Loading ToolBench
Keyless wrapper over the Wayback Machine "available" API and the CDX search API. Look up the archived/historical version of a web page, list every captured snapshot of a URL over time, and find out when a URL was first archived. No authentication required.
Three well-named tools with clear, actionable descriptions (avg 180 chars, within 10 - 1024 baseline). All tools have complete input schemas with typed properties and required fields. Naming follows verb_noun pattern (get_*, list_) and is unambiguous. However, output schemas are not documented, LLMs cannot predict response structure for downstream chaining. Error handling returns HTTP status + truncated text but lacks recovery guidance. No parameter constraints (enums, ranges) despite opportunities (e.g., limit capped at 1000, timestamp format YYYYMMDD). Missing tool annotations (readOnlyHint, idempotentHint) despite all three being read-only, safe operations.
Count how many times a URL has been captured by the Internet Archive Wayback Machine, and find when a URL was first and last archived. Returns the total number of captures plus the earliest and latest capture timestamps (YYYYMMDDhhmmss).
Look up the archived/historical version of a web page in the Internet Archive Wayback Machine. Returns the closest available snapshot (its Wayback URL, timestamp and HTTP status). Pass a timestamp to find the snapshot nearest a specific date; omit it to get the latest archived version.
List the captured snapshots of a URL in the Internet Archive Wayback Machine over time, using the CDX index. Each entry includes the capture timestamp, the original URL, HTTP status code, MIME type, content digest and a direct Wayback snapshot URL. Optionally restrict to a date range with from/to (YYYYMMDD).
Output schemas not documented. LLMs cannot predict response structure (e.g., does get_snapshot return {snapshot: {...}} or {url, found, snapshot}?). Breaks tool chaining and forces agents to guess field names.
No parameter constraints. 'limit' accepts any number but is capped at 1000 in code; 'timestamp' expects YYYYMMDD or YYYYMMDDhhmmss but no regex/pattern enforced. LLMs will pass invalid formats, causing silent failures.
Error responses lack recovery guidance. httpError() returns {error: status, message: truncated_text}. LLM cannot determine if error is retryable, user-fixable, or fatal. No suggestions for next steps.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 61 | 2026-07-28+ | v2 |
Missing tool annotations. All three tools are read-only and idempotent (safe to retry), but no readOnlyHint or idempotentHint declared. Agents cannot infer safety properties.
Parameter descriptions lack format/constraint details. 'timestamp' says 'YYYYMMDD or YYYYMMDDhhmmss' in description but no pattern field. 'limit' says 'default 50, max 1000' but no minimum/maximum JSON Schema properties.