raisedadead/run

dp-cto:run

Current Stage

!"${CLAUDE_PLUGIN_ROOT}/bin/dp-engine" state 2>/dev/null || printf '{"stage":"idle","epic_id":""}'

Read ${CLAUDE_PLUGIN_ROOT}/shared/cto-boundaries.md and ${CLAUDE_PLUGIN_ROOT}/shared/dispatch-protocol.md before proceeding.

Context Detection

Before proceeding, detect the current state:

1. Check for active loop sessions:

   ls .claude/ralph/*.md 2>/dev/null
   

2. Check tracker availability:

   dp-engine state >/dev/null 2>&1 && echo ok
   

   If the tracker is NOT available, STOP. Report: "Tracker not available. Run /dp-cto:plan first to initialize and create a plan."

3. Check dp-cto stage:

   dp-engine state
   

   Returns JSON: {"epic_id":"...","stage":"..."}. If it fails or returns empty, treat as idle.

Route based on context:

| Condition | Action | | ----------------------------------------------------------------- | --------------------------------------------------------------------------------------- | | Running loop exists (.claude/ralph/*.md with status: running) | Present menu via AskUserQuestion: Continue running / Cancel loop / Start new dispatch | | No loop, stage is planned | Start default task-by-task dispatch | | Stage is idle | Report: "No planned work. Run /dp-cto:plan first." and STOP | | Args are non-empty (internal orchestrator hint) | Use args as mode: loop → loop mode, stop → cancel mode |

Default Mode

Task-by-task dispatch. Read the full execution protocol from ${CLAUDE_SKILL_DIR}/references/run-default.md and follow it.

Loop Mode

Iterative autonomous execution. Read the full loop protocol from ${CLAUDE_SKILL_DIR}/references/run-loop.md and follow it.

Pass remaining args as the loop's PROMPT, max-iterations, completion-promise, and quality-gate parameters.

Cancel Loop

1. Read each .claude/ralph/*.md file, extract session_id, status, current_iteration, max_iterations from YAML frontmatter.

   • Discard files where status is NOT running.

2. If no running sessions: say "No active loop sessions found." and STOP.

   • If one running session: show status, ask "Cancel this loop?"
   • If multiple: show numbered list with session ID, iteration count, first line of task. Ask which to cancel.

3. On confirmation, update the state file's YAML frontmatter:

   • Set status: cancelled
   • Set completed_at: to current ISO timestamp
   • Report: "Cancelled loop session {SESSION_ID} at iteration {N}. State file preserved at .claude/ralph/{SESSION_ID}.md"

Do NOT delete state files. They are permanent records.

NEVER

1. NEVER write application code yourself unless the task is tagged [direct] — delegate all other tasks
2. NEVER run default mode without a completed plan from /dp-cto:plan
3. NEVER spawn more than 4 agents in a single round
4. NEVER skip the builder/validator review for any task
5. NEVER declare loop completion without promise appearing verbatim (if set)
6. NEVER exceed max_iterations in loop mode
7. NEVER continue after detecting status: cancelled in a loop state file