goal-setting
PM-GATED. Raw vision into ratified OKRs and roadmap seeds.
SKILL.mdskills/goal-setting/SKILL.md
GitHubGoal-Setting — Vision to OKRs to Scaffolded Stubs
Posture: vision-in, OKR-out; neither rubber-stamp nor gate. Each KR must be weekly-perceptible (observable moving this week); one that fails is a later-impact aspiration, surfaced as a shaping question.
Entry points:
- Direct invocation — the PM arrives with raw vision, no formed OKR yet.
- Pickup-from-goal-seed — a deferred
kind: goal-seedstub; run this skill on its vision. - Conform intake from a sizing-object —
coordinator:sizinghands an optionalstate/sizings/<id>.yamlahead of Step 1 (route: pm-decision+xl_exit: roadmap, legacyroute: roadmap, or direct-routedroute: goal-settingXXL). Receive-and-use-if-present, never a wall: absent one, Step 1 runs exactly as today, and the sizing lobby never gates or refuses agoal-settinginvocation. Field-crossing table: wiki.
Produces: a ratified OKR (one Objective + ≤5 weekly-perceptible Key Results) via
coordinator-doc-new --type goal; kind: roadmap-seed stubs pre-tagged to the goal, one per
roadmap-worth-of-work; optionally kind: goal-seed stubs for deferred vision-slices; an OFFER to
chain into /roadmap-planning (PM-gated, never auto-run).
When NOT to invoke
- Problem not yet converged →
coordinator:shapeorcoordinator:brainstormingfirst. - Single-feature scope, no broader OKR arc →
coordinator:plan. - Roadmap already has goals (an existing roadmap-seed against a ratified goal) →
/roadmap-planningdirectly. - Several seeds or goal-setting sizings at once →
coordinator:goal-blitz, where the VP-Product Reviewer sets and ratifies the OKRs.
Ceremony
Step 1 — PM states Objective and candidate Key Results
The PM names the Objective (qualitative, direction-setting) and one or more candidate KRs (raw is fine). Do NOT prompt for a format — structure comes from critique, not intake.
Sizing-object present (Entry point 3): it pre-populates this step's raw framing in body prose
only, never new frontmatter (appetite is never a KR or stub cost:). Crossing table: wiki.
Direct-routed XXL: once the PM utterance happens, write pm_resolution.xl_route_assent on
the sizing-object.
Step 1b — Offer to record competitive context (optional)
If the PM's vision names or implies a domain and state/strategic/self-description.yaml's
competitors[] is empty, offer once (else skip): "This reads like <inferred domain> work — want
me to record inspirations / peers / competitors / aspirational-targets in your strategic
self-description? (Skip freely.)" On opt-in, record via
coordinator:strategic-self-description-refresh's scaffold path (schema:
coordinator/schemas/strategic-self-description.schema.json; provenance: curated), never a
second marking store. On decline, go to Step 2 without re-offering.
Step 2 — Dispatch the VP-Product Reviewer as full OKR critic
Do not ask whether to dispatch — invoking this skill IS the request for the dispatch this step names; it dissolves no gate this skill's own body names.
Dispatch via Agent(subagent_type: "coordinator:vp-product", model: "opus"). Inline verbatim:
You are the VP-Product Reviewer (VP of Product, they/them). You are reviewing a draft OKR set for strategic rigor. Your job is to act as a full OKR critic, not a rubber-stamp.
Assess: (1) Is the Objective a real Objective — qualitative, direction-setting, not a metric or tactic in disguise? (2) Are the KRs real Key Results — measurable outcomes, not activity/output proxies? (3) Is the SET reasonable — ≤5 KRs; more means the Objective is unfocused? (4) Weekly-perceptibility per KR — can an agent or PM observe it move this week? If not, flag as a later-impact aspiration with a weekly-perceptible rewrite, or note the PM should defer it.
