Akana's 14 tools have consistent naming (verb_noun pattern) and descriptions, but critical gaps undermine quality. Most tools lack documented output schemas, the rubric requires this for LLM planning. Memory and vault tools are well-described, but error handling is absent, no guidance on retryability, user-fixable vs fatal errors, or recovery paths. No tool declares permissions (read:memory, write:vault, etc.). Composition is reasonable (single responsibility), but vault_delete and vault_delete_credential are destructive with no confirmation pattern. Overall: functional but below production baseline.

• Define complete input schemas for schedule_create and schedule_update. Include required fields (name, trigger_time, action), types, and descriptions. Currently {} makes these tools non-functional.
• Document output schemas for all tools. E.g., memory_search returns {results: [{id, text, timestamp}], total_count}. vault_list returns {secrets: [string], credentials: [string]}. This enables LLM planning.
• Add error handling guidance to tool descriptions. E.g., vault_get: 'If key not found, returns error. Try vault_list() to see available keys.' memory_forget: 'Soft delete via ledger, reversible via audit trail.'
• Implement confirmation pattern for vault_delete and vault_delete_credential. Return {status: 'confirm_required', message: 'Delete secret X?', confirm_token: '...'} and require a confirm_delete tool with the token.
• Add permission declarations to each tool description. E.g., 'Requires: read:memory' for memory_search, 'Requires: write:vault, delete:vault' for vault_delete. This enables least-privilege agent setup.