# copilot-researcher agent operating guide You are working in a copilot-researcher repository. Treat GitHub Copilot CLI as the orchestration harness and the copilot-researcher ledger as the durable source of truth for research work. ## Durable research workflow 1. If the user wants persistent research tracking, call `researchflow_get_workspace_status` before doing major work. 2. If no workspace exists and durable tracking is requested, call `researchflow_init_workspace`. 3. Use `researchflow_list_templates` before creating a new session and pick the template that best matches the task. 4. Persist meaningful work with the durable ledger tools: - hypotheses via `researchflow_add_hypothesis` - supporting or contradicting findings via `researchflow_add_evidence` - datasets, notes, scripts, and reports via `researchflow_add_artifact` - rankings via `researchflow_score_session` - final markdown via `researchflow_generate_report` ## Scientific rigor - Prefer evidence-backed claims over speculation. - Capture both supporting and contradicting evidence whenever available. - Keep uncertainty explicit; do not present unresolved questions as facts. - When recommending next steps, bias toward falsifiable experiments and reproducible artifacts. ## Copilot CLI orchestration - Use skills for repeatable methods and custom agents for specialist roles. - Prefer `/fleet` after plan mode when the work breaks cleanly into independent threads such as literature review, contradiction search, experiment design, and report drafting. - Use autopilot only for bounded, well-specified tasks where autonomous continuation is safe. - When editing TypeScript, preserve the local-first ledger behavior and do not remove existing `copilot-researcher` CLI commands unless explicitly asked.