SkillsSafe defines 2 tools with clear, domain-specific purposes. Both tools have well-written descriptions (200+ chars) that explain WHAT they do, WHEN to use them, and what they return. Input schemas are present and properly typed with JSON Schema. The scan_skill tool uses oneOf to enforce mutually exclusive parameters (url vs content), which is sophisticated. However, output schemas are not documented, the response structure is not formally specified, forcing LLMs to infer what fields to expect. Error handling is mentioned in descriptions but not formalized with recovery guidance. Tool names follow verb_noun convention (scan_skill, get_report) and are unambiguous. Parameters are well-described with enums (lang) and clear constraints. No security issues detected, no credentials exposed as parameters.

• Document output schemas for both tools. For scan_skill, specify: {decision: 'INSTALL'|'REVIEW'|'BLOCK', risk_score: 0-100, threats: [{category: string, severity: string, description: string}], scan_id: string}. For get_report, specify: {scan_id: string, report_url: string, generated_at: ISO8601}.
• Add error handling section to scan_skill description: 'Errors: invalid_url (malformed or unsupported scheme), timeout (URL unreachable after 30s), invalid_content (empty or non-text), rate_limit (60 req/hour). Retry with exponential backoff for timeout; ask user to verify URL for invalid_url.'
• Specify URL format constraints: 'url must be a valid HTTP(S) URL (https://... or http://...), max 2048 chars, must be publicly accessible. Supports GitHub raw URLs (raw.githubusercontent.com) and ClawHub URLs (clawhub.ai).'
• Add content size constraint: 'content must be non-empty UTF-8 text, max 1MB (1,048,576 bytes). Larger files should be uploaded via URL parameter instead.'
• Formalize scan_id format in get_report: 'scan_id is a unique identifier returned by scan_skill, format: ss_[8-16 hex chars]_[timestamp]. Invalid or expired scan_ids return 404 with available recent scans.'