bert-lab provides 16 tools with mostly complete schemas and descriptions. Naming is generally clear (verb_noun pattern: memory_search, read, write, edit, bash, grep, glob, webfetch, websearch). Descriptions range 50 - 250 chars, meeting the 10 - 1024 baseline. However, several tools lack parameter descriptions (identity, list_findings, read_ledger_rows, assemble_evidence_bundle, synthesize_artifact_body, analyze_evidence_holes), and output schemas are not documented. Error handling is minimal, no recovery guidance or categorization. The Spawn-subagent tool is vague ('handle a task') and lacks detail on failure modes. Overall: solid foundation, but incomplete parameter documentation and missing output schemas prevent a higher score.

• Document output schemas for all 16 tools. Include field names, types, and what data the LLM should expect. Example: memory_search returns {chunks: [{id, text, score, metadata}], total_count}.
• Expand Spawn-subagent description to explain: What happens if the subagent fails? Can it be retried? What does success look like? When should the parent agent use this vs. inline logic?
• Add parameter descriptions to assemble_evidence_bundle, synthesize_artifact_body, analyze_evidence_holes. Explain the structure and purpose of 'findings', 'ledger_rows', 'evidence', 'artifact', 'objective'.
• Add error handling to Bash tool: validate command syntax, document sandbox tier behavior, return actionable errors (e.g., 'Command timed out after 30s. Try with a longer timeout or break into smaller steps.').
• Implement recovery guidance for all tools. Example: 'File not found. Available files: [list]. Did you mean one of these?' or 'Query returned no results. Try a broader search term or check memory_search documentation.'