bug-blame
Skill4 starsChanged 6 months agoApache-2.0View on GitHub
Trace a bug or incident back through captured history to the change that caused it, the prompt behind that change, and its blast radius. Use when something broke and you need to know what changed and why.
In this repository14 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- blameSkill
- logSkill
- showSkill
- blameSkill
- context-primerSkill
- file-couplingSkill
- logSkill
- regent-skillsSkill
- rewindSkill
- showSkill
- skill-factorySkill
- style-factorySkill
The file
We split this skill into the rules it gives an agent
3 rules. About 0 are common, 1 is distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 15
If given a path or path:line, start there:
- Very specific: line 36
Read the user message on that step.
- Distinctive: line 41
Check the assistant message too.
≈ “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
Answer four questions about a defect, in order: what changed, when, why, and what else moved with it.
git blame answers the first two. Only re_gent answers the third — the prompt that
caused the edit — and the fourth is what turns a fix into a safe fix.
1. What changed
If given a path or path:line, start there:
rgt blame <path>[:<line>]Each line carries the step hash that last wrote it. Take the step for the suspect
line. If given only a symptom, find candidate files first with
rgt log --json -n 50 and match against the described behaviour.
2. When, and in what company
rgt show <step-hash>This gives the full step: timestamp, tool calls with their arguments and results, every file the turn touched, and the surrounding conversation.
3. Why — the prompt
Read the user message on that step. This is the instruction the agent was
following. State it verbatim. Very often the bug is not a coding mistake but a
faithful execution of an ambiguous or wrong instruction, and that distinction
changes the fix: rewrite the code, or rewrite the expectation.
Check the assistant message too. If the agent explained a tradeoff or flagged an assumption, the defect may be a known compromise rather than an accident.
4. Blast radius
The step's files list is the immediate radius — everything that turn changed
under one intention. Widen it:
- Other steps in the same session, before and after (
rgt log --session <id>), since a multi-turn task spreads one intention across several steps. - Files that habitually travel with the suspect file — invoke the file-coupling skill. A file coupled to the culprit but not touched in this step is the highest-value thing to check: it usually should have changed and did not.
Report
Symptom <what the user described>
Cause step <hash> · <timestamp> · <tool>
Prompt "<the user instruction, verbatim>"
Reasoning <what the agent said it was doing, if relevant>
Changed <files in that step>
Also check <coupled files that did NOT change — likely gaps>
Session <session id>, step N of MFinish with a recommended next action: fix forward, or rgt rewind <step> to
restore the workspace to before the change. Name the rewind target explicitly and
say that rewind snapshots the current state first, so it is reversible.
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.

