omp-flow: Skill for Claude Code

OMP-Flow Brainstorm

Preconditions

• The user consented to task creation or selected an existing task.
• The active task is a readable Bundle with a root index.
• The assignment identifies the Bundle root, bounded objective, entry Concept, and allowed output.
• The entry normally points to brainstorm.md or a linked framing Concept.

First-Principles Orientation

For non-trivial Explore, form or revise the Router's first-principles anchor（第一性锚定） in authored prose before broad Research. Name the concrete actor, action, outcome, human boundary, principal contradiction（主要矛盾）, and evidence that could change the framing. Mechanical, low-ambiguity work may proceed directly.

Concentrate interaction on the material decision frontier:

• For purpose, values, risk tolerance, and non-negotiable outcomes, the human states a position first; then give the strongest counter-case, counterexample, and consequence.
• For evidence-led technical choices, an Agent may lead with a recommendation only when it also gives the strongest counter-case and a falsifier: the evidence that would overturn it.
• Use targeted high-intensity Grill only while a question can change the problem core or outcome. Do not exhaust branches, prescribe a question count, or let the Agent silently rank human values.

Interview Contract

1. Read the entry Concept and follow only useful links from the Bundle.
2. Record the initial request, concrete actors/action/outcome, current principal contradiction, and known repository facts in the assigned framing Concept.
3. Read the material the request points to, then code, tests, docs, prior tasks, relevant history, and applicable durable knowledge through the native omp-flow-wiki Skill, before asking anything they already answer.
4. Ask the user one material decision question at a time. Ask only about purpose, value, risk tolerance, preference, or use that repository and other relevant practice evidence cannot establish, and say what the answer would change. When the request already says what is wanted, play it back with what you found and ask only about the one open point whose answer would most change what gets built.
5. Keep confirmed facts, assumptions, disagreements, alternatives, constraints, success criteria, and open questions distinct.
6. Present materially different directions only when a real choice exists. Follow the human-first and recommendation/counter-case order above instead of anchoring value choices with a model default. Clarify early which technical or medium choices the human wants to own versus delegate; mark any agent-selected defaults or guesses explicitly as agent's call / provisional assumption in playback and in brainstorm.md rather than blending them into human-confirmed requirements.
7. Update the linked framing Concept as decisions change. Chat is not the durable source.
8. When a question needs evidence, link a bounded research Concept or assignment and state what anchor assumption, contradiction, or practical decision the result may change. When research confirms, revises, or falsifies the framing, update the authored anchor and next question. This is a practice test（实践检验）spiral, not a one-way gate.