Adopt the Agent Skills specification - #7
Merged
Merged
Conversation
Add scripts/validate-skills.sh, checking the hard requirements from https://agentskills.io/specification: the closed set of frontmatter fields, name rules including the directory match, and description and compatibility length limits. It also checks the allowed-tools format, which the spec defines as a space-separated string, and this repo's convention that every skill ships a README.md. Body length and short descriptions are reported as warnings so existing skills do not fail the run. Wire it into test-all.sh as Test 5.
|
Verdict: APPROVE Files reviewed:
Findings:
Action required: None — ready to merge |
Two jobs. The first runs test-all.sh, which docs/TESTING.md had listed as a planned option. The second installs skills-ref, the reference validator from the spec authors, and runs it over every skill. skills-ref is deliberately unpinned. Tracking its latest release is how this repo learns that the spec has moved, so a failure after an upstream release is information rather than a broken build. It stays out of test-all.sh because it needs Python and the local suite is meant to run with no dependencies.
Skills get copied out of this repo one folder at a time. Once a folder is copied it carries no provenance, so the license should travel with the file. license is an optional field in the Agent Skills spec.
Structure and spec conformance are now checked deterministically by validate-skills.sh and skills-ref. Asking a model to repeat them costs tokens and can disagree with the validators. The prompt keeps overlap and quality, where a model actually adds something.
Reposition the README around the open format while keeping Claude Code named as where the skills are developed and tested, since compatibility with other hosts has not been verified. Record the frontmatter rules in SKILL_GUIDELINES and in the skills README checklist, add the new script to SCRIPTS, and explain there why CI runs both validators, so neither gets removed later as duplication. In TESTING, move GitHub Actions out of the future options, and fill in configure-settings.sh, which was tested but undocumented.
check() takes the description first and the command after it. The example in the Adding New Tests section had them the other way round, so following it produced a test that always reported the wrong result.
decebals
force-pushed
the
feat/agent-skills-spec
branch
from
August 28, 2026 19:44
a621500 to
11553e9
Compare
|
Verdict: APPROVE Files reviewed: 18 skill SKILL.md files + infrastructure changes Findings:
Action required: None — ready to merge Summary: This PR adds the MIT license field to all 18 existing skill files and introduces infrastructure for Agent Skills specification compliance. No overlaps with existing skills, no quality concerns — all changes are additive metadata and validation tooling. The new |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Closes #5.
The skills already met the spec's hard requirements, so nothing about them changed except the added
license: MIT. The rest of this is checks and documentation so it stays that way.scripts/validate-skills.shcovers the spec rules plus the two things the reference validator does not enforce: theallowed-toolsformat, and this repo's convention that every skill ships a README. Body length and short descriptions are warnings, so the eight skills currently over 500 lines do not fail the build. Splitting those intoreferences/is separate work.CI gains a Test workflow with two jobs. One runs
test-all.sh, which docs/TESTING.md had been listing as a planned option since the beginning. The other installsskills-reffrom the spec authors and runs it over every skill. Both validators run because they cover different ground, and SCRIPTS.md explains why, so neither gets deleted later as duplication.skills-refis unpinned on purpose: if the spec moves, a red build is how we find out.The skill-review prompt no longer checks structure. The validators do it deterministically and for less.
No rename, and no claim that these work on hosts I have not tried. The README says they follow the spec and are developed and tested with Claude Code.
Verified locally: the suite passes, both validators pass on all 18 skills, the bash one catches all 12 broken fixtures I threw at it, and a fresh session still loads all 18 skills with the new frontmatter field.