tooling-feedback-triage
Skill2 starsChanged 4 months agoMITView on GitHub
Review and act on accumulated feedback reports about agents, skills, and tools — the field reports left by the tooling-feedback skill under .agent-feedback/. Finds the reports, investigates each against the agent, skill, or tool it names, drives it to a resolution, and cleans up handled reports. Defers the actual remedy to the repo's own conventions or to the user — it does not invent fixes. Trigger on "triage feedback", "review agent feedback", "process tooling feedback", "handle the field reports", or a periodic feedback sweep.
In this repository3 other files it gives its agents.
- review-lensesSkill
- technical-editorSkill
- tooling-feedbackSkill
The file
We split this skill into the rules it gives an agent
4 rules. About 0 are common, 3 are distinctive, and 1 is 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.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 24
If that directory is absent or empty,: check whether the repo keeps feedback elsewhere — a single tracked file, or a path named in …
- Very specific: line 34
Read it: If it carries a session id and you're on the machine that produced it, you may open that transcript for fuller context.
- Distinctive: line 59
Only delete reports you've actually closed out — leave unresolved ones in place.
- Distinctive: line 61
Keep this proportionate — triage is a periodic sweep, not a standing process.
≈ “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
Tooling feedback triage
Acts on the field reports recorded by the tooling-feedback skill. Each report is about an agent, a skill, or a tool that fell short — not just tool usage. That skill only records; this one processes: it gathers the outstanding reports, works out which are real, gets each to a resolution, and clears it away.
What it deliberately does not do is decide the remedy on its own. The right response to a report — edit an agent or skill file, open an issue, adjust config, update docs, or judge it not worth acting on — depends entirely on the project. This skill's job is to surface, validate, and route; the fix comes from the repo's conventions or from the user.
Find the reports
- Default location:
.agent-feedback/, a directory of per-report markdown files (the format tooling-feedback writes). Each has frontmatter (subject— the agent, skill, or tool the report is about — pluskind,date,branch, and an optionalsession) and a body describing what went wrong. - If that directory is absent or empty, check whether the repo keeps feedback elsewhere —
a single tracked file, or a path named in
CLAUDE.md/AGENTS.md/ contributing docs. A quick search for "feedback" across those files usually settles it. - If you still can't find any, say so and stop — there's nothing to triage. Don't manufacture work.
Process each report
For every report found:
- Read it. If it carries a
sessionid and you're on the machine that produced it, you may open that transcript for fuller context; otherwise work from the report alone — it's written to stand without the session. - Validate against the named subject. Open the agent, skill, tool, or config the report names and check the issue is real and still present — not already fixed, not a one-off misread. Reports come from observed failures, so weight them over hypothetical concerns, but confirm before acting.
- Decide the route — using the repo or the user, not your own invention.
- First look for an established convention: does the project document how feedback is resolved (a fix-the-file pattern for the agent/skill, an issue tracker, a changelog)? If so, follow it.
- Absent a convention, present your analysis to the user — the report, whether you think it's valid, and a recommended action — and let them direct. Don't autonomously rewrite tooling or file issues unless the repo's pattern or the user clearly calls for it.
- Carry out the agreed action.
Work newest-first, or group by subject if several reports point at the same agent, skill, or
tool — fixing it once can resolve a cluster.
Clean up
Once a report has been resolved — fixed, filed, or deliberately dismissed — delete the file. git history preserves the record, so there's no need to archive it. Note what was done (or why it was dismissed) in the commit message that carries the deletion.
Only delete reports you've actually closed out — leave unresolved ones in place.
Keep this proportionate — triage is a periodic sweep, not a standing process. On a small or solo project, a short pass that fixes the clear ones, routes the unclear ones to the user, and tidies up afterwards is the whole job.
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.

