hansei
Skill11 starsChanged 2 months agoMITView on GitHub
Reflect on failures, incidents, reviews, and completed software work to turn evidence into durable tests, guardrails, documentation, and process improvements.
In this repository13 other files it gives its agents.
- andonSkill
- genchi-genbutsuSkill
- jidokaSkill
- kaizenSkill
- kansoSkill
- kataSkill
- kodawariSkill
- monozukuri-blueprintSkill
- monozukuri-coreSkill
- monozukuri-routerSkill
- nemawashiSkill
- poka-yokeSkill
- shukkaSkill
The file
We split this skill into the rules it gives an agent
10 rules. About 0 are common, 10 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈10
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 36
Use logs, traces, tests, diffs, deployments, timelines, and direct reports.
- Distinctive: line 42
Ask how the defect could be introduced, remain undetected, reach users, and persist.
- Distinctive: line 54
Prefer a small number of specific actions that change the system: a regression test, invariant, validation rule, alert, runbook, decision …
- Distinctive: line 65
Use a timeline and concrete artifacts where possible.
- Distinctive: line 70
Do not use reflection to identify scapegoats.
- Distinctive: line 71
Do not create a long report with no owner or completion condition.
- Distinctive: line 72
Do not stop at “be more careful” when a mechanical guardrail is possible.
- Distinctive: line 73
Do not reopen unrelated design debates unless the evidence connects them to the failure.
- Distinctive: line 75
Do not treat a completed meeting as a completed improvement.
- Distinctive: line 83
If a referenced skill is not installed, apply its named lens inline instead of trying to invoke it.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Hansei
Hansei is disciplined reflection. In software, examine what happened, why it happened, and how the system allowed it so the lesson becomes a durable improvement. This is a Japanese-inspired engineering metaphor, practiced without blame or ritual for its own sake.
Use this skill when
- an incident, outage, regression, or escaped defect occurred;
- a review found a recurring quality problem;
- a release or project completed and important lessons should be preserved;
- a workaround has become normal behavior; or
- the team wants to improve its engineering system rather than only patch code.
Outcome
Produce a concise reflection with:
- the observed impact and timeline;
- contributing technical, process, and communication conditions;
- what made detection, diagnosis, or recovery easier or harder;
- what worked and should be preserved;
- actionable improvements with owners or clear scope; and
- a follow-up test, guardrail, document, signal, or process change.
Workflow
1. Establish the facts
Use logs, traces, tests, diffs, deployments, timelines, and direct reports. Separate what is known from what is inferred. Do not begin with blame or a preferred narrative.
2. Trace the system's contribution
Ask how the defect could be introduced, remain undetected, reach users, and persist. Look at missing tests, weak boundaries, confusing ownership, unsafe defaults, poor observability, rushed handoffs, and misleading documentation.
3. Preserve the good
Record the checks, habits, design choices, and response actions that reduced impact or sped recovery. Reflection should reinforce effective practice, not only list failure.
4. Choose durable actions
Prefer a small number of specific actions that change the system: a regression test, invariant, validation rule, alert, runbook, decision record, ownership boundary, or workflow improvement. Assign a clear completion condition.
5. Verify the lesson
Confirm that each important action would have prevented, detected, explained, or reduced the original problem. Remove vague actions that cannot be checked.
Evidence standard
Use a timeline and concrete artifacts where possible. Do not claim a root cause that the evidence cannot support. State uncertainty and residual risk plainly.
Boundaries
- Do not use reflection to identify scapegoats.
- Do not create a long report with no owner or completion condition.
- Do not stop at “be more careful” when a mechanical guardrail is possible.
- Do not reopen unrelated design debates unless the evidence connects them to the failure.
- Do not treat a completed meeting as a completed improvement.
Handoff
Hand off concrete actions to poka-yoke, andon, kaizen, or shukka as
appropriate. Revisit the reflection after actions land and verify that the
original failure mode is now harder to repeat.
If a referenced skill is not installed, apply its named lens inline instead of trying to invoke 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.

