Cognis Recall exposes 4 tools with basic schemas and descriptions, but critical gaps undermine production readiness. Tool names lack action verbs (scan, relevant, add, audit are weak; should be scan_codebase, retrieve_documents, add_document, verify_audit_log). Descriptions are present but generic (10-80 chars, below the 50-200 char LLM-optimized baseline). Parameter descriptions exist but lack constraints, format guidance, and dependency hints. The 'passphrase' parameter appears in 3 tools as plaintext input, a critical security violation. Output schemas are not documented anywhere in the source. Error handling is absent, no recovery guidance, no categorization, no actionable messages. Tool composition is weak: 'relevant' and 'add' both require vault_path + passphrase, forcing repeated credential passing. No pagination, no batch operations, no idempotency markers.

• Rename tools to action-verb form: scan → scan_codebase, relevant → retrieve_documents, add → add_document, audit → verify_audit_log. This helps LLMs parse intent from the name alone.
• Move passphrase from tool parameters to server-side secret injection. Store vault credentials in environment variables or a secure vault, and pass them internally. Never expose secrets in tool parameters.
• Document output schemas for all tools. For scan: {findings: [{id: string, sev: string, where: string}], score: integer}. For relevant: {documents: [{title, text, tags, score}], total: integer}. For add: {success: boolean, message: string}. For audit: {entries: [{timestamp, action, hash}]}.
• Add parameter constraints and format guidance. Example: 'vault_path: path to encrypted .vault file (required)', 'k: number of results to return (1-100, default 5)', 'query_text: search query (1-1000 characters)'.
• Implement error handling with recovery guidance. Return structured errors: {error: string, code: string, retryable: boolean, suggestion: string}. Example: 'Vault locked. Try again in 5 seconds or call unlock_vault() first.'