Loading ToolBench
CLI and MCP server that turns long videos into transcripts, visual notes, reconstructed code, and structured markdown your AI editor can read
vidistill provides 6 tools with basic descriptions and input schemas. Naming follows verb_noun convention (analyze_video, get_transcript, list_outputs). However, descriptions are minimal (50-150 chars, below the 194-char baseline), parameter descriptions lack detail about constraints and formats, and output schemas are not documented. No error handling guidance, no parameter validation rules, and no indication of idempotency or side effects. Tools are well-scoped (one responsibility each), but lack the LLM-optimization depth expected of production tools.
Run the full vidistill pipeline for a given input (YouTube URL, local file, or remote URL). Processes video through multiple passes to extract transcript, visual analysis, code reconstruction, and structured notes.
Read code files from an existing vidistill output directory. Returns all reconstructed code files with their content.
Read structured notes from an existing vidistill output directory. Returns synthesis result with overview, key decisions, concepts, topics, and suggestions.
Read people/participants information from an existing vidistill output directory. Returns extracted participant details including names, roles, organizations, and contributions.
Read transcript from an existing vidistill output directory. Optionally filter by time range.
List all vidistill output directories with metadata. Shows video titles, durations, types, generation dates, and file counts.
Output schemas not documented. LLMs cannot infer what fields to expect from get_transcript, get_code, get_notes, get_people, or list_outputs. This blocks downstream tool chaining and forces agents to guess at response structure.
Parameter descriptions lack format/constraint details. 'outputDir' is described as 'Path to vidistill output directory' but does not specify: required vs optional, absolute vs relative, what happens if path doesn't exist, or example format. 'startTime' and 'endTime' lack units (seconds? milliseconds?) and valid ranges.
No error handling guidance. Tools do not document what happens on failure (e.g., outputDir not found, invalid time range, file read errors). LLMs receive no recovery hints and cannot self-correct.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 56 | 2026-07-28+ | v2 |
analyze_video description does not state that it modifies state (writes files to disk). Agents need to know this is a destructive/side-effecting operation, not a read-only query. No mention of prerequisites (API keys, network access, disk space).
No pagination support documented for list_outputs. If many output directories exist, the tool may return thousands of items, exhausting context. No limit, offset, or cursor parameters visible.