upsert-skill
Skill4 starsChanged 6 months agoMITView on GitHub
Manages the lifecycle of reusable agent skills stored under .agents/skills/. Use when the user wants to add, modify, rename, delete, or audit skills in this repository — also triggered by "create a new skill", "set up a skill", "list my skills", "what skills do I have", or "remove a skill". DO NOT TRIGGER when: the user asks to review, evaluate, or critique a skill for quality (use review-skill instead); or asks for a full library health report, trigger conflict analysis, or workflow coverage audit (use audit-skills instead).
In this repository13 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- audit-skillsSkill
- create-commitSkill
- format-review-commentsSkill
- implement-planSkill
- refine-proseSkill
- review-agents-mdSkill
- review-planSkill
- review-prSkill
- review-skillSkill
- upsert-agents-mdSkill
- upsert-planSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
14 rules. About 0 are common, 2 are distinctive, and 12 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
12
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 28
Quality: use an action+object formula (search-orders, extract-pdf-text).
- Very specific: line 29
Namespace: if the repository already has 10 or more skills, add a shared prefix for related skills (e.g.
- Very specific: line 31
Create the directory: .agents/skills/<skill-name>/
- Distinctive: line 35
Write a clear Markdown body with instructions agents can follow directly
- Very specific: line 36
Verify plugin self-containment — re-read the instructions imagining no access to this repository's AGENTS.md.
- Very specific: line 37
Add scripts/, references/, or assets/ subdirectories only for content too large to inline in SKILL.md (large examples, schemas, scripts, …
- Very specific: line 44
Read the existing SKILL.md before making changes
- Very specific: line 45
Keep the name field in sync with the directory name — never rename one without the other
- Very specific: line 46
Update metadata.version when the instructions change meaningfully
- Very specific: line 47
If touching the description, review the natural-language trigger phrases — ensure the Use when...
- Very specific: line 50
Commit: feat(<skill-name>): or fix(<skill-name>): or docs(<skill-name>): depending on the nature of the change, followed by a lowercase …
- Very specific: line 55
Remove the entire .agents/skills/<skill-name>/ directory
- Distinctive: line 71
If the audit is read-only and no issues are found, no commit is needed
- Very specific: line 72
If audit findings require corrections, apply fixes and commit: fix(<skill-name>): <description of issue corrected>
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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
Upsert Skill
Manages the lifecycle of reusable skills stored under .agents/skills/.
Instructions
Creating a skill
- Choose a name — it must pass all of the following checks:
- Format: lowercase letters, digits, and hyphens only; 1–64 chars; no leading, trailing, or consecutive hyphens; must be unique under
.agents/skills/ - Quality: use an action+object formula (
search-orders,extract-pdf-text); avoid single-word or noun-only names; name the outcome, not the implementation; avoid reserved words (anthropic,claude) - Namespace: if the repository already has 10 or more skills, add a shared prefix for related skills (e.g.
orders-search,orders-refund) - For the full naming specification and worked example, read
references/skill-format.md
- Format: lowercase letters, digits, and hyphens only; 1–64 chars; no leading, trailing, or consecutive hyphens; must be unique under
- Create the directory:
.agents/skills/<skill-name>/ - Create
SKILL.md— required frontmatter fields:name(must match directory),description(what it does and when to use it); use the "Does X. Use when Y." template — start with the action verb in third person, then state the triggering condition; for full specification and worked example, readreferences/skill-format.md - Document natural-language trigger phrases in the
description— include the phrases users are likely to say (e.g. "create a commit", "review this PR", "make a plan"). Use the inlineUse when...form for simple skills; add explicitTRIGGER when:/DO NOT TRIGGER when:lines when the skill overlaps with others and disambiguation is needed. See the Natural Language Triggers section inreferences/skill-format.mdfor both patterns and concrete examples. - Omit
allowed-toolsunless the user explicitly requests tool restrictions - Write a clear Markdown body with instructions agents can follow directly
- Verify plugin self-containment — re-read the instructions imagining no access to this repository's
AGENTS.md. Any guidance the skill needs must be in theSKILL.mditself or in areferences/file shipped with the skill. - Add
scripts/,references/, orassets/subdirectories only for content too large to inline inSKILL.md(large examples, schemas, scripts, reference docs). Seereferences/skill-format.mdfor the full progressive disclosure model. - Apply the
refine-proseskill to thedescriptionfrontmatter field and theSKILL.mdbody — run it silently before saving - Review
AGENTS.md— update if the new skill affects documented structure, conventions, or workflow - Commit:
feat(<skill-name>): add <skill-name> skill — <one-line summary>
Updating a skill
- Read the existing
SKILL.mdbefore making changes - Keep the
namefield in sync with the directory name — never rename one without the other - Update
metadata.versionwhen the instructions change meaningfully - If touching the
description, review the natural-language trigger phrases — ensure theUse when...condition and anyTRIGGER when:/DO NOT TRIGGER when:lines reflect the updated skill scope. See the Natural Language Triggers section inreferences/skill-format.mdfor guidance. - Apply the
refine-proseskill to anydescriptionor body text that is rewritten or newly authored — run it silently before saving. - Review
AGENTS.md— update if the changes affect anything documented there - Commit:
feat(<skill-name>):orfix(<skill-name>):ordocs(<skill-name>):depending on the nature of the change, followed by a lowercase imperative description
Deleting a skill
- Confirm with the user before deleting
- Remove the entire
.agents/skills/<skill-name>/directory - Review
AGENTS.md— remove any references to the deleted skill - Commit:
chore(<skill-name>): remove <skill-name> skill — <reason>
Reviewing / auditing skills
- Discover all skills: use
Globwith pattern.agents/skills/*/SKILL.md - For each skill found, validate:
SKILL.mdexists (confirmed by the glob result)- Frontmatter contains
nameand non-emptydescription namevalue matches the parent directory name exactlydescriptiondescribes both what the skill does and when to invoke it
- For complete format rules and validation criteria, read
references/skill-format.md - For quality evaluation of individual skills (instruction clarity, trigger discoverability, composability), use the
review-skillskill - Auto-fix minor issues (typos, table formatting, missing optional metadata fields); ask the user before removing or renaming skills
- Apply the
refine-proseskill to anydescriptionor body text that is rewritten or newly authored during the audit — run it silently before saving - If the audit is read-only and no issues are found, no commit is needed
- If audit findings require corrections, apply fixes and commit:
fix(<skill-name>): <description of issue corrected>
Examples
Creating a new skill called deploy:
"Create a skill that deploys the app to staging using the deploy.sh script."
Expected: agent creates .agents/skills/deploy/SKILL.md with appropriate frontmatter and instructions, commits with feat(deploy): add deploy skill — ...
Updating an existing skill's description:
"Update the skills skill description to mention it also handles auditing."
Expected: agent reads existing SKILL.md, updates the description frontmatter, bumps metadata.version, commits with docs(skills): update description to mention auditing
Deleting an obsolete skill:
"Remove the old
Expected: agent confirms with user, removes .agents/skills/pdf/ directory, updates AGENTS.md if needed, commits with chore(pdf): remove pdf skill — no longer used
Auditing all skills:
"List all my skills and check if any have issues."
Expected: agent globs .agents/skills/*/SKILL.md, validates each skill's frontmatter and name/directory alignment, reports findings, auto-fixes minor issues (e.g. typos, table formatting), commits any corrections with fix(<skill-name>): <description of issue corrected>
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.

