oduffy-delphi/shape

Shape

Converge with the PM on what the problem actually is before the EM goes solo on how to solve it. Plans fuse a PM-owned PRD half (the problem, what "solved" means) with an EM-owned SDD half (architecture, fix-locus, sequencing); the pipeline has rigor for the SDD half and almost none for the PRD half. /shape is that missing front half: a short, collaborative problem-convergence beat whose only exit is a ratified problem-set that altitude-routes to the horizon-appropriate downstream ceremony (coordinator:plan, coordinator:roadmap-planning, or coordinator:goal-setting).

Strategic ceremony, not tactical clarification. Operates at the altitude of product intent, scope boundaries, and success criteria — questions only the PM can resolve. Tactical uncertainties (naming, file structure, test framework, commit shape, refactor mechanics) are EM remit and do NOT belong in a /shape exchange, however genuinely unsure the EM is — resolving them is the EM's job (global CLAUDE.md § PM Altitude).

Announce at start: "Using /shape to converge on the problem before we plan a solution."

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.

When NOT to invoke

• Trivial work (single-file change, obvious scope) → just do it.
• PM doesn't know what to build / needs approaches explored → coordinator:brainstorming. Discriminator: PM has a problem and wants confirmation you understood it → /shape; PM doesn't know what to build at all → brainstorming.
• Problem already converged (a prior /shape, an existing approved spec, a known next-step on a live workstream) → straight to the downstream skill (§ Transition).

/shape and coordinator:brainstorming are siblings: brainstorming turns vague intent into a design spec (docs/specs/); /shape converges on the problem (docs/problems/) and defers solutioning entirely. Both chain into a downstream planning ceremony.

Conform intake — an incoming sizing-object (route: shape)

coordinator:sizing sometimes routes here rather than straight to plan/roadmap-planning: (a) large size with unclear JTBD, or (b) a well-trodden problem/solution space where the ask wants a step-change. If the firing condition is known from the sizing exchange, name it; otherwise state both as possible. A large-but-clear ask skips shape entirely — this is not a universal sizing route. Shape keeps its own § Transition horizon-routing regardless — sizing routes INTO shape, it does not perform shape's downstream routing. This is a receive-and-use-if-present detent: /shape runs exactly as without a sizing-object, and sizing never gates or refuses a bare /shape invocation.

<HARD-GATE> Once `/shape` has started, do NOT invoke any implementation skill, write code, scaffold anything, or dispatch an executor until the problem-set is written and PM-ratified. The only exit is a ratified problem-set that transitions via § Transition. If the problem was already converged, skip `/shape` entirely instead. Once started, see it through. </HARD-GATE>

Process (lightweight — not brainstorming's full design dialogue)

1. Restate the problem(s) as received, in the PM's vocabulary, falsifiably — the actual problem, stated so the PM can catch a misread. With a route: shape sizing-object present, open from its intent field verbatim rather than re-deriving the raw ask.
2. Surface your single biggest uncertainty per the forced-articulation contract below — the load-bearing step.
3. PM corrects/confirms. Iterate to a shared mental model; the written problem-set is its residue, not its substitute.
4. If the PM has a proposed solution-shape, reflect it back and flag pushback — do NOT design. Solutioning is coordinator:plan's job.
5. On convergence, scaffold the problem-set (below), fill its body, set estimated_horizon, ratify with the PM, confirm the horizon, and chain into the downstream ceremony (§ Transition).

Forced-articulation contract

A yes/no "do I have the shape? ✓" is a banned response shape — EM confidence is coupled with helpfulness, so it self-reports green every time. Step 2 must instead produce the thing that reveals a misunderstanding:

• (a) The single least-certain item must be the scope boundary whose wrong guess costs the most rework — "name the boundary that, if guessed wrong, invalidates the most of the plan", not "name something I'm unsure about."
• (b) State the probability-weighted consequence: "If I'm wrong about X, then Y and Z are rework." A low-stakes selection self-evidently fails this.
• (c) Must be a PM-altitude question — product intent, scope boundary, success criteria. Tactical/EM-decided uncertainties are disqualified as off-altitude regardless of how unsure you are (global CLAUDE.md § PM Altitude) — resolving them is your job, surfacing them is noise.
• Also flag any intent you inferred that the PM did not state — scope, priority, constraint.

The problem-set artifact

Scaffold via (POSIX-host form; PowerShell hosts: Shape W, snippets/resolve-coordinator-bin.md): coordinator-doc-new --type problem-set --title "<problem title>" — emits docs/problems/YYYY-MM-DD-<slug>.md, status: draft, canonical skeleton. Tiny by design: a bulleted problem list, not a design doc.

Fill and ratify:

1. List problems under ## Problems (numbered, NOT prioritized — no P<n> prefix).
2. List non-goals under ## Out of scope (architectural reasons) with hard architectural reasons, not "later."
3. On PM convergence: flip status: draft → ratified, fill ratified_by, ratified_date, estimated_horizon, stamp > Ratified by PM <name> <date>.

<!-- engine-gap: field=shape.convergence.detected producer=unknown memo=2026-08-14-doe-claude-em-three-cut-obligations-from-the-corpus-grind.md -->

A problem-set without ratification is status: draft and does NOT count as an oracle. Plans link it via problem_set:.

Transition

Once ratified, route by horizon — detect-then-confirm: surface your horizon read at ratification, let the PM confirm before chaining. Never pick a rung silently.

| estimated_horizon | Work shape | Downstream ceremony | |---|---|---| | session | One or two sessions | coordinator:plan — problem-set becomes the plan's coverage oracle (plan-coverage-checker). | | week | ~week, bounded | coordinator:roadmap-planning — seeds a roadmap with sprint-sized stubs. | | initiative | Weeks-to-quarters | coordinator:goal-setting — seeds an OKR-shaped goal artifact. May fan out 1→N goal-seed stubs when the vision decomposes into multiple goal-slices. |

Precedence: a PM axiom/directive naming a specific downstream skill, or a forced-articulation tier flag that resolves the horizon unambiguously, overrides the router — the router is the default for ambiguous or undirected convergence.

Ambiguous horizon: ask explicitly — "Does this feel like a session, a week, or an initiative?" Never pick silently.