Skip to content

putervision/behavior-mcp / .agents/skills/video-ingest

video-ingest

Verified

Teaches the agent to process, ingest, analyze, and compare WebM, MP4, and GIF video recordings using vision-memory-mcp and state-memory-mcp.

Video Frame Digesting & Temporal Memory Skill (video-ingest)

This skill provides step-by-step guidance, best practices, and operational patterns for digesting WebM, MP4, and GIF video recordings using @putervision/vision-memory-mcp.


1. When to Use Video Ingestion

Use video ingestion whenever you encounter:

  • E2E Playwright / Cypress / Selenium Test Artifacts: Recorded .webm screenchunks or .mp4 test run videos.
  • Bug Reproduction Videos: User-uploaded screen recordings demonstrating UI glitches or crashes.
  • UI Walkthrough Recordings: Demonstrations of complex multi-step user workflows.
  • Visual Regression Diagnostics: Comparing a passing baseline video run against a failing test run.

2. Ingestion Strategies & Parameter Tuning

ScenarioRecommended ParametersWhy
Action Event Timestamps (Highest Precision)action_timestamps: [1.2, 3.5, 7.0]Samples keyframes at exact interaction timestamps (clicks, types, navigation events) from test runners or state-memory logs.
Dynamic UI / Animationsscene_threshold: 0.3, fps: 1Combines scene-change detection (gt(scene,0.3)) with 1 fps background sampling to capture major screen transitions without frame bloat.
High-Speed Test Runsfps: 2 or fps: 5Increases frame rate sampling for rapidly switching UI test steps.
Long Screen Recordingsfps: 0.5, scene_threshold: 0.4Lowers sampling rate to conserve storage while extracting unique keyframe states.

3. Mandatory Dual-MCP Evidence Workflow

When diagnosing bugs or linking test runs to task nodes:

  1. Ingest Video: Call manage_video (action: "ingest") with file path and action timestamps.
  2. Extract Evidence Payload: Read the returned evidence_payload containing source_video_id, frame_range, and timestamps_ms.
  3. Build Evidence Pack: Call create_evidence_pack linking keyframe_state_ids with state-memory-mcp task or blocker node IDs.
  4. Compare Trajectories (On Failure): Call compare_states(video_a_id: "...", video_b_id: "...") to pinpoint exact frame divergence points.

4. CLI Quick Reference

# Ingest WebM / MP4 video recording with custom category
vision-memory-mcp video ingest ./recording.webm --category e2e_test

# Ingest with explicit action timestamps
vision-memory-mcp video ingest ./bug.mp4 --category bug_repro --action-timestamps 1.2,3.5,8.0

# Inspect chronological timeline & keyframes for a video ID
vision-memory-mcp video inspect vid_d7fc21c0ad52f861

# List all ingested video memory records
vision-memory-mcp video list

5. Consolidated Video MCP Tools Reference

  • manage_video (action: "ingest"): Ingests video file/base64, extracts keyframes, dHash deduplicates, and generates CLIP vector embeddings.
  • manage_video (action: "timeline"): Fetches step-by-step keyframes, exact timestamps (timestamp_ms), OCR snippets, and grounded target handles.
  • manage_video (action: "search"): Searches video memory by description query, category, tags, or file path.
  • compare_states (video_a_id, video_b_id): Calculates similarity score between two recordings and pinpoints exact timestamp divergence.
  • create_evidence_pack: Produces an immutable, cryptographically hashable evidence pack payload linking keyframes to task graph nodes.
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 469kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 321kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 321kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 321kUpdated Claude CodeCodexVerifiedInstall