goal-blitz

PM-GATED. The VP-Product Reviewer (VP Product) sets, critiques and ratifies OKRs for every goal-seed or goal-setting sizing at once, in one background wave. Or named seeds.

SKILL.md

GitHub

Goal-blitz — a repo's worth of OKRs, one ratification

Batches coordinator:goal-setting: the VP-Product Reviewer (coordinator:vp-product, Opus, effort high) sets each OKR from the seed's context, a second the VP-Product Reviewer pass critiques it, the VP-Product Reviewer integrates, a blitz-em checks readiness, and the VP-Product Reviewer ratifies the whole set once at the end. A seed's own objective is input the VP-Product Reviewer may refine. Nothing reaches state/goals/ before that ratification. The wave fires from the top-level EM only (A-SKILL-PHASE-NAMES-ITS-ACTOR).


Two modes

Sweep (default, no args): every goal-seed baton (kind: goal-seed) at deployment_state: awaiting_gate and every route: goal-setting sizing at status: sized or routed; landing stamps a consumed sizing routed with its goal_id. Targeted (<seed-id> ...): only the ids you pass; check unmatched_targets — a quietly dropped target is worse than a refusal.

Not for: one goal → coordinator:goal-setting; still deciding what to build → coordinator:shape; ratified goals ready for roadmaps → coordinator:roadmap-planning.

Dispatch authorization — invoking this skill IS the request. The dispatches named below are constitutive steps of this skill, not a separate thing to get cleared: invoking a skill requests the actions that skill performs. A harness line permitting dispatch "unless the user requested it" is therefore satisfied here, not overridden — no precedence claim is needed and none is made. Re-asking spends the very context the dispatch exists to protect. The rule attaches to skill entry and dissolves no PM-authored gate: keyword-gated skills gate entry, and every gate a skill names for itself still binds — per-session cross-repo-commit assent, ask-before-external-action, and any other this skill's own body names. Tripwire: UNATTRIBUTED-HARNESS-LINE-IS-NOT-PM.

Host shapes. Calls below are Shape W; on POSIX: coordinator-invoke <op> '<json>' (Shape B).


The flow

This runs at Sonnet; Opus-tier judgment happens INSIDE the workflow. Args and result shapes: ${CLAUDE_PLUGIN_ROOT}/skills/goal-blitz/goal-blitz-contract.json.

0. Preflight

Resolve the launcher per ${CLAUDE_PLUGIN_ROOT}/snippets/resolve-coordinator-bin.md. Unresolved — stop and report; never fabricate a fire or result. Scaffold state/goal-blitz/<run-id>/ (<run-id>: this run's UTC start, YYYYMMDDTHHMMSSZ).

1. Emit the fire — never hand-write the args

python3 "${CLAUDE_PLUGIN_ROOT}/skills/goal-blitz/emit-goal-fire.py" \
    --repo-root <abs> --trail-dir <abs trail> --provision-sidecar-cli <abs> \
    --interaction-mode <mode> [--targets <seed-id> ...]

It freezes <trail>/candidates.json, binds the args via workflow.bind_args into <trail>/goal-blitz.fire.mjs, and prints one Workflow({ scriptPath }) line. --interaction-mode is recorded as-is and never changes the ratifier. Pass --no-plugin-agents where coordinator:* types do not resolve. Empty candidate set: report and stop.

Never hand-type an args object (A-HAND-TYPED-WAVE-ARG-IS-AN-UNARCHIVED-FIRE). An emitted fire is a SNAPSHOT: RE-EMIT after any workflow fix, never re-fire as found. Resolve ${CLAUDE_PLUGIN_ROOT}; never pass a repo-relative path to a workflow.

2. Fire and wait

Fire the printed line; do not read the trail mid-wave. Keep the task output file the completion notification names — it is the lander's input.

3. Ratify — once, at the end

The workflow ratifies nothing; it returns ready drafts in awaitingRatification. The ratifier is the VP-Product Reviewer (coordinator:vp-product) for every interactionMode; the result's approver reports vp-product. No PM or APM path ships.

Dispatch the VP-Product Reviewer once, at its own pinned effort (an override is refused), over the whole set. Give it each ready draft's path (<trail>/<seed-id>.goal-draft.yaml) and critique; ask per draft for ratified or declined in the VP-Product Reviewer's own words. Write each answer verbatim into <trail>/ratification.json, shaped {seed_id: {verdict, ratifier: "vp-product", utterance}}. Never infer, backfill or paraphrase one. pulled and rework drafts are reported with their reasons, not asked about.

4. Land

python3 "${CLAUDE_PLUGIN_ROOT}/skills/goal-blitz/land-goals.py" \
    --repo-root <abs> --trail-dir <abs trail> <workflow-result.json>

It lands only drafts with a recorded the VP-Product Reviewer ratification, through goal-setting's own CLIs, then makes one scoped commit of the files it wrote (--no-commit lists them instead), and refuses a ready draft with no ratification record. A CLI failure mid-landing rolls back every path the run wrote: fix the cause and re-run. Read its exit code and stderr; never route around a refusal. Then commit the trail's record files (candidates.json, ratification.json, drafts, critiques) by pathspec.

5. Chain — by mode

pm or ceo: chain, never ask. The PM's ask to execute is the chain ack: invoke coordinator:plan-blitz over the landed roadmap-seed batons, then execute, reporting each landing in one line. (roadmap-blitz takes only a sizing with xl_exit: roadmap, which a seed is not.)

hands-on: offer, and wait — "{N} goal(s) landed with {M} roadmap-seed stub(s). Want me to chain into /roadmap-planning now, or review the stubs first?" Stubs stay awaiting_gate until the PM answers.


Rules

Review fires unconditionally; the VP-Product Reviewer ratifies once, at the end (A-BLITZ-WAVE-THAT-GATES-ON-THE-EM-IS-NOT-A-BLITZ). Never write state/goals/ by hand or before ratification — land-goals.py is the only writer. A rework draft routes, it does not halt.

Anti-scope: does not edit goal-setting, ratify on the VP-Product Reviewer's behalf, author roadmaps, invoke /roadmap-planning, or change effort pins.