NightShade CLI provides 12 tools with consistent naming (verb_noun pattern) and reasonable descriptions (avg ~120 chars). All tools have input schemas with typed parameters. However, parameter descriptions are sparse or missing for nested objects (e.g., todo_write's 'todos' array items lack field descriptions). Output schemas are undocumented, LLMs cannot predict return structure. Error handling is minimal; most tools return plain strings without recovery guidance. The 'task' tool (subagent delegation) lacks documentation of expected subagent behavior and failure modes. Security concerns: shell tool is unrestricted and dangerous; no mention of input sanitization or permission gates.

• Document output schemas for all tools. Example: read_file returns {lines: string[], total_lines: number, path: string}. This enables LLMs to chain tools and extract data reliably.
• Add descriptions to nested object properties. For todo_write, document that 'content' is a string (max 500 chars) and 'status' is an enum of 'pending|in_progress|completed'.
• Restrict shell tool: add a 'command_type' enum parameter (e.g., 'build|test|lint|run') and validate against an allowlist. Require explicit confirmation before executing destructive commands (rm, dd, etc.).
• Enhance error messages with recovery guidance. Instead of 'Error: file not found', return 'File not found: /path/to/file. Use list_dir() to explore available files.' This guides the LLM's next step.
• Expand task tool documentation: explain what the subagent can do (same tools as parent?), how to structure the prompt, what the return value contains, and how to handle failures (e.g., 'If task fails, try breaking it into smaller subtasks').