Six tools with mixed quality. Tool naming is action-oriented (state, read_comments, write_comments, compile, image, listen), but descriptions are terse (10-50 chars) and lack context for LLM selection. Input schemas are present for most tools but lack descriptions for critical parameters. The write_comments tool has a complex discriminated union anchor type, but no guidance on when to use each anchor kind. Output schemas are not documented. Error handling and recovery guidance are absent from all tool descriptions.

• Expand tool descriptions to 50-200 chars. Example for state(): 'Retrieve the current project state: file paths, compilation status, document sections with line ranges, and pending comment counts. Call this first to understand the document structure and what needs review.'
• Document output schemas for all tools. For read_comments, specify: returns array of comment objects with fields {id, text, anchor, author, created, updated, revision, replies}. For compile, specify: {success: bool, errors: [{line, col, message}], warnings: [...]}.
• Add descriptions to all parameters. For write_comments.action, explain: 'add: create new comment; reply: respond to existing comment; suggest: propose text edits; withdraw: remove comment; edit: modify existing comment; delete: permanently remove comment.'
• Add error handling guidance to tool descriptions. For compile(): 'Returns {success: false, errors: [...]} if LaTeX compilation fails. Each error includes line number and message. If errors occur, display them to the user and ask for corrections.'
• Provide anchor selection guidance in write_comments description. Example: 'Use paper anchor for document-level comments, section anchor for comments tied to a section, source_range anchor for comments on specific lines of code. Include the exact text fragment (old field) that the comment references.'