Merlion Agent has 15 tools with basic descriptions and minimal parameter documentation. Most tools lack detailed input schemas, parameters are present but lack type constraints, enums, or validation guidance. Descriptions are terse (10-50 chars for most tools), below the 50-200 char baseline for LLM-optimized docs. No output schemas documented. Error handling is absent from tool definitions. Security-critical tools (bash, bash_docker, bash_ssh) expose command execution without dry-run or confirmation patterns. Memory and skill tools are underspecified. Overall, the server follows a minimal definition pattern typical of early-stage agent frameworks.

• Expand tool descriptions to 80-150 chars. Include WHAT (action), WHEN (use case), and SIDE EFFECTS (irreversible, state change). Example: 'Execute a shell command in a sandbox. Use for file ops, builds, deployments. WARNING: irreversible, no undo. Returns {stdout, stderr, exit_code}.'
• Add input schema constraints: bash/bash_docker/bash_ssh should declare command as string with min/max length, forbidden patterns (rm -rf /), or regex validation. read/write should constrain path to prevent directory traversal.
• Document output schemas for all tools. Example for read: {content: string, size_bytes: int, encoding: string}. For grep: {matches: [{line_number: int, text: string}], total_matches: int}.
• Implement confirmation pattern for irreversible tools. Add optional 'dry_run' parameter to bash tools. Return {would_execute: string, changes: [...]} on dry_run=true, requiring explicit confirmation before actual execution.
• Split task tool into separate tools: task_create, task_update, task_get, task_list. Each with its own schema and clear parameters. Alternatively, use discriminated union schema with action-specific required fields.

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

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