Sutradhaar exposes 8 tools with significant quality gaps. Tool naming is inconsistent: skill_view lacks verb clarity, mcp__* tools are opaque proxies without individual schemas, and skill__* tools use double-underscore namespacing that obscures intent. Descriptions exist but are generic (10-50 chars for most); none explain WHEN to use the tool or what it returns. Input schemas are present for skill tools but lack type constraints (e.g., max_results is integer but no min/max bounds). Error handling is absent: no recovery guidance, no categorization of retryable vs fatal errors. No tool annotations (readOnlyHint/destructiveHint) despite clear risk levels declared in metadata. Output schemas are undocumented. The skill_view tool is always available but its purpose (progressive disclosure) is not explained in its description.

• Add explicit input schemas to all skill tools with type constraints: max_results (min: 1, max: 100), range (regex pattern for A1:Z100 format), spreadsheet_id (non-empty string). Document in parameter descriptions.
• Expand tool descriptions to 50-200 characters, answering: What does it do? When should the LLM call it? What does it return? Example: 'Search Gmail messages by query string. Returns up to N matching message IDs, subjects, and senders. Use this to find emails before reading or forwarding them.'
• Add tool annotations: mark skill__sheets-writer__append and mcp__filesystem__* with destructiveHint=true; mark read-only tools with readOnlyHint=true. Include idempotentHint for tools safe to retry.
• Implement error handling with recovery guidance. For skill__sheets-reader__read, return 'Spreadsheet not found. Try list_spreadsheets() to find the correct ID.' For skill__gmail-reader__search, return 'No results found. Try a simpler query or check the account has messages.'
• Document output schemas for each tool. Example for skill__gmail-reader__search: 'Returns {results: [{id, subject, sender, date}], total_count, has_more}'. This enables agents to chain tools and extract required fields.