Loading ToolBench
LangGraph-powered AI agent that solves the NYT Wordle puzzle daily
This server has significant definition quality gaps. Of 5 tools, 3 are write operations (send_message_to_slack, send_whatsapp_message, post_comment_in_nyt) with minimal descriptions and incomplete schemas. The two Wordle-specific tools (attempt_wordle_guess, is_word_in_previous_answers) have reasonable descriptions but lack depth. Critically, send_whatsapp_message and post_comment_in_nyt have extremely sparse descriptions (under 50 chars). Parameter schemas are present but inconsistently detailed. No output schemas are documented. Error handling is absent, tools provide no guidance on retry logic, failure modes, or recovery paths. The mix of game-logic tools and social-media side-effects suggests poor composition and unclear domain focus.
Attempts a guess at the Wordle puzzle. Args: guess: The word to guess. Returns: A list of integers representing the feedback for the guess. 1 = Correct position 2 = Wrong position 0 = Not in word
Checks if a word has been used as a previous answer in Wordle. Args: guessword: The word to check. Returns: True if the word has been used as a previous answer, False otherwise.
Post a comment on New York Times.
Sends a message to a default Slack channel for notifications and alerts. This tool posts AI-generated updates, results, or alerts to a predefined Slack channel. Perfect for logging agent actions, sharing computation results, or team notifications without needing channel configuration. Args: message: The text content to send (up to 40,000 characters). Returns: Success confirmation with timestamp, or detailed error message. Raises: SlackApiError: Authentication, rate limits, or API failures. Example: send_message_to_slack("AI agent completed analysis: 95% success rate")
Critical: send_whatsapp_message description is 63 characters ('Sends a WhatsApp message to a list of contacts defined in the environment.'). This violates the 10-1024 character baseline and lacks critical context: What environment variable? What format? What happens on failure? LLM cannot determine when to use this tool vs send_message_to_slack.
Critical: post_comment_in_nyt description is 36 characters ('Post a comment on New York Times.'). Under the 50-character floor for actionable descriptions. Missing: What authentication is required? Where does the comment get posted? What is the input format? Can it fail? LLM has zero guidance.
High: No output schemas documented for ANY tool. The rubric requires documented return types (baseline: 100% of A+ tools have them). attempt_wordle_guess returns a list of integers but the format is not formally specified, is it [feedback_for_pos_0, ..., feedback_for_pos_4]? send_message_to_slack returns 'Success confirmation with timestamp' but no schema. This forces LLMs to infer response structure and plan blindly for downstream operations.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 55 | 2026-07-28+ | v2 |
Sends a WhatsApp message to a list of contacts defined in the environment.
High: send_whatsapp_message schema has a 'contacts' parameter with type 'string' and default '+919442807217'. The description says 'List of phone numbers as comma-separated string' but the default is a single number. Ambiguous: should the LLM pass a single contact or a CSV list? The schema type should be clarified, and the default should be empty or null to avoid accidental message sending to a hardcoded number.
High: No error handling guidance. All tools lack descriptions of failure modes, recovery steps, or actionable error messages. If send_whatsapp_message fails due to invalid phone number or rate limit, what should the agent do? Retry? Ask the user? This violates pattern:recovery-guide.
Medium: Composition issue, this server mixes game logic (attempt_wordle_guess, is_word_in_previous_answers) with social-media operations (send_message_to_slack, send_whatsapp_message, post_comment_in_nyt). These are unrelated domains. A Wordle solver should focus on game mechanics. External notifications should be a separate MCP server or configurable callback, not embedded tools. This violates pattern:tool (each tool should do one thing).
Medium: Credentials and secrets risk. send_whatsapp_message relies on 'contacts defined in the environment' but there is no documentation of the expected environment variable names. The pyproject.toml lists 'slack-sdk>=3.39.0' and 'pywhatkit>=5.4', suggesting API keys are expected, but they are not mentioned in tool descriptions. If secrets are embedded in environment or code, they risk leaking into logs.
Medium: attempt_wordle_guess description says 'Returns: A list of integers representing the feedback for the guess. 1 = Correct position, 2 = Wrong position, 0 = Not in word'. This is helpful but incomplete, what is the length of the list? Is it always 5 elements (one per letter)? In what order? This should be a formal output schema with field definitions, not prose.
Medium: send_message_to_slack parameter 'message' has type 'string' with description 'The text content to send (up to 40,000 characters).' but no minLength or maxLength constraint in the schema. LLMs cannot validate this constraint at runtime and may send messages exceeding the limit, causing API failures.