Skill Spec Validator
Checks Agent Skills against the agentskills.io spec and Claude Code best practices.
The skill is primarily read-only file inspection and does not request credentials, sensitive data, or destructive actions, so its baseline permission risk is low. It does not disclose data flow, confirmation, isolation, rollback, or dependency-security practices; attribution is limited to an author field and MIT license, with unverified publisher identity, so points are deducted.
The checklist, procedure, and report format are reasonably clear, but the skill assumes the undeclared fd command and provides no script, tests, abnormal-input handling, or diagnostic failure behavior. It also targets skills/ in the body while the assessed path is .claude/skills/validate-skills, creating path-alignment uncertainty. Static evidence warrants a conservative score.
The audience and invocation method (/validate-skills) are clear, and the scenario fits skill-specification and documentation checks. However, input/output boundaries, non-fit cases, precise trigger conditions, Chinese-language support, and environment or network assumptions are not defined, limiting applicability.
The document is readable and includes metadata, a checklist, procedure, sample output, Markdown-link guidance, and MIT licensing. It lacks versioning, changelog, maintenance ownership, update path, dependency notes, and troubleshooting guidance; referenced external specifications are not pinned locally.
The checklist can guide the core validation task and the sample format is reusable. However, no validator implementation, committed tests, or representative results are provided; the process remains manual and evidence for output completeness or marginal benefit over manual review is limited.
The rules and sample report are auditable from the source, but there is no third-party execution evidence, automated test suite, or CI coverage of the skill's key paths. Static review therefore cannot verify operational accuracy, warranting only limited credit.
- The body targets skills/ while the assessed path is .claude/skills/validate-skills; confirm the intended directory and repository layout before use.
- The skill assumes fd and provides no fallback, error handling, or automated tests, so abnormal environments may require manual diagnosis.
- External specification links are not version-pinned, and Chinese-language or mainland-China network compatibility is not documented.
What does this skill do, and when should you use it?
This is a development skill for validating Agent Skills against the agentskills.io specification and Claude Code best practices. It reviews each skill directory’s SKILL.md and checks naming, descriptions, optional fields, body length, progressive-disclosure links, and Markdown link usage. It produces a PASS/FAIL report organized by skill directory. The skill lives at .claude/skills/validate-skills/SKILL.md, while its instructions target a skills/ directory.
Finds skill directories under skills/; reads each SKILL.md; checks name format and directory-name matching, description length, optional fields, third-person description style, body length, direct discoverability of reference files, Markdown link formatting, redundant content, and concision; then produces a validation report containing PASS/FAIL results for each check.
- A skill author wants to check a new SKILL.md before submitting it against the agentskills.io rules.
- A repository maintainer needs a consistent review of multiple skills against Claude Code best practices.
- A team has changed reference files and wants to detect hidden multi-level progressive-disclosure chains.
- A publisher wants a directory-by-directory issue report before releasing a skill collection.
What are this skill's strengths and limitations?
- Provides concrete checks for specification compliance, description quality, body length, and reference structure.
- Uses a clear report format for locating issues by skill.
- Released under the MIT license and attributed to Callstack.
- The instructions target skills/, while the skill itself is stored under .claude/skills/, creating a scope mismatch in the source.
- The source provides no test results, test suite, or broader platform verification evidence.
- Its `/validate-skills` trigger is Claude Code-oriented and requires adaptation elsewhere.
- The source does not describe automatic remediation; it only specifies validation and reporting.
How do you install this skill?
The source does not provide a dedicated installation command for this skill. It is included in the callstackincubator/agent-skills repository at .claude/skills/validate-skills/SKILL.md. The README documents installation methods for the collection across Claude Code, Codex, and other assistants, but does not list a separate validate-skills installation command.
How do you use this skill?
In a Claude Code environment that supports the skill, use /validate-skills. The instructions call for checking every skill under skills/ and suggest fd -t d -d 1 . skills/ to find skill directories. The report format uses ## Validation Results, per-directory checks, and [PASS]/[FAIL] entries.