Loading ToolBench
A read-only Solana account monitor for ZeroClaw with no wallet custody.
Four read-only tools with clear naming (verb_noun pattern: solana_*) and basic descriptions. All tools have input schemas with typed parameters and defaults. However, descriptions are terse (10-20 chars below baseline 194), parameter descriptions lack detail on constraints/formats, and output schemas are not documented. No error handling guidance, no parameter validation rules stated, and no composition hints for multi-tool workflows. Naming is consistent but descriptions undershoot the 50-200 char LLM-optimized range.
Return the confirmed SOL balance of a public address.
Return a compact summary of the latest transaction for a public address.
Return up to 20 recent confirmed signatures for a public address.
Build a balance and recent-activity report for human review.
Tool descriptions are 10-20 characters, well below the 50-200 char LLM-optimized baseline. 'Return the confirmed SOL balance of a public address' lacks context on when to use this vs other tools, what happens on error, or what the response structure contains.
Parameter descriptions are missing or minimal. 'cluster' param has description 'Solana cluster' but does not explain valid values (mainnet|devnet), format, or consequences of choosing each. 'limit' lacks range constraints (1-20 stated in docstring but not in param description).
Output schemas are not documented. Code shows tools return dicts with fields like 'lamports', 'sol', 'slot', 'read_only', but LLMs cannot infer downstream field names or types. No documentation of response structure for chaining or field extraction.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 54 | 2026-07-28+ | v2 |
No error handling guidance. Tools call external Solana RPC and can fail (network timeout, invalid address, RPC error). No recovery hints in descriptions (e.g., 'If address is invalid, try validate_address first' or 'Retry on timeout'). SentinelError exceptions are caught but not surfaced to LLM with actionable guidance.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). All four tools are read-only and idempotent, but this is not declared in the tool definition. LLMs cannot infer safety properties and may treat them as stateful or destructive.