Loading ToolBench
Local, read-only HTTP Archive (HAR) analysis server for AI agents.
HAR Analysis server has 16 well-named tools with consistent verb_noun patterns (load_, list_, get_, search_, find_, detect_, analyze_, compare_, trace_). All tools have descriptions (avg ~120 chars, within baseline 34-392 range). Input schemas are present and typed for all tools. However, descriptions are often terse and lack actionable context for LLM selection. Output schemas are not documented, LLMs cannot predict response structure. Error handling returns structured error objects but lacks recovery guidance. No tool annotations (readOnlyHint, etc.). Security is sound (read-only operations, path validation, secret redaction), but composition could be tighter, some tools overlap (list_requests + search_requests, list_endpoints + list_domains).
Deep analysis of an endpoint — param classification, header patterns, etc.
Diff two requests across all dimensions.
Detect API patterns (REST, GraphQL, JSON-API, etc.) in the HAR.
Detect authentication mechanisms used in the HAR.
Find requests with 4xx/5xx status codes or failed requests.
Find requests exceeding a time threshold.
Get a high-level overview of a loaded HAR.
Output schemas not documented. LLMs cannot predict response structure for any tool. Descriptions state 'Returns summaries' or 'Get full details' but do not specify fields, types, or pagination behavior.
Error responses lack recovery guidance. _error() returns {error, code, message} but does not tell LLM what to do next. E.g., 'HAR_NOT_LOADED' should suggest 'Call load_har() first with a valid .har file path.'
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 54 | <=2025-11-25 | v2 |
Get full details of a specific request.
Get full details of a response.
Get aggregate statistics about a loaded HAR.
List all domains in the HAR with request counts.
List unique API endpoints (method + normalized path) with request counts.
List requests in a loaded HAR with optional filters. Returns summaries without body content.
Load a HAR file from disk and return a summary.
Search across all requests and responses in a HAR file.
Find where a specific value appears across all requests/responses.
Tool descriptions are terse (avg ~65 chars for analysis tools). Lack context on WHEN to use each tool. E.g., 'detect_auth' vs 'detect_api_pattern', when should LLM choose one over the other? No guidance.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). All tools are read-only, but LLM cannot infer this from schema. Annotations would clarify safety and retry semantics.
Overlapping tool responsibilities. list_requests + search_requests both filter/search requests; list_endpoints + list_domains both enumerate resources. LLM must reason about which to call, wasting tokens.