Harness Kit

skill

shape

Shape a raw idea into something buildable. Product + technical exploration. Spec, design, critique, plan. Output is a context packet. Use when: "shape this", "write a spec", "design this feature", "plan this", "spec out", "context packet", "technical design". Trigger: /shape, /spec, /plan, /cp.

/shape

Source: skills/shape/SKILL.md

What it does

Shape a raw idea into something buildable. Product + technical exploration. Spec, design, critique, plan. Output is a context packet. Use when: "shape this", "write a spec", "design this feature", "plan this", "spec out", "context packet", "technical design". Trigger: /shape, /spec, /plan, /cp.

Workflow role

Plan primitive in the Harness Kit operating loop.

Source contract preview

This generated excerpt gives readers the beginning of the live primitive contract before they jump to GitHub.

Shape a raw idea into something buildable. Output is a **context packet** —
the unit of specification that precedes implementation.

## Workflow

Size the shape before raising the floor. Trivial shapes are mechanical
one-file fixes, typo-level docs, or already-decided packet formatting.
Non-trivial/M+ shapes are everything with product judgment, architecture
choice, user workflow, data model, dependency, privacy, or rollout risk. The
diversity, roster, matrix, and review-bench gates apply to non-trivial/M+
shapes.

### Phase 1: Understand

Accept: raw idea, backlog.d/ item, issue ID, or observation.

## Delegation Floor

Delegation floor applies: probe the roster first; dispatch two or more
providers for substantive work; direct solo only for mechanical, emergency,
user-forbidden, or fewer-than-two-providers cases. See
`harnesses/shared/AGENTS.md` (Roster).

Local lane guidance: Use one lane to map repo constraints and another for prior-art or premise challenge; native in-thread subagents may add signal but do not satisfy the roster floor.

Spawn parallel sub-agents to gather context fast: one to map the relevant
codebase area (files, patterns, constraints), another to search for prior art
(how do other projects solve this? check codebase first, then /research).
For non-trivial shapes, fill the bench instead of running a two-agent
formality: include at least one repo investigator, one product/premise critic,
one architecture critic, one implementation-risk reviewer, and one test/oracle
reviewer when the available harness can support it. Use the lead model for
synthesis; record meaningful external lanes as delegation receipts.
Synthesize their findings before proceeding.

If `exemplars.md` exists at project root, read it. Include relevant exemplar
techniques in the context packet with specific files to study during build.
...

What to verify