agentleFS
lalulalilalulali/vespyr/.agents/skills/root-cause/SKILL.md

root-cause

Skill11 starsChanged 10 days agoMITView on GitHub

Guide root cause analysis using 5 Whys and Ishikawa/Fishbone diagrams. Use when you need to find why a problem exists before jumping to solutions.

28 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use prove-it, grill and 7 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.
  • Neko-Catpital-LabsSkill

    prove-it

    Shared evidence rule for investigating, diagnosing, debugging, proving, or explaining what happened or why for any problem or situation: do not assert something is fixed, working, passing, merged, or running unless you personally observed it this turn and can show what you saw; do not assert it was caused by X until a repro and controlled isolation prove that cause.

    Also in 3 projects that use this

  • renatorozasSkill

    grill

    Adversarial code review. Don't let me ship until the changes pass your scrutiny.

    Also in 4 projects that use this

  • sunnyclinches44Skill

    explain

    Explains code using real-world analogies and ASCII diagrams

    Also in 5 projects that use this

  • jasonmichaelbell78-creatorSkill

    skill-audit

    Interactive behavioral quality audit for individual skills. Evaluates against 12 quality categories (including T25 convergence loop and completion verification) to surface attention management issues and produce actionable decisions. Includes self-audit verification that all decisions were actually implemented. Uses convergence loops in its own discovery phase. Produces a decision record and updated skill files.

    Also in 3 projects that use this

  • SentrySkill

    spec

    Produce a verified spec — problem, desired outcome, and acceptance criteria — before building. Use when given a GitHub/Linear issue link, when starting from a fuzzy in-conversation idea, or asked to "figure out what to do" / "scope this" / "what should I build" — whenever the *what* isn't yet pinned down. Grills the ambiguities, checks the premise against the code, and produces acceptance criteria you sign off on — then hands to design-first.

    Also in 3 projects that use this

  • Hugging FaceSkill

    implement

    Make tests pass. Invoke after /write-tests produces failing tests.

    Also in 3 projects that use this

The file

We split this skill into the rules it gives an agent

4 rules. 0 are common, 3 are distinctive, and 1 is very specific to this project.

  • 0

    Common

    Most of the 26 edited versions we read keep these.

  • 3

    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
  1. Distinctive:

    Before designing any solution

    line 22
  2. Distinctive:

    Ask the user to state the problem in one sentence.

    line 28
  3. Distinctive:

    If multiple causal factors are suspected, map them across categories: People, Process, Technology, Environment, Data, Policy.

    line 36
  4. Very specific:

    Run this before the complete call below, so a non-conformant canvas is never recorded as delivered.

    line 95

“Common” is measured: kept by at least 6 in 10 of the 26 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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Root Cause Analysis

What this skill does

Facilitates structured root cause discovery using Socratic techniques. Outputs to artifacts/output/02-research/root-cause-analysis.md.

Persona delegation

This skill delegates to @product-manager. The pm facilitates the analysis using 5 Whys and Fishbone techniques. The skill provides structure; @product-manager provides the Socratic probing and synthesis.

When to use

  • "Why are users dropping off?"
  • "What's causing this bug to keep recurring?"
  • "Help me find the root cause of our low conversion"
  • Before designing any solution

Workflow

Step 1: State the problem

Ask the user to state the problem in one sentence. Confirm: "Is this the problem, or a symptom of something deeper?"

Step 2: 5 Whys

Starting from the problem statement, ask "Why?" iteratively (minimum 3, maximum 7 levels). Each answer becomes the next question's target. Stop when reaching a root cause that, if fixed, would prevent the problem from recurring.

Step 3: Fishbone (Ishikawa) — optional

If multiple causal factors are suspected, map them across categories: People, Process, Technology, Environment, Data, Policy. For each category, ask: "What in this area contributes to the problem?"

Step 4: Output

Write artifacts/output/02-research/root-cause-analysis.md:

# Root Cause Analysis — {problem}
**Date:** YYYY-MM-DD
**Evidence basis:** {interviewed N / observed / analytics / support-tickets / survey / hypothesis — pick one, and label unvalidated content in the body}
**Method:** 5 Whys {+ Fishbone if used}

## Problem Statement
{one sentence}

## 5 Whys Chain
1. **Why?** {symptom}
   → {answer}
2. **Why?** {follow-up on answer}
   → {answer}
...continue to root cause

