AgentMemory exposes 3 tools with partial schema coverage and inconsistent descriptions. Tool names lack action verbs (mem::file-context, mem::index-persistence-compact, mem::retrieval-index-verify). Descriptions range from adequate (file-context: 127 chars) to verbose (retrieval-index-verify: 156 chars), but parameter descriptions are sparse or missing. Input schemas are present with types, but lack depth in constraints and validation guidance. No output schemas documented. Error handling and recovery guidance absent. Tools appear to manage internal memory indices rather than expose user-facing operations, limiting composability.

• Rename tools with action verbs: 'get_file_context', 'compact_memory_indices', 'verify_retrieval_index'. This enables LLMs to infer intent from the name alone.
• Document output schemas for each tool. E.g., file-context should return {observations: [{file: string, importance: number, content: string}], totalMatched: number}. This enables downstream tool chaining.
• Add parameter constraints and ranges. E.g., 'timeBudgetMs: number (100 - 300000, default 60000)', 'importance threshold: 1 - 10 (default 4)', 'files: array of file paths (glob patterns supported, e.g., src/**/*.ts)'.
• Add error handling guidance to descriptions. E.g., 'If compaction times out, retry with a smaller timeBudgetMs or call with dryRun=true to estimate duration. If verification detects drift > 50%, repair=true is recommended.'
• Expose higher-level tools that match user intents. E.g., 'get_code_context_for_file(file_path, max_observations)' that internally calls file-context and formats results for LLM consumption, rather than exposing raw index operations.