Server has 8 tools with complete input schemas and descriptions, but quality is inconsistent. Naming follows verb_noun pattern (open_gdb_*, launch_debug, run_gdb_cmd, close_*) which is good. However, descriptions are terse (avg ~60 chars, well below the 194-char baseline for A+ tools). Parameter descriptions exist but lack actionable detail on constraints, formats, and error recovery. Output schemas are not documented, responses are free-text markdown strings rather than structured objects. Error handling is minimal; no guidance on retryability or recovery paths. The run_gdb_cmd tool has a critical design flaw: three mutually exclusive parameters (core_dump_path, remote_address, program_path) with no clear validation or error messaging. No tool annotations (readOnlyHint, destructiveHint) despite clear risk levels assigned.

• Expand tool descriptions to 100 - 200 characters. Include WHEN to use (vs similar tools), prerequisites, and what the output contains. E.g., 'Open a core dump for post-mortem analysis. Call this after a crash to inspect signal info, backtrace, and threads. Requires the executable and optional debug symbols.'
• Convert all tool responses from markdown strings to structured JSON objects. Return {backtrace: [...], threads: [...], registers: {...}} instead of concatenated markdown. This enables LLMs to extract and chain specific fields.
• Add tool annotations to the schema: set readOnlyHint=true for open_gdb_*, list_core_dumps; set destructiveHint=true for close_debug; set idempotentHint=true for read-only tools. This signals to agents which tools are safe to retry.
• Redesign run_gdb_cmd to use a single 'session_id' parameter (or 'session_type' enum: core|remote|launch) instead of three mutually exclusive params. Document the enum values and add validation that rejects invalid combinations with a clear error message.
• Add error recovery hints to all error responses. E.g., 'Core dump not found at /path/to/core. Call list_core_dumps(directory_path="/path/to") to find available files.' This guides LLM self-correction without extra discovery calls.

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

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