Static security scanner for AI agent Skills (SKILL.md + bundled scripts). Zero runtime dependencies.
SkillsGuard exposes a single tool, scan_skill, with a basic HTTP POST interface. The tool has a description (70 chars), but the input schema is minimal and lacks proper JSON Schema structure with type definitions. Parameters 'content' and 'filename' are documented in the description but not formally typed in a schema object. No output schema is documented. The tool performs static analysis on skill files for security vulnerabilities, a legitimate read-only operation, but the definition quality falls well below production standards. The server is HTTP-based (not STDIO), which is positive for transport, but the tool definition itself is underdeveloped.
Scans skill content (SKILL.md or bundled scripts) for security vulnerabilities using static analysis rules. Accepts content via HTTP POST with optional JSON body containing content and filename fields.
Input schema lacks proper JSON Schema structure. Parameters 'content' and 'filename' are mentioned in the description but not formally defined with type, required, and constraint metadata. The schema shown in the tool definition is incomplete.
No output schema documented. The tool returns a JSON object with 'filename', 'findings', and 'safe' fields, but the structure of 'findings' (array of what?) is not formally specified. LLMs cannot plan downstream operations without knowing the response structure.
Tool description is generic and lacks context for LLM selection. It does not explain WHEN to use this tool, what prerequisites exist, or how findings should be acted upon. A 70-char description is below the 194-char baseline for production tools.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 39 | 2026-07-28+ | v2 |
No error handling guidance. The server returns HTTP 400/413/500 errors with generic messages ('Invalid JSON body', 'Payload too large', 'Internal Server Error'). LLMs receive no actionable recovery steps or error classification (retryable vs. fatal).
Parameter 'filename' is optional with a default, but the description does not explain the default value or when to override it. The 'content' parameter lacks constraints (max length, encoding, format).