publish-skill
Skill0 starsChanged 7 days agoMITView on GitHub
Maintainer workflow for publishing an Agent Skill to the lothlorienblade/agent-skills collection on GitHub. Brings a new or changed skill up to the collection's publication criteria, wires it into the repository (docs, trigger evals, README, plugin marketplace, changelog, versions), validates it, pushes it, tags a collection release, publishes the GitHub Release, and updates the local installs. Use whenever the user wants to publish, upload, release, or add a skill to their skills repository or to agent-skills, to update a published skill there, or to check whether a skill is ready for publication, even if they only say "залей скилл в репозиторий" or "опубликуй скилл". Do not use for creating or improving a skill's own content when publication is not the goal, or for publishing to other repositories.
- Installs packages
13 other projects use a version of this.
Adopted by
Most recently taken up by
- dvy1987/agent-loomtheir own version, first seen Nov 15, 2026
- mqmalagris/agent-skillstheir own version, first seen Nov 5, 2026
- djfwilkinson/skillstheir own version, first seen Aug 26, 2026
- Kislonfreak/publish-skilltheir own version, first seen Aug 4, 2026
- skilld-dev/skillstheir own version, first seen Aug 4, 2026
- minilittlechanges-ai/skillstheir own version, first seen Jul 14, 2026
Other versions13 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 13
- glebis/claude-skills★ 414same name
- alexanderop/walkthrough★ 130same name
- dvy1987/agent-loom★ 4same name
- drewdrewthis/claude-skills★ 2same name
- skilld-dev/skills★ 2same name
- h-mascot/Enterprise-Crew-skills★ 1same name
- Kislonfreak/publish-skill★ 1same name
- MaxLiuyy/Claude-Code-Skills★ 1same name
- minilittlechanges-ai/skills★ 1same name
- djfwilkinson/skills★ 0same name
3 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 repository4 other files it gives its agents.
- codebase-auditSkill
- codex-imagegenSkill
- modular-architecture-refactorSkill
- text-distillSkill
The file
We split this skill into the rules it gives an agent
8 rules. 0 are common, 7 are distinctive, and 1 is very specific to this project.
0
Common
Most of the 12 edited versions we read keep these.
7
Distinctive
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
- Very specific: line 25
If a clone exists, git pull --ff-only and check git status is clean.
- Distinctive: line 32
Generalization: Remove the author's personal details: paths, user names, project names, personal model or tool preferences.
- Distinctive: line 34
Agent-neutral wording: No tool names of one agent ("call SendUserFile").
- Distinctive: line 37
If the draft is installed locally under a different form (another language, another script), the repository version replaces it at step 6.
- Distinctive: line 56
Run the same checks as CI, in a virtual environment outside the clone:
- Distinctive: line 71
Commit as the repository author, with the message in a file.
- Distinctive: line 73
Wait for both CI runs (branch and tag) to succeed.
- Distinctive: line 75
Verify the release through the public API.
“Common” is measured: kept by at least 6 in 10 of the 12 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
Publish a skill to lothlorienblade/agent-skills
Take a skill from "it works for me" to a published, released, and installed part of the collection at https://github.com/lothlorienblade/agent-skills. The repository is public and every skill in it is installed into other people's projects, so the bar is the collection's criteria, not the draft's.
The repository's own CONTRIBUTING.md, CHANGELOG.md, and README.md are authoritative. Read them at the start of every publication. If they disagree with this skill, follow the repository and update this skill in the same change.
1. Get a clean clone
Clone the repository into a temporary or scratch folder, never inside the user's project, and work only there:
git clone https://github.com/lothlorienblade/agent-skills.git <scratch>/agent-skillsIf a clone exists, git pull --ff-only and check git status is clean. Read CONTRIBUTING.md, CHANGELOG.md, README.md, .claude-plugin/marketplace.json, and one published skill with its docs page and evals: the new skill must look like its neighbours.
2. Bring the skill to publishable form
Go through every item in references/criteria.md. The items that most often need work on a draft:
- Language. Skills, docs, and evals are in English. Trigger evals also include Russian prompts.
- Generalization. Remove the author's personal details: paths, user names, project names, personal model or tool preferences. Turn preferences into documented defaults that a flag or environment variable overrides.
- Portability. Scripts are standard-library Python, cross-platform, with distinct exit codes. Not PowerShell or Bash only.
- Agent-neutral wording. No tool names of one agent ("call SendUserFile"); say what to do ("show the result to the user").
- Honest claims. Something observed once is stated as observed, not as a rule. What was not tested is not presented as working.
If the draft is installed locally under a different form (another language, another script), the repository version replaces it at step 6.
3. Wire it into the repository
For a new skill <name> (details and formats in references/criteria.md):
skills/<name>/withSKILL.md,agents/openai.yaml, andreferences/orscripts/as needed.evals/<name>/trigger-cases.json.docs/skills/<name>.md.- A row in the README Skills table and an install line in the plugin marketplace section; widen the README intro and
marketplace.jsontop-level description if the collection's scope grew. - A plugin entry in
.claude-plugin/marketplace.json. - A
CHANGELOG.mdentry under the new collection version, and the original prompt indocs/original-prompts/<name>.<lang>.mdif the skill grew from one.
For a change to a published skill: bump metadata.version, the marketplace version, and add a changelog entry; if the description changed, add a trigger case for what prompted the change.
Choose versions with semantic versioning: a new skill starts at 1.0.0. The collection release is minor for a new skill or capability, patch for fixes and wording, major for behaviour users rely on. Changes to repository tooling only (CI, maintainer skills like this one) go under ## Unreleased and ship with the next release.
4. Validate
Run the same checks as CI, in a virtual environment outside the clone:
python -m venv <scratch>/venv
<scratch>/venv/bin/pip install skills-ref==0.1.1 # Scripts\ on Windows
for skill in skills/*/ .agents/skills/*/; do <scratch>/venv/bin/agentskills validate "$skill"; done
python scripts/check_repo.pyExercise every bundled script on its success path and on each error path you can trigger without spending the user's money or quota. Delete __pycache__ folders and other run artifacts, then check git status shows only intended files. Report anything you could not test.
5. Publish and release
Pushing to a public repository and publishing a release are outward-facing. Do it when the user asked to publish; otherwise show the summary and ask first. Follow references/publishing.md exactly; it records the pitfalls of this environment. In short:
- Commit as the repository author, with the message in a file. Verify the commit exists before anything else.
- Push
main, then tagvX.Y.Zon that commit and push the tag. - Wait for both CI runs (branch and tag) to succeed.
- Publish the GitHub Release
vX.Y.Z — <headline>with notes in the established format, using scripts/create_release.py. It reads the token Git stores for github.com, which needs the user's explicit permission in chat. - Verify the release through the public API.
6. Update the local installs
Reinstall the published version for Claude Code and Codex, replacing any hand-made local copy, and check that the installed files match the repository. Commands are in references/publishing.md.
7. Report
Tell the user, in their language: what was published (skill, version, collection release), links to the commit, release, and docs page, the CI result, what was installed where, anything that went wrong and how it was fixed, and what remains untested.
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.

