$npx -y skills add jmstar85/oh-my-githubcopilot --skill autoresearchStateful research improvement loop with evaluator gates and decision logs. Activate when: autoresearch, research loop, improve research, evaluate research.
| 1 | # Autoresearch |
| 2 | |
| 3 | > **OMG adapted scope:** This workflow is driven by visible checkpoints and evaluator gates, not by an external OMC runtime loop. |
| 4 | |
| 5 | ## Workflow |
| 6 | 1. Define the research mission, success criteria, constraints, and stop conditions. |
| 7 | 2. Break the mission into hypotheses or research questions. |
| 8 | 3. Gather evidence from local docs, code, and permitted web sources. |
| 9 | 4. Evaluate findings against the success criteria. |
| 10 | 5. Iterate only while the next pass is likely to improve the answer meaningfully. |
| 11 | 6. Record final decisions, rejected alternatives, and confidence. |
| 12 | |
| 13 | ## Evaluator Gates |
| 14 | - Evidence sufficiency |
| 15 | - Contradiction check |
| 16 | - Confidence level |
| 17 | - Stop/continue decision |
| 18 | |
| 19 | ## Rules |
| 20 | - Prefer primary sources. |
| 21 | - Label uncertainty clearly. |
| 22 | - Do not loop indefinitely without measurable improvement. |