| name | co-think |
| description | Pre-task thinking for claude-codex-orchestration (grill / office-hours style). Source-first: inspect repo/docs before asking. Product/Design, Technical/Approach, and Domain/Docs lanes. Ask only user-owned or genuinely ambiguous questions, one at a time, with recommendation + confidence. High-verification-risk fuzzy work becomes a Run Contract before dispatch. Always ends with Premise Challenge. Optional Codex cold-read second opinion. Delegates to references/thinking-decision.md. |
/co-think — Pre-Task Thinking
Full protocol: ~/.claude/skills/claude-codex-orchestration/references/thinking-decision.md §/co-think.
Quick run
Source-first rule:
- Check named docs,
AGENTS.md, CONTEXT.md, CONTEXT-MAP.md, docs/,
ADRs, and adjacent code patterns before asking.
- Ask only user-owned decisions: target user, demand premise, public contract,
security/privacy, data model, paid infra, external service, phase split.
- Derive mechanical choices from the repo; recommend defaults for taste choices.
- Ask one question at a time, with recommendation + confidence.
Detect lane from context; ask only if unclear:
- Product/Design — new feature / API / workflow
- Technical/Approach — implementation unclear
- Domain/Docs — project vocabulary, docs, ADRs, or contracts constrain work
Product mode (5 questions, one-at-a-time):
- What's the narrowest version proving the core idea?
- Who exactly is this for, what are they doing instead today?
- What's the most likely failure mode?
- What would a senior engineer say is unnecessary?
- If you had 2 hours to demo, what would you build?
Technical mode (5 questions):
- What's the coolest version? What makes it good?
- What existing pattern gets you 50% there?
- What would you add with unlimited time? (then cut)
- Fastest path to something verifiable?
- Which assumption is most expensive if wrong?
Domain/Docs mode (4 questions):
- Which existing doc/code path is source of truth?
- Which terms/contracts must not be renamed or reinterpreted?
- Which prior decision or ADR constrains this?
- What would make this incompatible with current users or agents?
Escape hatch: "just do it" or fully formed plan → skip to Premise Challenge only.
For fuzzy + high-verification-risk work, lock a Run Contract before dispatch.
Premise Challenge (always, after questions):
PREMISES:
1. [statement] — valid / questionable
2. [assumption] — valid / questionable
3. [dependency] — valid / questionable
Optional: offer Codex cold-read (not auto-run): "Want Codex second opinion?"