Loading ToolBench
AI agent framework with on-chain identity and messaging via AlgoChat on Algorand
Scoring was not performed
27 of 29 tools have no visible input schema definition with type and description metadata. Only corvid_send_message and corvid_recall_memory show documented schemas. This prevents LLMs from understanding parameter requirements and validation rules.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 28 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 41 | - | v1 |
No output schemas documented for any tool. Agents cannot plan downstream calls or validate data extraction when they do not know what fields to expect. This violates the documented-return-types baseline (100% of A+ tools).
Tool descriptions are generic and lack context. Examples: 'List all sessions with optional project filter' (no mention of what a session is, what fields returned, when to use it), 'Get detailed information about a specific session' (too vague, what details?). Most descriptions under 80 chars lack WHEN/WHY guidance.
corvid_send_message and corvid_ask_owner serve overlapping purposes (communication with external entities). No clear differentiation or composition rules. LLMs will struggle to pick the right one.
No error handling guidance in any tool description. Descriptions do not indicate whether tools are idempotent, retryable, or destructive. Agents cannot reason about failure recovery or side effects.
Tools like corvid_list_sessions and corvid_github_list_prs have no documented pagination support. No mention of limit, offset, page_size, or result count limits. Large datasets will blow context windows.
Parameter descriptions missing or minimal. For corvid_send_message, 'to_agent' is described as 'Agent name or ID to message' but no guidance on how to discover valid names/IDs or what happens if ambiguous. No validation rules stated.
No mention of authentication or permission gates in tool descriptions. Tools like corvid_github_create_issue and corvid_github_fork_repo are write operations but descriptions do not clarify access requirements or failure modes if user lacks permissions.
Duplicate or overlapping tool groups without clear composition. GitHub tools (16 total) lack a unifying discovery or filtering strategy. No guidance on which to call first, how results chain, or when to batch vs. single calls.