FORGE defines 6 tools with clear verb-noun naming (forge_prepare, forge_metrics, forge_plan, forge_context, forge_run_status, forge_run_complete). All tools have descriptions (avg ~120 chars, within baseline 34-392 range). Input schemas are present with type definitions and descriptions for all parameters. However, output schemas are NOT documented, critical gap for LLM planning. Parameter descriptions are adequate but lack constraint details (ranges, enums, format specs). Error handling and recovery guidance are absent. Tool composition is sound (each does one thing), but descriptions could be more LLM-optimized per pattern:tool-description baseline.

• Document output schemas for all 6 tools. Specify field names, types, and what data each field contains. Example: forge_prepare returns {context: string, risk: string, commands: string[], modelTier: string}.
• Add constraint details to parameter descriptions. For 'cwd', clarify: 'Absolute path or relative to current working directory. Must be readable and contain a valid repository.' For 'task', specify: 'Free-form description of the coding outcome (1-4000 chars).'
• Add error handling guidance to each tool description. Example: 'If cwd is invalid, returns error with available paths. If repo is unreadable, suggests permission checks.'
• Implement dry-run or preview mode for forge_run_complete. Return a summary of changes before atomically committing evidence, allowing the agent to confirm or abort.
• Add tool annotations (readOnlyHint, destructiveHint, idempotentHint) to schema definitions. Mark forge_run_complete with destructiveHint: true to signal irreversibility.