Loading ToolBench
Drop a video and a markdown file into a folder. Every platform publishes itself.
socialdrop exposes 4 tools with clear verb-noun naming (list_drops, doctor, publish, get_stats). All tools have descriptions (avg 115 chars) and input schemas with types. However, parameter descriptions are sparse or missing entirely. The publish tool lacks descriptions for 'only' and 'now' parameters. Output schemas are not documented, LLMs cannot predict return structure. Error handling is absent; tools will fail silently or return raw exceptions. No input validation guidance. The server is functional but falls short of production-grade agent tooling.
Report per-platform authentication/setup status and drop-file validation errors.
Refresh consolidated Insights tables in every md file and return the metrics.
List every video drop (md + video pair) in a folder with its publish status.
Publish all due video drops in a folder. Set only=[...] to restrict platforms, now=True to ignore future schedules. Returns per-video results.
Output schemas not documented. LLMs cannot predict return structure (list_drops returns DropInfo objects, doctor returns dict with 'token_storage'/'platforms'/'problems' keys, publish/get_stats return dicts). Without documented schemas, agents cannot plan downstream operations or extract required fields.
Parameter descriptions missing or incomplete. 'publish' tool has 'only' and 'now' parameters with no descriptions in the schema. 'folder' parameter descriptions are generic ('Folder path...') without guidance on what happens if the folder is empty or contains invalid drops.
No error handling or recovery guidance. Tools will raise ValueError or other exceptions without actionable messages. E.g., if folder validation fails, the LLM sees 'is not a directory' but has no guidance on what to do next (try a different path? create the folder?).
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 63 | 2026-07-28+ | v2 |
Destructive operations (publish, get_stats) lack confirmation or dry-run support. 'publish' modifies state across multiple platforms with no preview or rollback. Agents cannot safely test before committing.
No pagination or result limits. 'list_drops' and 'doctor' return all results without limit. A folder with thousands of drops will blow the context window. No 'limit' or 'offset' parameters.