Keel provides 6 tools with basic schemas and descriptions, but falls short of production quality. Tool names lack consistent verb prefixes (e.g., 'read' vs 'search' vs 'bash' vs 'process.run'). Descriptions are present but minimal (10-50 chars for most), providing insufficient context for LLM selection. Parameter descriptions exist but are sparse. Critical gaps: no output schemas documented, no error handling guidance, no pagination support for search results, no idempotent/destructive hints. The 'bash' and 'process.run' tools are high-risk irreversible operations with no confirmation or dry-run patterns. File access tools lack path traversal validation documentation.

• Add tool annotations: mark 'read' and 'search' with readOnlyHint=true; mark 'write', 'edit', 'bash', 'process.run' with destructiveHint=true; mark 'read' with idempotentHint=true.
• Expand tool descriptions to 100-200 chars. For 'bash': 'Execute arbitrary bash commands. Use for shell scripting, file operations, and system tasks. Returns stdout/stderr. WARNING: irreversible, no undo. Prefer process.run for structured commands.'
• Document output schemas for all tools. E.g., search returns {results: [{path, lineNumber, content}], totalCount, hasMore}; read returns {content, lineCount, byteCount}.
• Add confirmation/dry-run pattern to bash and process.run. E.g., add 'dryRun' boolean param; return what would execute without executing; require explicit confirmation for destructive commands.
• Implement pagination for search: add 'limit' (default 20, max 100) and 'offset' params; return totalCount and hasMore flag to guide agent iteration.