AI-driven video pipeline: describe a video in plain words, your AI assistant writes a spec, one command renders it (voiceover, music, SFX, motion graphics, Shorts, thumbnails).
VideoPip has 14 tools with consistent naming (verb_noun pattern: get_, list_, analyze_, init_, write_, preflight, render_, verify_, make_, audition_). All tools have descriptions (avg 180 chars, within 10-1024 baseline). However, input schemas are incomplete: 6 tools have empty input objects (get_guide, get_schema, list_assets, analyze_clips missing contact_sheets type, preflight missing exact_timing/scan_cards types). Output schemas are undocumented, no tool declares what it returns. Parameter descriptions exist but lack format/constraint details (e.g., 'path' doesn't specify file vs folder, 'voice' lacks enum of valid voices). Error handling is minimal: write_project returns structured errors, but most tools lack recovery guidance. No tool declares permissions, audit trails, or idempotency. Composition is sound (tools chain well: analyze_clips → init_project → write_project → preflight → render_start), but no batch variants or pagination for list_assets.
Analyze a clip or a folder of clips: duration, size, static stretches (likely logo/text cards), scene cuts, suggested safe ranges and action peaks. Contact sheets (timestamped frame grids) are written next to the clips in `_sheets/` - LOOK at them before trusting the suggestions.
Render a short line to a WAV so the user can check a voice or a pronunciation fix.
How to turn a user's loose video request into a valid VideoPip project.yaml (read this first).
Title/description/chapters/tags text built from the last render.
JSON schema for project.yaml.
A commented starter project.yaml. Names: product-countdown, vertical-short.
6 tools have empty input schemas (get_guide, get_schema, list_assets, and 3 others with missing type declarations on boolean/string params).
No tool documents its output schema. LLMs cannot plan downstream calls or extract fields without knowing return structure. Violates pattern:tool and pattern:response-shaper.
Parameter descriptions lack format/constraint details. 'path' doesn't specify file vs folder; 'voice' lacks enum of valid voices (only default shown); 'name' in get_template doesn't list valid template names. Violates pattern:tool-description and review:param-validation-rules.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 64 | 2026-07-28+ | v2 |
Create a starter project.yaml in `folder` (optionally pre-filled from a folder of clips).
Builtin sounds, overlay templates and project templates.
(Re)build the project's vertical shorts from the last render.
Check a project before rendering. Returns errors/warnings with suggested fixes and a runtime estimate. exact_timing renders (and caches) the voiceover for precise timings.
Start rendering in the background (takes minutes). Poll render_status(job_id).
Status and last log lines of a render job.
QA a finished video: freezes, black frames, loudness, dimensions.
Validate YAML against the schema and write it to `path` only if valid.
No error handling guidance. Most tools lack recovery hints (e.g., 'If preflight fails, check project.yaml syntax'). Only write_project returns structured errors. Violates pattern:recovery-guide and pattern:error-classification.
No tool declares permissions, audit trails, or idempotency. Destructive tools (render_start, make_shorts, audition_voice) lack confirmation patterns. Violates pattern:permission-gate, pattern:audit-trail, and pattern:confirmation-request.