Behaviorlock defines 5 tools with complete input schemas and descriptions. All tools follow verb_noun naming (validate, compare, explain, fingerprint, render). Descriptions are clear and domain-specific (avg ~100 chars), exceeding the 10-char minimum. All parameters have type definitions and descriptions. However, output schemas are not documented, LLMs cannot predict response structure. Error handling is minimal; tools return generic JSON without recovery guidance. No tool annotations (readOnlyHint, idempotentHint) despite all being read-only. Parameter descriptions lack format/constraint details (e.g., 'Path to trace file' doesn't specify file extension, encoding, or validation rules). No pagination or result-limiting patterns despite tools that could return large datasets.

• Add outputSchema to all tool registrations. Document the JSON structure returned by each tool (e.g., behaviorlock_compare returns {scenarios: [{assertions: [{id, status, baseline, candidate, reason}]}]}). This enables LLMs to extract fields and chain tools.
• Add readOnlyHint: true to all tool registrations via the tool annotation pattern. This signals to LLMs that these tools are safe to retry and do not modify state.
• Enhance parameter descriptions with format constraints. Change 'Path to trace file' to 'Path to portable trace file (JSON, UTF-8 encoded, must conform to trace.schema.json)'. Add examples of valid paths.
• Implement structured error responses with recovery guidance. Instead of {error: 'not_found'}, return {error: 'assertion_not_found', assertion_id: '...', available_assertions: [...], next_step: 'Call behaviorlock_compare to generate assertions'}.
• Document the BEHAVIORLOCK_CONTRACT environment variable fallback in tool descriptions. Add: 'If contract is omitted, defaults to BEHAVIORLOCK_CONTRACT env var or behaviorlock.json in current directory.'