Nodestradamus has 9 well-named tools following verb_noun patterns (project_scout, analyze_deps, get_impact, etc.), all with READ_ONLY risk classification. However, critical gaps prevent a higher score: (1) Input schemas are visible but INCOMPLETE, most tools accept only repo_path with minimal parameter detail; depth and threshold parameters exist but lack bounds, validation rules, or constraints in descriptions. (2) Descriptions are GENERIC and lack WHEN-to-use guidance, 'Scout the project structure' (38 chars) and 'Build and analyze the dependency graph' (37 chars) are below the 50-200 char baseline and fail to explain what the LLM should do with results. (3) Output schemas are UNDOCUMENTED, no specification of what these tools return, forcing LLMs to guess at response structure. (4) Error handling is absent, no guidance on what happens when repo_path is invalid, git history is unavailable, or embedding models fail. (5) Tool #5 (codebase_health) and tools #6-9 have INFERRED definitions only, visible in file references but NOT explicitly registered in callable handler code we can inspect. Schema quality is further hampered by missing bounds (depth: int with no min/max), missing enums (threshold: 0-1 but not declared as constrained), and no documentation of what structure is returned.

• Add complete output schemas for all 9 tools. Document return types, required fields, and nested structure. Example: project_scout returns {files_count: int, languages: [str], directories: int, ...}
• Expand tool descriptions to 50 - 200 characters with WHAT, WHEN, and key fields. Example: 'get_impact: Analyze how changes to a file ripple through the codebase. Use this BEFORE modifying critical code to understand dependencies and reverse-dependencies. Returns affected_files, affected_functions, and risk_level.'
• Add explicit bounds and enums to parameters: depth (min: 1, max: 10), commits (min: 1, max: 10000), threshold (enum: [0.5, 0.6, 0.7, 0.8, 0.9, 0.95] or min: 0.0, max: 1.0), top_k (min: 1, max: 100).
• Document parameter formats: repo_path must be absolute or relative path to local git repository; query must be < 1000 characters; file_path must use / separators relative to repo_path.
• Add error handling to descriptions or tool documentation: 'Raises: InvalidRepositoryError if repo_path does not exist or is not a git repo. Returns: empty results if no git history is available. Timeout: 60s per analysis.'
• Provide recovery guidance: 'If repo_path fails, call project_scout first to validate the repository structure. If embedding models unavailable, call semantic_search with timeout=5 for fallback.'
• Implement input validation at handler level: validate repo_path exists, validate depth and commits are in bounds, validate threshold is 0.0 - 1.0, validate query length < 1000 chars.

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry