skill-harvest
Skill0 starsChanged 4 months agoMITView on GitHub
Use when the user asks to harvest skill learnings, improve the skills from recent builds, or process skill notes — "/skill-harvest", "update the skills from what we learned", "process the skill notes from <project>". Runs in its own short session, never inside a build.
2 other projects use a version of this.
Adopted by
Most recently taken up by
- vindreshsingh/engineering-skillstheir own version, first seen Oct 20, 2026
- GerardoRdz96/claude-skillstheir own version, first seen Jul 13, 2026
Other versions2 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 2
- GerardoRdz96/claude-skills★ 0same name
- vindreshsingh/engineering-skills★ 0same 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.
- product-blueprintSkill
- saas-builderSkill
- saas-maintainerSkill
The file
We split this skill into the rules it gives an agent
6 rules. About 0 are common, 4 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
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 22
Do not read build transcripts, whole repos, or unrelated references.
- Very specific: line 54
Keep descriptions trigger-only.
- Very specific: line 58
Delete processed notes from docs/skill-notes.md so nothing is double-harvested.
- Distinctive: line 63
Never run inside a build session.
- Distinctive: line 64
Never batch-create rules for hypothetical failures — one field observation, one fix.
- Distinctive: line 66
If two harvests in a row touch the same section, the section is wrong at a deeper level — redesign it instead of patching a third time.
≈ “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
Skill Harvest
Turn field friction into skill commits at minimum token cost. Capture was
already done during builds (one-line entries in each project's
docs/skill-notes.md); this skill batches the analysis.
Inputs — read ONLY these
docs/skill-notes.mdfrom the project(s) the user names.- The specific skill sections the notes implicate — not whole skill files unless the edit requires surrounding context.
Do not read build transcripts, whole repos, or unrelated references. If a note is too vague to act on, ask the user one question or drop it — do not excavate.
Notes are symptom reports, not verdicts
They were usually written by a fast model mid-build. Re-derive every diagnosis yourself from the skill text and the reported event; ignore a note's suggested fix unless it survives your own triage. Signal ranking: a verbatim user correction is the strongest evidence there is; "an instruction couldn't execute as written" is strong; an agent's self-diagnosis is the weakest. Run the harvest on the strongest available model — this is a thinking phase.
Triage each note with the improvisation filter
- Real gap — a decision got improvised mid-build because no file answered it → fix the owning file with the smallest edit that closes it.
- Binding failure — the guidance existed but didn't fire → don't add words; anchor the rule to an executable moment (a numbered step, a checkpoint message, a wave boundary), and add a user gate if the decision is taste or irreversible. Prose that isn't a step gets skipped.
- Routing failure — the wrong skill ran → sharpen the trigger descriptions or add an explicit handoff rule.
- One-off / project-specific / "a serious company would have this" → drop it. Note why in the commit message only if it was borderline.
- User preference, not a skill gap → save it to persistent memory instead of editing a skill.
Apply
- One owner per topic — edit the file that owns it; other files point.
- Keep descriptions trigger-only; keep SKILL.md bodies lean (details go to references); keep every rule proportionate to the risk it manages.
- One commit per theme, with the field evidence in the message.
- Sync the private skills repo to the public one and push both.
- Delete processed notes from
docs/skill-notes.mdso nothing is double-harvested.
Rules
- Never run inside a build session; harvest context stays small on purpose.
- Never batch-create rules for hypothetical failures — one field observation, one fix.
- If two harvests in a row touch the same section, the section is wrong at a deeper level — redesign it instead of patching a third time.
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.