improve-skill
Skill39 starsChanged 23 days agoMITView on GitHub
Fix a Last Mile skill that did the wrong thing, from inside the project where it went wrong. Decides whether the lesson belongs in the project's rules file or in the skill, makes the smallest change to the skill in The Last Mile repo (never the synced copy), keeps every project detail out because the repo is public, runs the checks and opens a pull request. Also adds a store or build trap other apps would hit to the ship skill's traps, through a fork when the user cannot push. Use when the user says a skill misfired, "improve the skill", "fix session-end", "that skill should have done X", or when session end lists a skill problem.
14 other projects use a version of this.
Adopted by
Most recently taken up by
- fang2hou/skillstheir own version, first seen Nov 1, 2026
- jorgenschaefer/skillstheir own version, first seen Oct 19, 2026
- razbakov/skillstheir own version, first seen Oct 17, 2026
- openai/pluginstheir own version, first seen Oct 13, 2026
- mgiovani/cc-arsenaltheir own version, first seen Sep 25, 2026
- bmaltais/skillstheir own version, first seen Sep 6, 2026
Used with
Projects with it also use create-skill.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
ElasticSkillcreate-skill
Interactively create a new Claude Code skill and add it to the elastic-docs-skills catalog. Use when the user wants to generate a new skill, scaffold a slash command, or build automation for docs tasks.
Also in 3 projects that use this
Other versions14 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 14
- openai/plugins★ 7.8ksame name
- av/skills★ 28same name
- mgiovani/cc-arsenal★ 8same name
- razbakov/skills★ 4same name
- RCmerci/skills★ 4same name
- bmaltais/skills★ 2same name
- fang2hou/skills★ 2same name
- nunomen/claude-skills★ 1same name
- a-poor/skills★ 0same name
- farzanmrz/meta-dev★ 0same name
4 more not listed
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 repository21 other files it gives its agents.
- .claude/CLAUDE.mdCLAUDE.md
- build-a-baselineSkill
- connect-toolsSkill
- extract-lessonsSkill
- fact-checkSkill
- feature-demoSkill
- lean-outSkill
- mine-communitySkill
- one-time-surfaceSkill
- port-featureSkill
- reference-huntSkill
- researchSkill
- screen-atlasSkill
- search-pagesSkill
- see-it-workSkill
- session-endSkill
- set-goalsSkill
- setupSkill
- shipSkill
- store-listingSkill
- syncSkill
The file
We split this skill into the rules it gives an agent
7 rules. 0 are common, 4 are distinctive, and 3 are very specific to this project.
0
Common
Most of the 13 edited versions we read keep these.
4
Distinctive
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 13
Write it in three lines, from the conversation:
- Distinctive: line 19
If the conversation does not make this clear, ask one question, then wait.
- Distinctive: line 38
Cloud session: add the repo to the session with push access (for example an add repository tool), clone it into its own folder, and branch …
- Distinctive: line 56
Nothing from the project goes in: no app name, screen names, copy, user data, keys, URLs, device ids or analytics.
- Very specific: line 61
Add one line under the unreleased heading in CHANGELOG.md.
- Very specific: line 65
Run node scripts/validate.mjs.
- Very specific: line 66
Commit only the files you changed, push the branch and open a pull request that follows .github/pull_request_template.md.
“Common” is measured: kept by at least 6 in 10 of the 13 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Improve skill
The skills get better from real use. This turns one thing that went wrong into one small, general fix in The Last Mile, without leaking the project it happened in.
0. What went wrong
Write it in three lines, from the conversation:
- Skill: which one.
- What it did: the step and the result.
- What it should have done.
If the conversation does not make this clear, ask one question, then wait.
1. Decide where the fix belongs
| The lesson is... | Fix it in |
| About this project only: a command, a file, a service, an app rule | The project's rules file (CLAUDE.md or AGENTS.md): the Settings table, Gotchas or Standing decisions. Stop there. |
About a skill listed in keepLocal in .claude/last-mile/config.json | That skill in .claude/skills/<name>/, which is the project's own version. Then ask whether the general version needs the same lesson; if yes, carry on below. |
| True for any app using the skill | The Last Mile, below. |
A store rejection or build trap other apps would hit, not yet in skills/ship/references/traps.md | A new trap in that file, below. |
Most misfires are a missing project setting, not a broken skill. Check that first.
2. Reach The Last Mile
The source repo is source in .claude/last-mile/config.json
[namedtoaster/the-last-mile].
- A local clone exists: work there, on a new branch from an up to date
main. - Cloud session: add the repo to the session with push access (for example an
add repository tool), clone it into its own folder, and branch from
main. - The user cannot push to it (most people): fork it with
gh repo fork namedtoaster/the-last-mile --clone, branch in the fork, and open the pull request from the fork in step 4. Withoutgh, open an issue instead: the "This caught me" form for a trap, the skill problem template for a skill.
Name the branch improve/<skill>-<few-words>.
3. Make the smallest change
Edit skills/<skill>/SKILL.md, or the reference file the step points to. For a
trap, add one entry to skills/ship/references/traps.md under the right heading, in
the shape CONTRIBUTING.md gives (What happened, Rule, How to check, Source, Status).
Mark it From experience unless you opened the official page this session.
- Change the step that went wrong. Do not rewrite the skill around it.
- Word it for any app. Swap this project's names for a plain example.
- Nothing from the project goes in: no app name, screen names, copy, user data, keys, URLs, device ids or analytics. The repo is public. If the lesson cannot be told without them, it belongs in the rules file (step 1).
- Date anything that can go stale. Follow
CONTRIBUTING.md: plain words, no long dashes (em or en). - Add one line under the unreleased heading in
CHANGELOG.md.
4. Check and open the pull request
- Run
node scripts/validate.mjs. Fix what fails until it passes. - Commit only the files you changed, push the branch and open a pull request
that follows
.github/pull_request_template.md. Say what went wrong in general terms and what the change does.
5. Finish
- Never edit the synced copy in the project's
.claude/skills/<skill>/. The next sync overwrites it. For the rest of this session, follow the fixed step by hand. - Tell the user, in a few lines: what went wrong, where the fix went, and the pull
request link. Once it is merged, the project picks it up on the next sync: the
"Sync The Last Mile skills" workflow (daily, or Run workflow from the Actions
tab), or
node .claude/last-mile/sync.mjsand a commit.
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.