coord-requeue

Recover stuck coord tasks. Status flip + hash reset, optionally with a model write in the same coord update call.

Project-root preflight:

• Resolve the canonical project checkout before reading, updating, staging, committing, or pushing: COORD_MAIN=$("${COORD_TOOLS:-$HOME/Projects/coord-wright}/bin/coord-project-root") && cd "$COORD_MAIN"
• This is required when the skill is invoked from a sibling task worktree; requeue must update the main queue that the launchd worker reads.

Syntax

/coord-requeue <task-id> [--model=<name>]

Workflow

1. Read current state.
   • python3 "${COORD_TOOLS:-$HOME/Projects/coord-wright}/bin/coord" show <id> → capture status and assigned.
2. Validate source status.
   • Accepted: pending, claude-working, codex-working.
   • Refuse needs-review, review-passed, review-failed, needs-brainstorming, done, shaping. Print the current status and which skill to use instead:
     • needs-brainstorming → /coord-promote after shaping
     • done → /coord-discard (already finished) or audit
     • review queue → diagnose, do not silently rescue
     • shaping → /coord-promote
3. Validate --model (if supplied).
   • Infer agent from model name: opus|sonnet|haiku|*claude* → claude; gpt-*|*codex* → codex.
   • Inferred agent must equal task assigned. Reject cross-agent model swaps with a clear message: cross-agent model swap requires explicit re-assignment first: run /coord-assign <id> --agent=<other> --model=<m>, then /coord-requeue <id>
4. Apply changes.
   • If status was *-working: python3 "${COORD_TOOLS:-$HOME/Projects/coord-wright}/bin/coord" update <id> --status=pending (add --model_<agent>=<value> to the same call when --model was given).
   • If status was already pending and --model was given: python3 "${COORD_TOOLS:-$HOME/Projects/coord-wright}/bin/coord" update <id> --model_<agent>=<value>.
   • If the update exits 3 as off-baseline for the task's complexity, report it and stop; do not retry with other values. A different rung needs the complexity changed in the same call (--complexity=<value>), and only after the user agrees.
   • If status was already pending and no --model: the coord CLI only clears the content hash on a non-runnable → runnable transition. A pending → pending write does NOT clear it. In this case, also run python3 "${COORD_TOOLS:-$HOME/Projects/coord-wright}/bin/coord" update <id> --status=pending to force a write, but warn the user that the hash may not be cleared and the task may not be picked up as fresh work. True hash-clearing for an already-pending task requires a pending→shaping→pending round-trip or a direct CLI flag if one is added.
   • Each coord update auto-commits-and-pushes. Do not run manual git add, git commit, or git push.
5. Verify hash was cleared.
   • python3 "${COORD_TOOLS:-$HOME/Projects/coord-wright}/bin/coord" show <id> and check that content_hash_<assigned> is absent from the output. If still present on an already-pending task, warn the user about the limitation above.
6. Report.
   • requeued: <id> (was <prev-status>, now pending) — include model_<agent>=<value> if changed.
   • Note the commit sha from coord update's output.

Rules

• Never use --force or --no-verify.
• Never cross agents via --model — reject and point at /coord-assign.
• Never touch review-queue states (needs-review, etc.) — those need a human look, not a kick.

Token-out recovery flows

• Same agent, cheaper model: one command — /coord-requeue <id> --model=<cheaper>.
• Switch agent entirely: /coord-assign <id> --agent=<other> --model=<m> then /coord-requeue <id>. The agent switch is a deliberate decision, not bundled into recovery.