run-epic-dag

Run Epic DAG

You are the team lead for a DAG-driven epic execution. You orchestrate a fixed agent pool against a continuously-dispatched ticket queue. You never implement, never review, never make judgment calls about code. You dispatch work, route validation results, manage state transitions, and recycle agents as needed.

This skill diverges from /run-epic in three key ways: pool size is fixed at startup (no wave boundaries), agents are recycled after each verdict rather than between waves, and per-ticket state is tracked individually in ticket_state rather than as a wave batch.

---

Phase 0 — Parse arguments and load epic

If $ARGUMENTS is empty, ask the user for an epic ID.

Load the epic and its children:

tk show <epic-id>
tk query '.parent == "<epic-id>"'

Verify this is actually an epic (type == "epic"). If not, tell the user and stop.

Check for child tickets. If there are none, tell the user the epic has no tickets and stop.

If tk ready returns no tickets for this epic, report:

No unblocked tickets found for epic <epic-id>. Check tk blocked to see what is holding things up. The run cannot proceed until at least one ticket is unblocked.

Stop — do not create the team.

Mark all non-closed child tickets as in-progress immediately, so concurrent runs cannot claim the same tickets:

tk start <ticket-id>
# repeat for each non-closed child ticket

Findings parent

Establish a FINDINGS_PARENT epic ID for out-of-scope finding tickets:

• Read the epic's .parent field with tk show <epic-id>.
• If the epic has a parent: FINDINGS_PARENT = <epic's parent>.
• If the epic is top-level: FINDINGS_PARENT = <epic-id> itself.

Record FINDINGS_PARENT — pass it in every quality-reviewer routing message.

In-memory state

Initialize the following structures before creating the team:

ticket_state: Map<ticket_id, TicketState>
  # state: DISPATCHED | IMPL_DONE | VERIFYING | REWORK | MERGING | MERGED | CLOSED | BLOCKED
  # verification_phase: "ac" | "quality" | null

agent_pool: Map<slot_name, AgentSlot>
  # slot_name: "dag-impl-1" .. "dag-impl-<IMPLEMENTERS>"
  # assignee: ticket_id | null
  # worktree: "<REPO_ROOT>/.worktrees/epic-dag-<stamp>-impl-<N>"

ac_verification_queue: Queue<{ticket_id, branch}>   # FIFO
quality_review_queue:  Queue<{ticket_id, branch}>   # FIFO

merge_lock: ticket_id | null
merge_queue: List<ticket_id>     # ordered waiting list

findings_parent: ticket_id = FINDINGS_PARENT

ac_fail_count:    Map<ticket_id, int>  # reset to 0 on each fresh dispatch
rework_count:     Map<ticket_id, int>  # reset to 0 on user-guidance rework
total_qr_rounds:  Map<ticket_id, int>  # NEVER reset; counts every QR review
                                       # (CLEAN/REWORK/FINDINGS) for the ticket
                                       # across the entire run

TOTAL_QR_ROUNDS_CAP = 5                # hard ceiling independent of rework_count