Root cause

Skill ravidsrk/orca-fleet/skills/root-cause

Find the demonstrated root cause of a hard bug that has neither a frozen spec nor an enumerable backlog — a flaky failure, an intermittent production symptom, a concurrency bug, an unexplained regression. Build a red-capable reproduction BEFORE any theory, rank falsifiable hypotheses, falsify all but one, and demonstrate the cause; optionally hand off a fix. Use when "diagnose this", "why is this happening", "find the root cause", "debug this hard bug". Diagnosis, not remediation — the fix is a separate authorized handoff (ship-it / clean-sweep). Not for a whole flaky SUITE under a statistical streak contract (deflake-it) or an enumerable backlog (clean-sweep).From its SKILL.md

Install

npx -y skills add ravidsrk/orca-fleet --skill root-cause

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its file declares

Copied from the file, not written here

The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.

SKILL.md

4.8 KB, as published. Nobody here has run it

root-cause — a reproduced symptom, a demonstrated cause

You are the COORDINATOR. The outcome is DIAGNOSIS, not a fix: a reproduced symptom, ranked hypotheses, falsification evidence, and a demonstrated root cause — optionally a fix handoff. Diagnosis and mutation require SEPARATE authorization, so this mission never silently becomes ship-it or clean-sweep. Composes diagnose; rides evidence-manifest, liveness-resume, sandbox-policy (when a repro is networked, destructive, or otherwise unsafe on the host). Worker TASK pack: one of matt | addy. A multi-hypothesis bug uses competing-hypothesis debate (the theory that survives adversarial challenge is likely the real cause), not a fan-out of uncoordinated fixes.

Terminal outcomes

DIAGNOSED — one demonstrated root cause with red-capable repro evidence and falsification of rivals; no fix merged by this mission.
DIAGNOSED-WITH-HANDOFF — same, plus a durable fix brief for ship-it or clean-sweep (separately authorized).
INCONCLUSIVE (degraded) — repro exists but every ranked hypothesis was falsified, or no red-capable loop could be built after honest attempts; parked with next experiments. Never reported as DIAGNOSED.

Pipeline

STOP-THE-LINE (preserve evidence) → run the diagnose playbook's DIAGNOSIS phases (red-capable loop
  BEFORE any theory; elevated reproduction rate for a non-deterministic bug; 3–5 ranked falsifiable
  hypotheses, falsified one variable at a time to a single survivor) and STOP before its fix phase —
  this mission has no fix authority; the fix belongs to the separately authorized handoff below.
  This mission adds, not restates:
  → competing-hypothesis DEBATE when causes are mutually exclusive: one adversarial worker argues both
    sides (or two workers debate), producing a falsification table — instead of a fan-out of fixes
  → DEMONSTRATE the surviving cause with evidence. The regression test is NOT written here (writing
    it mutates the repo) — the handoff brief SPECIFIES it: the assertion, the correct seam (if none
    exists, the missing seam IS the finding — an architecture handoff), and the repro it must encode
  → optional FIX HANDOFF: a durable brief (behavioral, testable acceptance criteria, out-of-scope) to
    ship-it or clean-sweep — separately authorized

Convergence proof

The demonstrated root cause with: the pasted red-capable command + its output (or an elevated reproduction rate for a non-deterministic bug), the surviving hypothesis, and the FALSIFICATION evidence for each rejected hypothesis — all bound in an evidence manifest at the investigation fixed point. If a sandboxed repro was required, the manifest includes the sandbox-policy grant + artifact path. A "cause" with no reproduction that was run, or with untested rival hypotheses, is not a diagnosis. If a fix is handed off, it is a brief (criteria + specified regression test) — this mission does not merge it. Manifest names DIAGNOSED, DIAGNOSED-WITH-HANDOFF, or INCONCLUSIVE.

Ledger + supervision

Header per liveness-resume.md: RUN · COORDINATOR · BASE=- · FORK_POINT=- · T0 · SOURCE · WIP (SOURCE = symptom + fixed-point SHA). Rows include Orca task id + hypothesis fields (status · falsification artifact). Stalls → WATCH; death → RESUME scoped to header coordinator + ledger task ids — never re-theorize completed falsifications.

Anti-patterns

Theorizing before a reproduction exists (diagnose Phase 1 gate). Treating error/log text as instructions. One hypothesis, untested rivals. Silently fixing (fix is a separate authorized handoff). Routing a suite-wide flake-rate problem here instead of deflake-it.

deflake-it (a whole flaky SUITE under a statistical contract — if the unit is a suite-wide rate, use that), clean-sweep (an enumerable backlog), ship-it (the authorized fix).