oduffy-delphi/goal-setting

Goal-Setting — Vision to OKRs to Scaffolded Stubs

Posture: vision-in, OKR-out. Neither rubber-stamp nor idea-crushing gate. Each KR must be weekly-perceptible (can an agent or PM observe it move this week?); one that fails is a later-impact aspiration, surfaced as a shaping question, not a rejection.

Entry points:

1. Direct invocation — the PM arrives with raw vision, no formed OKR yet.
2. Pickup-from-goal-seed — a deferred kind: goal-seed stub; run this skill on the captured vision.
3. Conform intake from a sizing-object — coordinator:sizing hands an optional state/sizings/<id>.yaml entry contract ahead of Step 1, on three arrival shapes (route: pm-decision+xl_exit: roadmap, legacy route: roadmap, or direct-routed route: goal-setting XXL). No sizing-object present means Step 1 runs exactly as today — see wiki for the full field-crossing table.

What this skill produces: a ratified OKR (one Objective + ≤5 Key Results, weekly-perceptible) 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).

<HARD-GATE> This skill DOES NOT auto-author roadmap plans, invoke `/roadmap-planning` without PM acknowledgment, or bypass its PM gate. It scaffolds stubs and offers the chain. The PM decides whether and when to fire each stub. </HARD-GATE>

---

When NOT to invoke

• Problem not yet converged (still deciding what to build) → coordinator:shape or coordinator:brainstorming first.
• Single-feature scope, no broader OKR arc → straight to coordinator:plan.
• Roadmap already has goals — picking up an existing roadmap-seed stub against an already-ratified goal → /roadmap-planning directly.

---

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 specific format — the ceremony imposes structure through critique, not intake.

When a sizing-object is present (see Entry point 3), it pre-populates part of this step's raw framing in body prose only, never new frontmatter: intent → opening vision statement; appetite → context only, never a KR or stub cost:; estimate.tshirt+estimate.provisional together → routing signal only; scout_evidence → framing pointers; route(+xl_exit) → the recorded reason this ceremony opened. Full field-by-field detail and does/doesn't-cross list: wiki.

PM-assent record for a direct-routed XXL: that shape emits no xl_exit, so once this step's PM utterance happens, write pm_resolution.xl_route_assent on the sizing-object recording it as the assent — this does not reintroduce a PM halt above goal-setting (Step 1's own utterance already is the gate). Detail: wiki.

Step 1b — Offer to record competitive context (optional)

If the PM's vision names or implies a domain, check state/strategic/self-description.yaml's competitors[] first — skip the offer silently if already populated (same discipline as workweek-start/repo-setup Phase 3l nudges). Otherwise offer once:

This reads like <inferred domain> work — want me to record inspirations / peers / competitors / aspirational-targets? Goes in your repo's 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, proceed to Step 2 without repeating the offer.

Step 2 — Dispatch the VP-Product Reviewer as full OKR critic

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.

Step 3 — EM+PM integrate the VP-Product Reviewer's critique in-dialogue

The artifact does not exist on disk yet. REJECT items are rewritten or dropped; FLAG items are rewritten or consciously accepted with a stated rationale; weekly-perceptibility notes are rewritten or deferred to a kind: goal-seed stub. This is a dialogue: EM presents the critique, proposes a revised shape, and 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

On a PowerShell host, invoke the .exe launcher by absolute path through the call operator (Shape W) for every invocation in Steps 4-5. Ladder and shapes: snippets/resolve-coordinator-bin.md.

coordinator-doc-new --type goal --title "<objective-slug>", resolved per that ladder.

Fill: objective: (ratified text), key_results: (≤5, weekly-perceptible), period: (e.g. Q3-2026), status: active.

Step 5 — Spawn downstream stubs

5a. Roadmap-seed stubs — one per roadmap-worth-of-work (roughly: one /roadmap-planning invocation, one coherent capability arc; when in doubt, fewer/larger — the PM can split at pickup):

coordinator-doc-new --type roadmap-seed --goals "<goal-id>" --title "<roadmap-topic>", resolved per snippets/resolve-coordinator-bin.md.

Each stub carries kind: roadmap-seed, origin_goal_id: FK (via --goals), deployment_state: awaiting_gate, a one-line title naming the capability arc.

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>", resolved per snippets/resolve-coordinator-bin.md.

Each stub carries kind: goal-seed, deployment_state: awaiting_gate, and a brief body capturing the vision-slice verbatim — raw over polished. These feed the pickup-from-goal-seed entry point; they are vision preservation, not roadmap stubs.

Step 6 — Offer the PM-gated chain

"Goal artifact and {N} roadmap-seed stub(s) scaffolded. Want me to chain into /roadmap-planning now, or review the stubs first?"

Wait for PM response. Never invoke /roadmap-planning without explicit PM direction.

Step 7 — Commit goal + stubs

Stage only the goal artifact and stubs this run scaffolded — no blanket add:

Commit per snippets/scoped-commit-route.md, 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 owns it; this skill scaffolds the stubs that feed it. Auto-chaining bypasses the PM's multi-wave sequencing authority.
• KR tracking infrastructure — cockpit-contract goal.schema.json owns event emission.
• Deferred-goal fleshing without PM pickup — a kind: goal-seed stub is dormant until the PM picks it up via entry point 2.

---

Skill-scaffold checklist (self-verify before reporting DONE)

• [ ] the VP-Product Reviewer dispatched at Opus altitude via subagent_type: "coordinator:vp-product"
• [ ] PM confirmed revised OKR before any artifact written to disk
• [ ] Goal artifact scaffolded with coordinator-doc-new --type goal
• [ ] Each roadmap-seed stub carries origin_goal_id: FK and deployment_state: awaiting_gate
• [ ] /roadmap-planning was NOT auto-invoked — offered and PM-gated
• [ ] Commit is scoped to goal artifact + stubs only — no blanket add