investigation-loop

Installation
SKILL.md

Default output: return only the result, blockers, and required evidence. Omit preambles, process narration, repeated context, confidence scores, and follow-up offers. Use at most five bullets unless a required artifact or schema needs more.

Investigation Loop

Long-form investigation skill for tickets where the answer requires walking multiple linked tickets, code paths, and Confluence pages. Hand the agent a ticket and a marker file; the cooking Stop hook keeps it iterating until either (a) a written hypothesis with at least three evidence citations exists, or (b) the iteration counter hits 10.

When to Use

  • A Jira (or similar) ticket whose root cause is non-obvious and requires reading code, related tickets, and historical decisions
  • The investigation will take 30+ minutes and you don't want to babysit it
  • You need a written hypothesis at the end, not just a fix — investigation is the deliverable

When NOT to Use

  • The fix is obvious — just do it; don't loop
  • The work is mechanical (test fixing, refactor) — use test-fix-loop or cross-repo-refactor
  • You don't have access to the systems involved (Jira, repo, etc.) — the loop will spin without progress

How to Run

Installs
1
GitHub Stars
8
First Seen
Aug 17, 2026
investigation-loop — v1truv1us/ai-eng-system