root-cause-analysis

root-cause-analysis

Field	Value
Kind	skill
Source	.github/skills/root-cause-analysis/root-cause-analysis
Invocation	Invoked directly as /root-cause-analysis, or loaded on demand by referencing agents
Interactive	No

What it does

Investigate incidents, software defects, and data or process failures by forming falsifiable hypotheses, executing tests against available evidence, and iterating to a verified causal explanation or an explicit evidence blocker. Use when performing root cause analysis, incident investigation, or postmortem analysis, with cost and model choice confirmed before a full investigation.

When to use it

Use this skill to investigate an incident, software defect, or data or process failure when the cause is not yet established. It is appropriate when evidence can be gathered or supplied, competing explanations need to be tested, and the result must distinguish a verified cause from a provisional or blocked investigation.

Use the incident-response skill instead when you need the broader operational workflow of triage, diagnosis, mitigation, communication, and post-incident documentation. Use an ordinary debugging workflow for a localized defect whose cause is already reproducible and does not require competing hypotheses or evidence-source analysis.

Results, including a Complete status, are AI-assessed and need validation by a qualified human reviewer before adoption. Do not use the output as the sole basis for regulatory incident reports, safety determinations, legal conclusions, or evaluating an individual's performance or conduct.

Install and govern the skill with the bundled references/setup-and-governance.md guidance. The statistical-hypothesis-testing and causal-evidence-likelihood companion skills provide bounded assessments but do not set RCA dispositions. They are internal-only capabilities that fail closed without an invocation context from this root workflow.

For a full real investigation, the skill first explains anticipated AI-credit and monetary costs and asks whether to keep the selected model, switch to the lower-cost gpt-5.6-luna model when available, or pause. Estimates depend on evidence volume, investigation length, and current host pricing; they are not guaranteed incident prices. Model selection and cost consent do not replace evidence-access approvals.

Example usage

Use the root-cause-analysis skill to investigate why checkout requests began timing out after
14:20 UTC. Compare affected and unaffected instances, examine the deployed change history, and
test at least one competing explanation. Keep the investigation read-only and return a resumable
checkpoint if the completion gate cannot pass.

The skill returns the investigation status, source coverage, hypotheses, executed tests, evidence, causal account, completion-gate assessment, cause-linked actions, and a resume checkpoint when the result is not complete.