root-cause-analysis
Skill0 starsChanged 57 days agoMITView on GitHub
Root-cause a bug whose symptom may mislead: intermittent, flaky, or already "fixed" once and back. Produces the causal chain with evidence and a solution proposal, not the implementation.
18 other projects use a version of this.
Adopted by
Most recently taken up by
- microsoft/hve-coretheir own version, first seen Nov 15, 2026
- nledford/engineering-review-boardtheir own version, first seen Nov 3, 2026
- azoghry-commits/Cognition-OStheir own version, first seen Nov 1, 2026
- Robert-Adunka/triz-skillstheir own version, first seen Nov 1, 2026
- KMaruthi2002/universal-llm-skillstheir own version, first seen Oct 25, 2026
- aictrl-dev/skillstheir own version, first seen Oct 22, 2026
Used with
Projects with it also use observability, adversarial-review and 6 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
OpenAISkillobservability
Vercel Observability expert guidance — Drains (logs, traces, speed insights, web analytics), Web Analytics, Speed Insights, runtime logs, custom events, OpenTelemetry integration, and monitoring dashboards. Use when instrumenting, debugging, or optimizing application performance and user experience on Vercel.
Also in 3 projects that use this
MicrosoftSkilladversarial-review
Adversarial review of a system design from 6 critical perspectives -- SRE, security, staff engineer, finance, operator, and developer advocate. Produces a unified risk assessment. Use for INTERACTIVE on-demand reviews during a design conversation (/adversarial-review). For RECIPE-DRIVEN reviews (where prior step context is needed), use the systems-design-critic agent instead.
Also in 3 projects that use this
GoogleSkilldomain-modeling
Build and sharpen a project's domain model. Use when the user wants to pin down domain terminology or a ubiquitous language, record an architectural decision, or when another skill needs to maintain the domain model.
Also in 4 projects that use this
GoogleSkillimprove-codebase-architecture
Scan a codebase for deepening opportunities, present them as a visual HTML report, then grill through whichever one you pick.
Also in 3 projects that use this
MicrosoftSkillhandoff
Compact the current conversation into a handoff document for another agent to pick up.
Also in 3 projects that use this
- +3 more from google
Other versions18 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 18
- microsoft/hve-core★ 1.6ksame name
- itseffi/agentic-os★ 114same name
- microsoft/cat-agent-skills★ 108same name
- Robert-Adunka/triz-skills★ 17same name
- nledford/engineering-review-board★ 11same name
- microsoft/aibast-agents-library★ 8same name
- viacheslav-tronko/claude-code-harness★ 6same name
- aictrl-dev/skills★ 4same name
- saitarrun/Devforge-ai★ 4same name
- azoghry-commits/Cognition-OS★ 3same name
8 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository7 other files it gives its agents.
- AGENTS.mdAGENTS.md
- implement-with-solSkill
- architecture-proposal-reviewSkill
- implement-with-opusSkill
- new-featureSkill
- pre-implementation-analysisSkill
- worktree-workflowSkill
The file
We split this skill into the rules it gives an agent
7 rules. 0 are common, 6 are distinctive, and 1 is very specific to this project.
0
Common
Most of the 17 edited versions we read keep these.
6
Distinctive
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 18
Use heddle/resources/decision-routing.md for ownership and question quality.
- Distinctive: line 26
Use the governing spec and applicable decisions for expected behavior.
- Distinctive: line 36
Do not invent a reproduction or treat an old successful run as current evidence.
- Distinctive: line 40
Follow the failing execution and state transitions far enough to identify the condition that produces the symptom.
- Distinctive: line 42
Use causal tracing, fault trees, or difference analysis as appropriate.
- Distinctive: line 53
Do not force an unchangeable upstream outage into a fictional locally fixable cause.
- Distinctive: line 96
Use the requested report format.
“Common” is measured: kept by at least 6 in 10 of the 17 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Root Cause Analysis
Explain the reported failure through a supported causal mechanism and propose the change that would remove it. Distinguish a verified mechanism from proof that it caused this particular incident. If the incident cannot be established, report the remaining uncertainty and the evidence needed to resolve it.
