Prompt file imported from jackofall1232/LOOPRITE (.codex/prompts/resume-loop.md). Copyright stays with the author.

Resume Loop Prompt

You are resuming a l00prite-managed project. Treat .l00prite/ as the shared source of truth across Claude, Codex, GPT, Gemini, and future agents.

This prompt performs one supervised loop iteration: one smallest useful step, verified, persisted, then stop. For an autonomous multi-iteration run that continues until a run boundary, use .l00prite/prompts/execute-loop.md instead — it requires its own pre-flight confirmation before starting.

Required context read

Before changing files, read:

• .l00prite/blueprint.md
• .l00prite/ledger.md
• .l00prite/memory.md
• .l00prite/constraints.md
• .l00prite/failures.md
• .l00prite/todos.md
• .l00prite/state.json
• .l00prite/heartbeat.json
• .l00prite/lock.json

Lock check (required before any write)

• Read .l00prite/lock.json before mutating any protected path (ledger.md, memory.md, state.json, heartbeat.json, failures.md, todos.md, events/, reviews/, sessions/).
• If status is unlocked, released, or expired, acquire the lock: set status: "active", a new lock_id, owner_agent, owner_session, acquired_at, expires_at (acquired_at + ttl_seconds), and purpose.
• If status is active, not expired, and owned by a different agent or session, do not write. Stop and report the lock as a blocker instead of proceeding.
• If status is active, not expired, and already owned by you (matching owner_agent/owner_session), continue writing — you do not need to re-acquire before each write within the same run.
• If status is active but expires_at has passed, or status is explicitly expired, treat it as stale: you may reclaim it (acquire as above), but record a ledger.md entry noting the reclaimed lock_id and why it was judged stale.
• If a step is likely to run longer than ttl_seconds (e.g. a slow test suite), refresh expires_at partway through rather than letting a still-running step look stale.
• Release the lock (status: "released") before stopping, once your memory updates are complete. See .l00prite/LOCKING.md for the full rules.

Precedence rules

• An active, non-expired lock you don't own wins over any write you were about to make.
• state.json.blocked: true overrides heartbeat.json.should_continue.
• Human review gates in heartbeat.json win over normal roadmap work.
• Failed CI, PR review, or other blocker-priority events outrank normal roadmap todos.md items.

Required loop behavior

1. State your understanding of the current project, current goal, status, blocker state, and next recommended action.
2. State what you will not retry, based on .l00prite/failures.md and ledger do-not-retry notes.
3. Check heartbeat limits and the precedence rules above before implementation. Stop if blocked, human review is required, completion is already reached, max iterations are reached, or the lock is held by a different agent or session.
4. Pick the next smallest useful step from .l00prite/todos.md or .l00prite/state.json.
5. Execute only that step. Do not expand scope without human approval.
6. Verify the step with the smallest meaningful test/check available.
7. Update .l00prite/ledger.md with goal, completed work, changed files, tests run (command, exit code, summary, evidence path if available, timestamp), failures, decisions, confidence, next action, do-not-retry notes, and lock status (lock_id acquired/released or none).
8. Update .l00prite/state.json with current goal, phase, active/last agent, last_updated, status, blocked, blocker_reason, and next_recommended_action.
9. Update .l00prite/todos.md to reflect completed and next work.
10. Update .l00prite/failures.md if anything failed or should not be retried.
11. Update .l00prite/heartbeat.json before stopping.
12. Release the lock (.l00prite/lock.json) before stopping.

Stop after the chosen step and memory updates. Every loop must update memory and release the lock before stopping.