Spec Completion Review Mode

Workflow is backend-split. Read workflow-common.md for Phase 0 (backend detection + philosophy), then read ONLY the file matching your active backend:

• BACKEND=codex → workflow-codex.md
• BACKEND=copilot → workflow-copilot.md
• BACKEND=cursor → workflow-cursor.md
• BACKEND=claude → workflow-claude.md
• BACKEND=host → workflow-host.md
• BACKEND=rp → workflow-rp.md

Do not load the others — only the active backend's file is needed.

Verify that the combined implementation of all tasks in a spec satisfies the spec requirements. This is NOT a code quality review (that's impl-review's job) — this confirms spec compliance only.

Role: Spec Completion Review Coordinator (NOT the reviewer) Backends (branch on the Phase 0 RP_ELIGIBLE probe):

• When RP_ELIGIBLE=1: RepoPrompt (rp), Codex CLI (codex), GitHub Copilot CLI (copilot), Cursor CLI (cursor), Claude Code CLI (claude), or host-native (host)
• When RP_ELIGIBLE=0: Codex CLI (codex), GitHub Copilot CLI (copilot), Cursor CLI (cursor), Claude Code CLI (claude), or host-native (host) — rp is macOS-only; never list it in guidance you surface (--review=rp stays accepted)

Preamble — execute Phase 0 exactly once

The executable Phase 0 lives in workflow-common.md §"Phase 0: Backend Detection" — Read it and execute it ONCE, before any other bash in this skill. It defines $FLOWCTL (bundled — NOT installed globally; which flowctl fails, expected), probes RP_ELIGIBLE, resolves $BACKEND via the single flowctl review-backend call, and handles the ASK / none cases. Never invoke flowctl review-backend a second time in the same run.

Exception: a --review=<backend> argument (see Backend Selection below) wins — when present, set BACKEND from the flag and skip Phase 0's review-backend call + ASK handling (still run its $FLOWCTL / RP_ELIGIBLE setup lines).

When RP_ELIGIBLE=0 (not macOS, no supported RepoPrompt CLI), never steer the user toward rp: every backend summary, recommendation, or override hint you surface presents only the runnable configured backends codex, copilot, cursor, claude, host (plus none). Suppression is not a ban: an explicit --review=rp, FLOW_REVIEW_BACKEND=rp, or review.backend=rp still resolves to rp and errors at runtime via require_rp_cli().

Backend Selection

Priority (first match wins):

1. --review=rp|codex|copilot|cursor|claude|host|none argument
2. FLOW_REVIEW_BACKEND env var — bare backend (rp, codex, copilot, cursor, claude, host, none) OR spec form (codex:<model>:xhigh, copilot:<model>, cursor:<model>, claude:<model>:<effort>); host is bare-only (host:<model> is rejected)
3. .flow/config.json → review.backend (same bare / spec forms)
4. Error - no auto-detection

Parse from arguments first

Check $ARGUMENTS for:

• --review=rp or --review rp → use rp
• --review=codex or --review codex → use codex
• --review=copilot or --review copilot → use copilot
• --review=cursor or --review cursor → use cursor
• --review=claude or --review claude → use claude
• --review=host or --review host → use host
• --review=none or --review none → skip review

If found, use that backend and skip all other detection.

Otherwise: Phase 0 resolves it

No --review flag → $BACKEND comes from workflow-common.md Phase 0 (executed once per the Preamble): the single flowctl review-backend "$SPEC_ID" call with ASK handling included. Do not re-resolve here.

Backend at a glance

The per-backend summary (models, env vars, --spec forms) and the backend[:model[:effort]] spec grammar live in references/backend-at-a-glance.md. Read it only when you surface backend guidance to the user (ASK branch, recommendation, override hint) — routing does not need it.

Critical Rules

Per-backend critical rules live in the backend file you route to (workflow-codex.md, workflow-copilot.md, workflow-cursor.md, workflow-claude.md, workflow-rp.md) — each opens with its own Critical rules section. The host safety invariant and the all-backends rules stay here because they gate routing itself.

For host backend: host is bare-only. After selection, read workflow-host.md. The review must use a fresh, tool-enforced read-only reviewer from a different model family and fail closed when no cross-family pin is available.

For all backends:

• If REVIEW_RECEIPT_PATH set: write receipt after SHIP verdict (RP writes manually after fix loop; codex writes automatically via --receipt)
• Any failure → output <promise>RETRY</promise> and stop. No-verdict transport failures are recorded and their reserved round refunded; never manually reset the review counter. Exit 5 / TRANSPORT_UNHEALTHY stops automatic retries until the backend is repaired.