Decision ownership
Use heddle/resources/decision-routing.md for ownership and question quality;
feature kickoff delivers the resolved host policy in full. Preserve this skill's
analysis/authoring boundary and applicable standing authorization. A review or
diagnosis request alone grants no implementation authority.
Scope and Authority
Investigate the reported symptom, relevant failure paths, and plausible causes. Use the governing spec and applicable decisions for expected behavior; use source, tests, logs, and reproducible observations for actual behavior. Read-only inspection and relevant hermetic diagnostic checks are in scope. A diagnosis request alone does not authorize implementation, permanent tests, live/paid reproduction, production mutation, or unrelated cleanup. Honor additional authorization already supplied for the task.
State expected versus actual behavior, affected inputs/environment, timing, scope, and reproducibility from available evidence. Ask for missing information only when it changes the investigation; finish independent checks while a question is open. Do not invent a reproduction or treat an old successful run as current evidence.
Causal Evidence
Follow the failing execution and state transitions far enough to identify the condition that produces the symptom. Compare working and failing cases when useful. Use causal tracing, fault trees, or difference analysis as appropriate; there is no required number of whys, branches, or hypotheses.
Evaluate plausible alternatives against both supporting and contrary evidence. Record an untested alternative as untested, not disproven. An absence of contrary evidence is not itself confirmation. Correlated timing identifies a lead; the mechanism and observations establish causation.
A supported root cause explains the observed manifestation conditions and identifies the actionable defect, violated contract, or external condition. Explain how the proposal removes the mechanism or prevents its recurrence under those conditions. Do not force an unchangeable upstream outage into a fictional locally fixable cause. Distinguish the cause from a mitigation, detection improvement, or recovery measure. Reverting a fix can restore a symptom for both a causal fix and a workaround, so reversion alone does not distinguish them.
For example, queue delay is a symptom when a trace shows another worker holding a shared connection throughout a large transaction. Reducing transaction scope is a causal proposal only if the transaction's atomicity is unnecessary under the actual contract. Priority scheduling may mitigate delay without removing connection starvation. Both proposals need their own evidence and trade-offs.
Reuse applicable investigation evidence. Recheck it when a relevant source, environment, input, or contradictory observation changes. Distinguish inspection, reproduction, and inference; a passing suite does not disprove an untested incident.
Named traps
- Premature termination: stopping at a plausible label without explaining the mechanism or the observed conditions. Trace the decision-relevant links; mark an inaccessible link uncertain instead of extending the chain by guesswork.
- Single-thread pursuit: accepting an early hypothesis without testing plausible alternatives. State what observations distinguish them and which checks were possible. Missing contrary evidence is neither confirmation nor proof that contrary evidence must exist.
- Symptom-fix conflation: presenting mitigation, detection, or recovery as removal of the cause. Explain which mechanism the proposal changes and under what conditions recurrence is prevented. An external cause can be established even when only a local mitigation is available.
Delegation
Delegate distinct hypotheses or independent path traces when that adds evidence
or saves time while the caller continues useful work. Intermittency or a prior
failed fix alone does not require parallel agents. Follow
docs/workflow/subagent-exploration-patterns.md for bounded briefs, model tier,
current tool capabilities, and evidence synthesis. A path trace is exploration
work; an investigation whose product is weighing contradictory evidence or
adjudicating a prior failed fix is the named case for the top tier. Give each
investigator the same observed problem and require supporting evidence, contrary
evidence, and unresolved limits.
Deliverable and Completion
Use the requested report format; otherwise provide:
- Finding: confirmed cause or leading explanation, confidence, and incident-specific uncertainty, followed by expected/actual behavior and conditions.
- Evidence: causal chain with source locations and observations, hypotheses tested, contrary evidence addressed, and relevant checks with their actual results.
- Proposal: recommended change and its causal connection, alternatives when materially different, affected contracts, regression risks, and required authority.
- Verification plan: the smallest meaningful check that distinguishes the diagnosis from its alternatives and proves the proposed behavior if implemented.
The investigation is complete when the evidence supports the cause and proposal, or when available in-scope checks are exhausted and the remaining evidence gap is explicit. An unsupported causal link stays uncertain. End with the diagnosis and next decision; do not implement the fix unless the user authorizes it.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

