Aphanis has 5 tools with reasonable naming (verb_noun pattern) and descriptions (100-200 chars), but critical gaps undermine quality. Input schemas are present but lack proper type definitions for several parameters, 'mode' and 'tone' are described as strings with enum-like values but not formally constrained as enums in JSON Schema. Parameters lack descriptions in some cases (e.g., 'perturb_statistical_watermarks' is described but its purpose is cryptic). No output schemas are documented, LLMs cannot predict what these tools return. Error handling is absent; no guidance on retryability, user-fixable errors, or recovery paths. The 'sanitize_file' tool modifies state (WRITE risk) but lacks confirmation/dry-run patterns. Overall, definitions are skeletal, suitable for a CLI but insufficient for reliable LLM agent integration.

• Add formal JSON Schema enum constraints for 'mode' (paranoid|aggressive|standard|minimal) and 'tone' (conversational|casual|tech-lead|academic|executive) parameters. Replace string descriptions with schema-level validation.
• Document output schemas for all 5 tools. Example: 'sanitize_text returns {sanitized_text: string, watermarks_detected: string[], confidence: number}'. This enables LLMs to chain tools and extract results.
• Expand parameter descriptions to explain intent and dependencies. E.g., 'perturb_statistical_watermarks: If true, rephrase common AI vocabulary markers (e.g., "in conclusion" → "so"). Use when mode=paranoid for maximum obfuscation.'
• Add error handling guidance. Document which errors are retryable (e.g., file not found) vs. fatal (e.g., invalid file format). Return actionable messages: 'File not found at /path/to/file. Check the path and try again.'
• Rename 'humanize_text_tool' to 'humanize_text' and clarify its relationship to the 'humanize' parameter in 'sanitize_text'. Consider merging if they serve the same purpose, or document when to use each.