new-feedback
Skill173 starsChanged 17 days agoMITView on GitHub
Log a harness lesson / post-incident so the SYSTEM improves, not just this one bug — fires on "log a harness lesson", "post-incident", "we should make this less likely", System-Evolution moments. Part of the Agentsmith harness; scaffolds a numbered docs/feedback/NNNN-*.md with the five-stage template (R9 — numbers are never reused).
In this repository11 other files it gives its agents.
- AGENTS.mdAGENTS.md
- autonomous-runSkill
- example-skillSkill
- grill-with-docsSkill
- handoffSkill
- harness-doctorSkill
- harness-helpSkill
- new-researchSkill
- verifySkill
- wayfinderSkill
- writing-rulesSkill
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
- Distinctive: line 18
Before filing or implementing a missing-item defect, confirm absence with an exact, count-based, or untruncated result.
- Very specific: line 22
Run agentsmith new-feedback "short symptom" — use .agentsmith/agentsmith on macOS/Linux or .agentsmith\\agentsmith.cmd on Windows when the …
- Very specific: line 29
Create docs/feedback/NNNN-<slug>.md (Status: open) with the five stages:
≈ “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
New feedback note (harness post-incident)
The point is not to fix THIS bug — it's to change the SYSTEM so this whole CLASS is less likely (core/60). Small, specific, traceable to the incident.
When this fires
You were corrected on something a rule could have prevented; you iterated more than you should have; a human stepped in; you re-derived a decision a past session already made. After fixing the immediate thing, capture the lesson.
Absence gate
Before filing or implementing a missing-item defect, confirm absence with an exact, count-based, or untruncated result. Truncated, paginated, filtered, or summarized search output is incomplete evidence.
Fast path — if the Agentsmith CLI is available
Run agentsmith new-feedback "short symptom" — use .agentsmith/agentsmith on macOS/Linux or
.agentsmith\\agentsmith.cmd on Windows when the command is not on PATH. It computes the next number and writes
docs/feedback/NNNN-<slug>.md with the five-stage template, Status open.
Fallback — no CLI
- Next number = (highest existing
docs/feedback/NNNN-*.md+ 1), zero-padded to 4 digits. Numbers are never reused (R9), even if an old entry was archived. - Create
docs/feedback/NNNN-<slug>.md(Status: open) with the five stages:- Evidence / symptom — what was observed, quoted.
- Failure mechanism — WHY the system allowed it (missing rule / vague instruction / absent guardrail / unreached tool / noisy context — almost never "the model was dumb").
- Bounded edit — the smallest change that prevents the whole class.
- Named surface — exactly where it lands (a core rule, a verify phase, a lint check, a
template, or a hook when the install runs hooks); prefer a deterministic surface over prose
the model can skip. Record it on the line
- **Surface:** file: <path> | text: <literal line>or- **Surface:** verify: <phase>. - Non-regression — the check that stays red on recurrence. Until it's real, Status stays open.
Report
Name the file, then land the bounded edit on its named surface and flip Status to applied.
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.

