| name | coe-investigation |
| description | Evidence-driven investigation workflow for turning vague issues into tracked cases with structured evidence chains. |
COE Investigation
Use this skill when a task should be handled as an evidence-driven investigation instead of an immediate fix.
Goals
- Turn a vague issue into a tracked investigation case.
- Record evidence before proposing or applying repairs.
- Use guardrails to decide whether the case is ready for patching or closure.
- Produce reviewer handoffs that preserve the evidence chain.
Workflow
- Read
investigation://profile to confirm the surface and capabilities.
- Open a case with
investigation.case.open.
- Keep the canonical root problem current with
investigation.problem.update, investigation.problem.set_status, and reference material when needed.
- Capture reusable evidence with
investigation.evidence.capture or investigation.evidence.capture_and_attach, then attach or revise evidence references through investigation.evidence.attach_existing and investigation.evidence_ref.update.
- Create hypotheses for unresolved explanations, open blockers when evidence or access is missing, and only create repair attempts after a hypothesis is confirmed.
- Run guardrails:
investigation.guardrail.check
investigation.guardrail.stall_check
investigation.guardrail.ready_to_patch_check
investigation.guardrail.close_case_check
- Use
investigation://cases/{caseId}/snapshot, investigation://cases/{caseId}/timeline, investigation://cases/{caseId}/graph, investigation://cases/{caseId}/evidence-pool, and investigation://cases/{caseId}/diff to summarize progress.
- Escalate reviewer-only actions instead of faking them in an agent session.
Rules
- Do not skip evidence collection.
- Do not write free-form conclusions when a structured
problem, hypothesis, blocker, repair_attempt, evidence, or evidence_ref object exists.
- Treat missing evidence, blocked access, and unresolved contradictions as structured blockers, not comments.
- Only advance to repair preparation when guardrails say the evidence chain is sufficient.