agentsclimarketplace

Identify root cause

Skill qa-veritas/skillpack/skills/identify_root_cause

Agent capability that loads only when the task needs it. The Skills layer of QA Veritas.

Install
npx -y skills add qa-veritas/skillpack --skill identify_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

  • 0 stars0 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 author says it does

Copied from the file, not written here

From all collected findings, pick the single most-likely root cause with a confidence level, the supporting evidence, and the cheapest test that would falsify it. Use as the synthesis step at the end of a triage.

SKILL.md

2.7 KB, as published. Nobody here has run it

identify_root_cause

Purpose

The synthesis step. After the plan, the log summary, the failure explanation, and any live-state checks, this skill commits to one root cause — with a confidence level and, critically, the test that would prove it wrong. A root cause you cannot falsify is a guess.

Inputs

  • findings_state — the accumulated structured findings (log summary, failure explanation, cluster health, leak verdict, prior-incident matches). In practice this is a state.json-style document with typed sections.

Outputs

root_cause: >-
  A data node crossed the storage flood-stage watermark; the cluster
  flipped indices to read-only; the test's writes were rejected. The
  test is correct; the environment ran out of disk.
cause_class: INFRA
confidence: high            # low | medium | high
supporting_evidence:
  - "First error is a write rejection, not a test assertion."
  - "Disk on node-2 at 94% at the failure timestamp."
  - "All downstream errors are timeout/retry cascade."
falsifying_test: >-
  If disk on every data node was below the watermark at 10:02:48, this
  cause is wrong — re-check node-2's disk timeline at that timestamp.
ruled_out:
  - "TEST-BUG: the assertion logic is unchanged and passed yesterday."

Prompt template

You are committing to a single root cause. Input: {findings_state}.

  1. Weigh the hypotheses. Prefer the one that explains the FIRST error and the cascade, not just the loudest symptom.
  2. Assign a confidence level and justify it from evidence, not vibes.
  3. Write the cheapest test that would falsify your conclusion. If you cannot write one, your confidence is at most medium.
  4. List what you ruled out and why — this is as valuable as the cause.
  5. If evidence is insufficient, say so and name the one missing piece that would decide it, rather than forcing a conclusion. Output the structured result.

Examples

  • Write rejection + disk at 94%. → INFRA, high confidence, falsifiable by the disk timeline.
  • Assertion failure on unordered data. → TEST-BUG, high confidence; falsifying test is whether the system ever guaranteed order.

Composition

The terminal node of the triage chain. Feeds create_verification_strategy so the fix is verified against the cause it actually identified, and feeds the journal entry in an operational-memory system.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.