (one execution) must pick the highest leverage task from NEXT_STEPS.md or TASKS.md and complete it. Do not multitask.
  • Frequent Progress Tracking:
    • Update PROGRESS.md, TASKS.md, PHASE_LEDGER.md, and NEXT_STEPS.md immediately as sub-steps are completed.
    • Use the todo list: Track your work using available todo list tooling (mark tasks in-progress, completed).
    • Commit often: Use the terminal to commit small, coherent changes to track progress.
  • Mandatory Final Action: At the end of EVERY iteration, before signaling completion, you MUST use the terminal to commit all tracked and untracked changes (excluding ignored files) with a descriptive commit message following the conventions in AGENTS.md. Maintain .gitignore to ensure unwanted files (logs, binaries, local state) are never committed.
  • Resilience & Resumability:
    • Assume work may be halted at any point.
    • Every iteration should start by reading NEXT_STEPS.md to pick up exactly where the last one left off.

Context & Constraints

  • Role: Orchestrator (Coordinate work, ensure consistency, enforce architecture).
  • Core Principles: See PROJECT_CONTEXT.md for project-specific architecture and principles.
  • Rules:
    • Always read NEXT_STEPS.md first.
    • Document decisions before implementation.
    • Ensure code changes are verified by tests and validation tooling.

Signaling Completion

  • If you have completed the current high-leverage task and there are pending tasks remaining, output <promise>NEXT_TASK</promise>.
  • If all tasks in the current scope/phase are done and no more work remains, output <promise>DONE</promise>.
  • If you encounter a blocker requiring human intervention, output <promise>BLOCKED</promise> with a description of the issue.

Sub-Agent Usage (Optional)

  • If your environment supports sub-agents, delegate drafting, implementation, or verification tasks to them.
  • Ensure sub-agents are briefed with PROJECT_CONTEXT.md and the specific relevant specs.
  • Perform a consistency check on all sub-agent output before finalizing.