Idea Validator
Use this skill before implementation when the user has an incomplete idea and needs a rigorous path from vague concept to grounded decision.
Workflow
1. Clarify The Idea
Interview the user one question at a time until the idea has a stable brief. Use references/interview-rubric.md.
Do not research or judge the idea too early. First clarify:
- problem
- target user
- proposed solution
- why now
- success criteria
- constraints
- unknowns
Stop this phase when you can summarize the idea in one paragraph and the user agrees or corrects it.
2. Confirm Research Mode
After the user confirms the stable brief, ask one focused mode question before research:
Do you want full multi-agent research fan-out for this idea? If yes, I will open five independent research-lane subagents, which Codex may show as visible child/forked conversations: existing work, technical feasibility, user/market value, risks and counterexamples, and implementation path. If no, I will run the same lanes sequentially in the main agent.
Do not proceed to research until the user chooses a mode. Do not treat a request for depth, rigor, or thoroughness as permission to spawn subagents.
If the user chooses full multi-agent mode but the environment does not support subagents or subagent use is unsafe, say so and continue with sequential single-agent research unless the user stops.
3. Research Fan-Out
When the user chooses full multi-agent mode and the environment supports subagents, spawn exactly one subagent for each research lane. Give each subagent a narrow, independent lens and ask for sources, evidence, caveats, and unresolved questions.
Treat those subagents as evidence collectors, not final decision makers. The main agent owns synthesis, skeptic loop, and final conclusion.
Recommended lanes:
- existing products, projects, papers, or competitors
- technical feasibility
- user value and market value
- risks, failure cases, and counterexamples
- implementation paths and architecture patterns
If the user chooses single-agent mode, does not approve subagents, subagents are unavailable, or web access is unavailable, do the same lanes sequentially and disclose the limitation.
Do not synthesize or start the skeptic loop until all five research lanes have returned or been completed sequentially.
Use references/research-rubric.md. Require each lane to return the structured lane output defined there.
4. Synthesize
The main agent synthesizes the five research lanes. Do not delegate final synthesis to a subagent.
Merge research results into:
- facts
- assumptions
- inferences
- open questions
- promising angles
- weak points
Keep source-backed evidence separate from your own reasoning.
5. Skeptic Loop
Immediately and automatically run a skeptic pass after synthesis using references/skeptic-rubric.md. Do not ask the user whether to start the skeptic loop; it is a required part of this skill.
Run the skeptic loop in the main agent by default. Do not spawn an extra skeptic subagent unless the user has explicitly authorized an additional skeptic subagent, separate from the five research-lane subagents.
If the skeptic identifies a specific evidence gap that could materially change the final verdict, run targeted re-research for only the affected lane. Do not rerun all five research lanes unless multiple verdict-changing gaps span all lanes.
Loop up to 5 times:
- skeptic identifies the strongest objections
- main agent answers or modifies the idea
- skeptic checks whether major objections remain
Stop when there are no unresolved major objections, or after 5 rounds. Do not try to remove every minor concern.
6. Final Conclusion
Only produce the final conclusion after clarification, research lanes, main-agent synthesis, and the skeptic loop are complete.
Use references/final-report-structure.md.
Return:
- feasibility verdict
- value verdict
- differentiation
- modified idea
- MVP
- implementation path
- validation experiments
- major risks
- remaining unknowns
- recommended next step
Be willing to say the idea is not worth doing, or worth doing only after modification.
Output Rules
- Ask clarifying questions before research when the idea is underspecified.
- Do not spawn subagents until the user confirms the stable brief and explicitly chooses multi-agent research fan-out.
- Always run the skeptic loop automatically after synthesis, regardless of whether research used subagents or sequential single-agent lanes.
- Use multiple independent research lanes for serious ideas.
- Cite sources when web research is used.
- Separate evidence from inference.
- Do not present consensus if sources disagree.
- Prefer a useful modified idea over a polite approval.