Single tool 'run_tests' has excellent schema completeness with Zod-validated input/output shapes, comprehensive parameter descriptions, and proper type definitions. Tool name is action-verb-based and clear. Description is detailed (280+ chars) and explains the tool's purpose, auto-detection behavior, and truncation logic. Input parameters (command, cwd, full) are all described with context and examples. Output schema is fully documented with typed fields (cruxVersion, runner, exitCode, summary, failures, runnerError). Tool annotations present (readOnlyHint, destructiveHint, idempotentHint, openWorldHint). However, the tool lacks error recovery guidance in descriptions, when runner auto-detection fails or a command is invalid, the description does not guide the LLM on what to do next. Error handling exists in code (CruxError catch) but is not surfaced in the tool definition itself.

• Add error recovery guidance to the tool description: 'If auto-detection fails (e.g., vitest not found), pass an explicit command like ["npx", "vitest", "run"] via the command parameter.'
• Document the runnerError kinds in the tool description with recovery steps: 'compile-error: fix the test file syntax; config-error: check vitest.config.ts; binary-not-found: install vitest or pass explicit command; unknown: check server logs.'
• Clarify the truncation trade-off: 'Set full: true only when debugging crux's sanitization or auditing original payloads, it increases response size and token cost.'
• Add a note about idempotency: 'This tool is idempotent, calling it multiple times with the same inputs produces the same output and has no side effects.'
• Consider adding a 'timeout' parameter (optional, default 30s) to prevent hung test runners from blocking the agent indefinitely.