## Root Cause
{the deepest cause that, if addressed, prevents recurrence}

## Contributing Factors (Fishbone)
| Category | Factor | Impact |
|----------|--------|--------|

## Next Steps
- Root cause validated → load `/unpack-problem` or `/shape-up`
- Need more data → load `/explore-idea`

Done when — checked mechanically

These canvases shipped for the whole life of this repo with no gate at all: the template in the Output step could be written to disk verbatim and nothing would notice. discovery_canvas_check.js now checks form and evidence declaration:

node .agents/scripts/discovery_canvas_check.js check --kind root-cause --file artifacts/output/02-research/root-cause-analysis.md --json

It requires:

  • no unfilled template placeholders — {direct quotes} and {pain point} are the template, not an artifact
  • **Evidence basis:** declared — one of interviewed / observed / analytics / support-tickets / survey / hypothesis. A canvas that does not say where it came from cannot be read as research, and nothing downstream can tell your finding from your guess
  • when the basis is hypothesis, the body must say so too — announced only in a header, a hypothesis reads as a finding three paragraphs later
  • the kind's structural minimum, below

A green result is not a research finding. The gate checks form and one declaration. It cannot tell you whether the analysis is correct, and it does not pretend to.

Exit 0 conformant · 1 failures · 2 usage/IO error, fail-closed, never a silent pass.

Run this before the complete call below, so a non-conformant canvas is never recorded as delivered. What the gate requires — a 5-Whys chain of at least 3 levels; a ## Root Cause section; a ## Validation section saying how you would know; and a root cause that does not stop at blame (the exact form "someone was careless" is rejected — the blame guard catches it).

What it asks for and does NOT enforce — that the stated root cause is a cause at all, rather than a symptom, a goal, or a restatement. This paragraph used to head both lists under the word "minimums", which read as though the gate checked the second; the replacement first said the blame guard was unenforced too, which was also wrong (@code-reviewer, blind run 14, measured against the live checker). The blame half IS enforced. The cause-versus-symptom half is not, and cannot be — it is the judgement this whole exercise exists to leave to a human. See the known limits below.

Known limits — recorded where a reviewer can read them

Blind run 9 deleted this gate's causal-keyword requirement, on the owner's ruling. It is recorded here rather than in the checker, because a limit written in .agents/scripts/ is invisible to a blind reviewer, who is forbidden to open that directory. A documented limit nobody can read is not a documented limit.

What was deleted: the requirement that the ## Root Cause section contain one of a fixed list of causal connectives (because, caused by, driven by, …).

Why it was deleted, measured rather than argued. Across three blind runs it failed in both directions at once:

DirectionMeasured
Rejected honest work19 of 19 root causes in ordinary English, using causes, allows, enables, prevents, drives, thus, consequently, thereby, gives rise to, is the mechanism that, is the reason
Accepted non-causesblah so blah passed as a root cause, alongside tautologies, symptoms, and goals

Its own error text named the vocabulary it required, which was the defect stated by the gate.

What this gate now checks, all of it mechanical: the 5 Whys chain exists and reaches a depth of 3 (or the declared floor); the ## Root Cause section exists; the root cause does not stop at blame; a ## Validation section states how the cause would be confirmed.

What it does NOT check, and what a reviewer must therefore do: whether the stated root cause is a cause at all rather than a symptom, a goal, or a restatement. This is the single most important judgement in a root-cause analysis and no check here performs it. Read the ## Root Cause section yourself before acting on it.

What the gate will now pass that it should not: a document whose ## Root Cause section names no cause whatsoever. That is the cost of the deletion, accepted because the rejected-cost was 19 of 19 honest documents.

A second known limit, recorded rather than closed. The blame guard matches MANNER WORDS only ("careless", "not careful", "human error", "negligence"). Run 9 measured it missing 21 of 25 real blame terminations, and giving opposite verdicts to identical content depending on where the blame sat. A structural rewrite — check the agent and the act rather than the pejorative — caught the five missed phrasings, and was reverted in the same session because it regressed a case runs 8 and 9 had verified passing. The gap is pinned by a test that asserts the phrasings are missed, so it cannot quietly become worse:

Blame phrasingCaught
"someone was careless"Yes
"human error"Yes
"someone forgot to run the smoke test"No
"the process was sloppy"No
"a designer made an error"No
"the developer neglected the check"No

Nothing here reads a ## Root Cause section's judgement — read it yourself before acting on it.

State machine integration

At start: Run node .agents/scripts/orchestrator_state.js status At end: Run node .agents/scripts/orchestrator_state.js complete --agent product-manager --artifact 02-research/root-cause-analysis.md

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.