Single tool 'convene_council' has a well-structured schema with proper types and enums, but description lacks LLM-optimized clarity on WHEN to use it vs alternatives. Parameter descriptions are present but generic. Output schema is not documented, LLMs cannot predict the structure of council results, dissent adjudication, or ledger format. Error handling is absent; no guidance on what happens if an advisor API fails, budget is exceeded, or synthesis is skipped. Tool does one thing (convene advisors) but composition with ledger writing is implicit, not exposed as separate tools.

• Document the full output schema for convene_council: return type (object), fields (question, stakes, plan, outcomes[], dissent, synthesis, ledger_path, latency_ms), and nested structures (AdvisorOutcome, AdvisorVerdict, etc.). Include example output shape.
• Expand tool description to 150-250 chars explaining: (1) What it does (convenes multiple LLM advisors), (2) When to use it (complex decisions requiring diverse perspectives, high-stakes deliberation), (3) What it returns (structured positions, dissent, synthesis), (4) Cost implications (stakes determine budget).
• Add error handling guidance: 'If an advisor fails, partial results are returned with status=error. If budget is breached before synthesis, positions are preserved but synthesis is skipped, ask the user to review incomplete council. If all advisors fail, the tool returns an empty council with status=degraded.'
• Clarify 'stakes' enum in parameter description: 'low (0.1 USD, full synthesis), medium (0.25 USD, full synthesis), high (1.0 USD, no synthesis, ends with open questions), audited (1.0 USD, no synthesis, audit trail enabled).'
• Expose ledger writing as a separate tool or document it as a side effect: 'This tool writes a markdown ledger entry to [ledger_path]. Ledger is append-only and immutable. To review past councils, read the ledger file directly.'