Three tools with partial schemas and descriptions. verify_tool_call and issue_evidence have input schemas with parameter descriptions, but descriptions are generic (under 100 chars). audit_mcp_config has minimal schema detail. No output schemas documented. No error handling guidance. Tool names are action-oriented (verify_, issue_, audit_) but descriptions lack WHEN/WHY context. Parameters lack enums, ranges, or format constraints. No mention of idempotency, side effects, or recovery paths.

• Add output schemas for all three tools. Example for verify_tool_call: {type: 'object', properties: {verdict: {type: 'string', enum: ['allow', 'deny']}, dimensions: {type: 'object'}, findings: {type: 'array', items: {type: 'string'}}}}
• Expand tool descriptions to 150-250 chars. Include WHEN to call: 'Verify a tool call against CCS 7 dimensions (structure, semantics, math, injection, identity, policy, risk). Use before executing untrusted agent tool calls to detect injection attacks, SSRF, command injection, and policy violations.'
• Add enum constraints to verdict parameter: {type: 'string', enum: ['allow', 'deny'], description: 'Verification outcome: allow (safe to execute) or deny (blocked by policy or security findings)'}
• Document error cases and recovery paths. E.g., 'If verification fails with security_findings, return the specific injection patterns detected and suggest sanitization steps.'
• For issue_evidence, add idempotentHint: false and document that repeated calls with identical inputs produce different evidence_hash values (due to timestamp/nonce). Clarify retry semantics.