About SkillVetAI — Mission & Standards
SkillVetAI is an independent research and catalog resource for understanding AI capabilities, concrete Agent Skill packages, permissions, workflow risks, and practical adoption choices.
What the site contains today
The /skills/ section contains capability guides. A capability guide explains a general workflow—such as code review, browser automation, PDF summarization, or email triage—and discusses typical use cases, permissions, risks, and guardrails.
The /packages/ section contains real package records imported from canonical GitHub repositories and the public ClawHub catalog. Each record is tied to a fixed commit or registry version and content hash. A package’s presence does not certify security, publisher identity, or platform compatibility.
That distinction matters:
- A capability guide helps you decide what kind of workflow may be useful.
- A Skill package record identifies concrete publisher attribution, source, version, content hash, install information when authoritative, and separately labelled platform or scan evidence.
Why SkillVetAI exists
Agent Skills can contain instructions, scripts, references, and access to external tools. A short marketplace description rarely answers the questions that matter before adoption:
- What does the workflow need to access?
- Can it write, execute, send, deploy, or purchase?
- Is the source still maintained?
- Which Agent platform and version has actually been tested?
- What changed between releases?
- Which conclusions come from the author, a scanner, a platform, or an independent test?
SkillVetAI is being built to make those questions legible and traceable.
Our evidence labels
We avoid using a generic Verified badge for different kinds of evidence.
- Editorial overview means the page contains editorial analysis of a capability. It is not package verification.
- Source verified means a canonical locator and immutable release identity have been recorded. Publisher attribution may still be unclaimed or unverified.
- Compatibility verified will mean a fixed release was reproduced on a recorded platform version and environment.
- Verification expired will mean the previous evidence is no longer current enough to rely on without retesting.
Automated scans are best-effort signals. A scan with no findings is not a guarantee that a Skill is safe.
Editorial approach
Current guides aim to provide specific use cases, typical permissions, risks, worked examples, alternatives, and practical guardrails. Utility, complexity, and risk labels are editorial assessments of the general workflow; they are not user ratings and do not describe every implementation.
Package records identify whether material claims are registry-supplied, source-declared, platform-official, automatically inferred, reproduced by SkillVetAI, or editorially assessed. See the Editorial Policy for the current rules.
Independence and commercial relationships
SkillVetAI is not affiliated with Codex, Claude, OpenClaw, or another Agent platform unless a relationship is explicitly disclosed. Sponsored placement must be labelled and cannot purchase a better risk conclusion, a Verified status, or a higher natural ranking.
Safety and limitations
- Content is informational and is not legal, security, financial, or compliance advice.
- Review the exact source and fixed version before installing third-party instructions or code.
- Test high-risk Skills in an isolated environment without production credentials.
- Platform behavior and upstream content can change after a page is published.
- Report errors, rights concerns, or security issues through the Contact page.
Direction
The source- and version-backed package catalog is now live as a Phase 1 dataset. The next work is higher-coverage classification, permission and dependency extraction, version changes, independent security scans, and reproducible compatibility tests. Existing capability guides remain educational and comparison layers; they are not relabelled as package records without a canonical source and immutable release.