log-mistake
Skill0 starsChanged 56 days agoMITView on GitHub
Write a logbook entry when something breaks, works unusually well, or a hard-to-reverse choice gets made. Use the moment it happens, not at the end of the session.
2 other projects use a version of this.
Adopted by
Most recently taken up by
- sheldor26/Blastdoorword for word, first seen Oct 8, 2026
- drewdrewthis/claude-skillstheir own version, first seen Apr 22, 2026
Other versions2 projects use a version of this.
Word for word · 1
- sheldor26/Blastdoor★ 0100% same
Edited · 0
None found.
Wrote their own version · 1
- drewdrewthis/claude-skills★ 2same name
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 repository3 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- close-sessionSkill
- recallSkill
The file
We split this skill into the rules it gives an agent
3 rules. About 0 are common, 3 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
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 60
Keep asking "and why was that allowed" until the answer is a property of the system rather than of the person or the model.
- Distinctive: line 76
If no guardrail is possible, say that explicitly and say why — an honest "unpreventable, detectable only by reading the output, so the …
- Distinctive: line 88
Check what already exists before inventing a tag:
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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
Log it now
The entry gets written when the thing happens, not at the end of the session. At the end the details are gone, and it is the first task to get cut when the session is running long. Four seconds now, or nothing ever.
node .bitacora/cli.mjs new mistake "Specific title" --tags area,failure-mode --severity high
node .bitacora/cli.mjs new learning "Specific title" --tags area
node .bitacora/cli.mjs new decision "Specific title" --tags areaThe command assigns the id and scaffolds the sections. You write the prose.
If a subagent found it, log it in the main session
A subagent runs in its own context window. It follows the main CLAUDE.md, so
it can read this skill and run recall — but when it returns, it returns a
summary, and everything specific it saw is gone with its context.
That matters more than it sounds, because subagents get the exploratory and verification work: reading unfamiliar code, checking whether something is actually broken. That is exactly where mistakes are discovered.
So when a subagent reports something that broke or surprised it, the main session writes the entry, and writes it before moving on — the detail is in the summary you are holding right now, and nowhere else. If the summary is too thin to support a real Guardrail, ask the subagent for the specifics before its findings go stale, rather than filing a vague entry.
Which kind
- Mistake — anything broke, surprised us, or had to be fixed twice. A near-miss counts. So does something that worked but only after three wrong attempts: the three wrong attempts are the entry.
- Learning — something worked unusually well and would change how the next problem is approached. If it would not, it is a changelog line.
- Decision — a choice that would be expensive to reverse, or that a future reader might quietly undo because the reasoning is invisible.
Writing a mistake
What happened. Concrete, specific, no blame. Enough detail that the shape is recognisable when it starts happening again. Name the actual values, files and commands — "the import overwrote 11 of 15 manually corrected prices" beats "data got overwritten".
Root cause. Why it was possible, not what broke. This is the distinction that determines whether the entry is worth anything:
"The importer overwrote the value" — that is what broke. "Nothing distinguished a verified value from a scraped one, so newer always won" — that is why it was possible.
Keep asking "and why was that allowed" until the answer is a property of the system rather than of the person or the model.
Guardrail. The check, test, type or refusal that makes the same failure
impossible rather than merely known. doctor rejects the entry if this section
is missing or near-empty, because this is the part that does the work.
A guardrail is not a lesson. It is code, config, or a test:
| Not a guardrail | A guardrail |
| "Be more careful with prices" | A priceVerifiedAt field the importer refuses to overwrite, checked in CI |
| "Remember the API is paginated" | A type that makes the un-paginated call unrepresentable |
| "Always run the build first" | A pre-commit hook, or a Stop hook that reports it |
| "Read the script header next time" | The script validates its input and exits with the reason |
If no guardrail is possible, say that explicitly and say why — an honest "unpreventable, detectable only by reading the output, so the output is now printed" is a real entry. Silence in that section is not.
Tagging
Two to four tags: one for the area, one for the failure mode. Good tags name a
thing in the system (pricing, auth, migrations, deploy) or a failure
mode (data-loss, race-condition, silent-failure, type-hole). Bad tags
describe the entry (important, bug, fixed) and match everything or
nothing.
Check what already exists before inventing a tag:
node .bitacora/cli.mjs statsReusing an existing tag is what makes recall work later, and what makes a
recurring failure mode visible as a count rather than as five unrelated
entries.
Finish it
node .bitacora/cli.mjs doctorGreen before moving on. An entry left half-written is how the logbook dies.
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.

