Investigation engine

The investigation layer (spec-my-A §25) answers: what broke, where, and why — before an implementation agent runs.

Pipeline

symptom / intent
  → scope resolution (flow, task, failed tests)
  → local discovery (grep, symbols, tests)
  → static analysis
  → flow / contract correlation
  → hypotheses (scored)
  → context pack + root-cause graph
  → targeted agent work

Outputs

Investigations produce versionable artefacts under .asagiri/investigations/<id>/ so humans and agents share the same evidence.

Integration points

• asa work --investigate-first
• asa verify --investigate-on-failure
• Runtime memory (investigation and hypothesis entries)

Related

• investigate CLI
• Executable Product Layer