Return: verdict per element (PASS/FLAG/REJECT), specific rewrite suggestions for flagged/rejected elements, a SET-level verdict (GO/REVISE/REFRAME). Give the EM and PM the material to revise it themselves — do not rewrite the whole OKR yourself.
The dispatch runs in the background: tell the PM to expect a wait; stand down until the VP-Product Reviewer's idle notification, no polling or re-dispatch.
Step 3 — EM+PM integrate the VP-Product Reviewer's critique in-dialogue
No artifact exists yet. REJECT items are rewritten or dropped; FLAG items rewritten or accepted
with a stated rationale; weekly-perceptibility notes rewritten or deferred to a kind: goal-seed
stub. EM presents the critique, proposes a revised shape, asks the PM to confirm.
Do NOT proceed to Step 4 without explicit PM confirmation on the revised OKR.
Step 4 — Scaffold the goal artifact
Resolve every CLI in Steps 4-5 per the ladder in
${CLAUDE_PLUGIN_ROOT}/snippets/resolve-coordinator-bin.md.
coordinator-doc-new --type goal --title "<objective-slug>".
Fill: objective: (ratified text), key_results: (≤5, weekly-perceptible), period: (schema enum
day|week|repo|quarter|year, e.g. period: quarter, period_value: Q3-2026), status: active.
Once filled, invoke emit-goal-from-artifact <path> (the goal-pipeline emitter per
goals-okr-system.md § Emit → Cockpit → Cockpit Pipe), where <path> is the state/goals/*.yaml
path coordinator-doc-new --type goal printed — scaffolding alone does not emit. Flags:
emit-goal-from-artifact --help.
Step 5 — Spawn downstream stubs
5a. Roadmap-seed stubs — one per roadmap-worth-of-work (one /roadmap-planning invocation,
one coherent capability arc; in doubt, fewer/larger — the PM can split at pickup):
coordinator-doc-new --type roadmap-seed --goals "<goal-id>" --title "<roadmap-topic>".
Each stub carries kind: roadmap-seed, origin_goal_id: FK (via --goals), deployment_state: awaiting_gate, a one-line title naming the capability arc. Split on the gate: work needing an
unbuilt dependency gets its own seed carrying that gate; work that does not is a separate, ungated
seed (A-GATED-SEED-PARKS-ITS-UNBLOCKED-HALF).
5b. Goal-seed stubs (optional) — for vision-slices out of scope this period, or KRs deferred
rather than rewritten: coordinator-doc-new --type goal-seed --title "<deferred-vision-slice>".
Each carries kind: goal-seed, deployment_state: awaiting_gate, and a brief body capturing the
vision-slice verbatim — raw over polished.
Step 6 — Offer the PM-gated chain
"Goal artifact and {N} roadmap-seed stub(s) scaffolded. Want me to chain into
/roadmap-planningnow, or review the stubs first?"
Wait for PM response. Never invoke /roadmap-planning without explicit PM direction. A PM
who already directed the goal through to done in this session ("throw it into the machine", "no
more questions") HAS given it: quote those words and chain without asking. Name any un-chained
seed in the DONE report as awaiting PM ack — never a silent park.
Step 7 — Commit goal + stubs
Commit per snippets/scoped-commit-route.md, no blanket add: subject goal-setting: ratify <objective-slug> + scaffold {N} downstream stubs, pathspec exactly the goal artifact plus the
stubs this run scaffolded.
Out of scope: roadmap authoring (/roadmap-planning), KR tracking infrastructure
(goal.schema.json), fleshing out a kind: goal-seed stub without PM pickup (entry point 2).
Self-verify before DONE: the VP-Product Reviewer dispatched; PM confirmed the revised OKR before any artifact hit
disk; goal artifact scaffolded then emitted; each roadmap-seed carries origin_goal_id: and
deployment_state: awaiting_gate; /roadmap-planning offered, not invoked; commit scoped.