The three hard invariants (never self-declare SHIP, never mix backends, never skip review silently) live with the shared anti-patterns in workflow-common.md §"Anti-patterns (all backends)".

Input

Arguments: $ARGUMENTS Format: <spec-id> [--review=rp|codex|copilot|cursor|claude|host|none]

• Spec ID - Required, e.g. fn-1 or fn-22-53k
• --review - Optional backend override

Workflow

REPO_ROOT="$(git rev-parse --show-toplevel 2>/dev/null || pwd)"

Step 0: Parse Arguments

Parse $ARGUMENTS for:

• First positional arg matching fn-* → SPEC_ID
• --review=<backend> → backend override
• Remaining args → focus areas

Step 0.5: Resume terminal status persistence before dispatch

Run this checkpoint after parsing SPEC_ID and before loading or dispatching any backend. Run the same checkpoint again immediately after host/rp records a verdict. It recovers a terminal status write that failed after the verdict round was durably consumed, without reserving or dispatching another review. Host and rp terminal status has one owner, review-rounds record --status-target completion (with a journaled receipt, that status leg lands when the receipt publishes); this checkpoint only repairs a write that did not land. A stored not_required (work's 3g policy skip) is neither ship nor unknown here: the checkpoint has no terminal attempt to resume for it, and an explicit manual invocation may still run a real review and overwrite it with ship/needs_work — the upgrade direction is legal, while the skip's own write stays gated on unknown.

TERMINAL_REVIEW_JSON="$($FLOWCTL review-rounds resume-terminal "$SPEC_ID" --review-type completion --json)" || exit $?
TERMINAL_ACTION="$(printf '%s' "$TERMINAL_REVIEW_JSON" | jq -r '.action')"
TERMINAL_STATUS="$(printf '%s' "$TERMINAL_REVIEW_JSON" | jq -r '.status')"
TERMINAL_EXIT="$(printf '%s' "$TERMINAL_REVIEW_JSON" | jq -r '.exit')"
case "$TERMINAL_ACTION" in
  continue) ;;
  retry) echo "<promise>RETRY</promise>"; exit "$TERMINAL_EXIT" ;;
  ship) echo "VERDICT=SHIP"; exit "$TERMINAL_EXIT" ;;
  superseded) echo "COMPLETION_REVIEW_STATUS=$TERMINAL_STATUS"; exit "$TERMINAL_EXIT" ;;
  escalate)
    if [ "$TERMINAL_STATUS" = needs_human ]; then
      echo "ESCALATE: reviewer requested human review"
    else
      echo "ESCALATE: completion-review did not converge within the verdict-round cap"
    fi
    exit "$TERMINAL_EXIT" ;;
  *) echo "Unknown terminal review action: $TERMINAL_ACTION" >&2; exit 1 ;;
esac

An exit-4 cap refusal before this run has delivered a completion verdict is non-terminal for completion status: surface ESCALATE: / NEEDS_HUMAN and do not invent a needs_work write. More than ${MAX_REVIEW_TRANSPORT_FAILURES:-2} consecutive transport failures stop separately with TRANSPORT_UNHEALTHY + exit 5; never write completion status or reset the verdict counter for transport health.

Unchanged-artifact terminal: NOT_RETRYABLE: artifact unchanged since last verdict exits 1 before dispatch. Stop for human action; never refund, reset, use --force, or redispatch autonomously. A human may edit the exact artifact, explicitly reset, or deliberately apply --force.

Step 1: Load Backend Workflow

1. $BACKEND was already resolved by workflow-common.md Phase 0 (Preamble) — do NOT re-run it.
2. Read only the file for that backend, per the routing table at the top of this file.

Do not read the other backend files. Each is self-contained for its backend; loading the others wastes context.

Step 2: Execute the backend workflow

Follow the phases in the per-backend file end-to-end. Each file owns its own Identify → Execute → Verdict → Receipt steps (and, for RP, the full Phase 1-4 setup-review (5-15 min, DO NOT RETRY) / chat-send (2-10 min, DO NOT RETRY) / receipt build).

Step 3: Fix loop and terminal status

Both are backend-agnostic and live in workflow-common.md — already in context from Phase 0:

• §"Fix Loop (INTERNAL - do not exit to Ralph)" — the round cap, the anti-patterns, and the parse → fix → commit → re-review cycle.
• §"Record the terminal verdict exactly once" — who writes completion_review_status, and when host/rp re-run the Step 0.5 checkpoint above.