AI-powered personal assistant with multi-agent orchestration, MCP server integration, and multi-channel messaging
LIA Assistant exposes 14 tools with significant quality gaps. Tool names lack consistent verb-noun structure (e.g., 'python_sandbox_tool' vs 'send_email'). Descriptions are present but generic (avg ~80 chars, well below the 194-char baseline for A+ tools). Input schemas are visible but lack depth: most parameters have minimal type info and no constraints (enums, ranges, patterns). No output schemas documented. Error handling is absent, no recovery guidance, no categorization of retryable vs fatal errors. Security concerns: python_sandbox_tool and mcp_tool_proxy expose irreversible operations without confirmation patterns. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite high-risk tools like delete operations.
Automate web browser interactions using Playwright for accessibility tree navigation
Create and manage calendar events via CalDAV
Extract text from documents (PDF, DOCX, XLSX, RTF)
Send emails via IMAP/SMTP connectors
Look up IP geolocation using GeoIP2 database
Proxy tool calls to user-configured MCP servers
Send push notifications via Firebase Cloud Messaging
Execute Python code in a sandboxed environment with optional network access
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite high-risk operations. python_sandbox_tool (IRREVERSIBLE), mcp_tool_proxy (IRREVERSIBLE), and write tools (email_send, telegram_send, calendar_event, push_notification) lack explicit risk markers.
Inconsistent tool naming. Names like 'python_sandbox_tool', 'web_fetch_tool', 'text_to_speech_tool' use '_tool' suffix (non-standard). Should follow verb_noun pattern: 'execute_python', 'fetch_web_page', 'convert_text_to_speech'. LLMs infer intent from name, inconsistency causes selection errors.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 38 | 2026-07-28+ | v2 |
Generate QR codes for TOTP enrollment and data encoding
Perform hybrid semantic and keyword search using BM25 and embeddings
Convert speech to text using Sherpa-ONNX offline STT
Send messages via Telegram Bot API
Convert text to speech using Edge TTS with Microsoft neural voices
Fetch and extract content from web pages using readability algorithm
No output schemas documented. Tools return unstructured responses with no field definitions. LLMs cannot plan downstream calls or extract required IDs (e.g., does web_fetch_tool return 'content' or 'text'? Does semantic_search_tool return 'results' or 'matches'?).
No error handling guidance. Tools lack recovery instructions, error categorization (retryable vs fatal), or actionable error messages. E.g., python_sandbox_tool execution failure returns no guidance on whether to retry, adjust code, or escalate.
Irreversible operations (python_sandbox_tool, mcp_tool_proxy) lack confirmation/dry-run patterns. Agents can execute arbitrary code or proxy calls without safeguards, risking catastrophic errors.