harishkotra Autonomous Research Solver
Autonomously research a hard, measurable problem and return a verified solution.
When to use
- Someone hands you a hard problem and wants it solved, not just described.
- The problem is measurable: there is a correct answer or a testable outcome.
Procedure
- Restate the problem as a falsifiable question with a success criterion.
- Break it into sub-questions; plan which tools (web search, code exec, math) answer each.
- Iterate: gather evidence, attempt a solution, run the verification, refine. Loop until the success criterion passes.
- Cite sources for every factual claim.
- Return: the answer, the method, the evidence, and what remains uncertain.
Pitfalls
- Don't stop at a plausible-sounding answer; always run the verification step.
- If the problem is not measurable, say so and ask for a success criterion before deep work.
Verification
- The final answer must pass the stated success criterion, with evidence you can point to.