메뉴
Use when asked to analyze, review, or describe a local video file (e.g., .mp4, .mov). Describes how to use the `mt watch` command to stream large videos to Gemini securely with server-side context caching and multi-turn thread support.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
| name | video-analysis |
| description | Use when asked to analyze, review, or describe a local video file (e.g., .mp4, .mov). Describes how to use the `mt watch` command to stream large videos to Gemini securely with server-side context caching and multi-turn thread support. |
Use this skill when you need to analyze a local video file. Large video files cannot be processed directly in your local context or uploaded as standard attachments without hitting token and size limits. The mt watch command provides a dedicated workflow to securely upload the video in chunks, cache it server-side, and stream Gemini model responses.
To start a new analysis on a video, run mt watch with the file path and your prompt:
mt watch "/path/to/video.mp4" "Describe what happens in this video in detail."
If the prompt is long or complex, you can construct it first and pass it via a variable:
PROMPT="Look closely at the UI transitions. What color is the primary button before and after the click?"
mt watch "/path/to/video.mp4" "$PROMPT"
The command will output the model's text response.
Note: For machine-readable output or extra diagnostics (like file deduplication or cache telemetry), append --json or --verbose.
When you run mt watch with the --verbose flag (or inspect the JSON output), you will receive a thread_id. You can use this thread_id to ask follow-up questions without re-uploading or re-processing the entire video from scratch.
# Add --verbose to see the thread_id
mt watch "/path/to/video.mp4" "Describe the video" --verbose
# Use the returned thread_id for a follow-up:
mt watch "/path/to/video.mp4" "What was the text on the second screen?" --thread <thread_id>
Upload local screenshot/video artifacts from agent runs and return renderable file embeds (`![[file ...]]`) plus `mighty-file:` refs for thread UI previews.
Use when the user uploads files or a .zip archive (like user research or logs). Describes how to locate attached files in the sandbox, extract archives if necessary, synthesize findings, and save results back to Mighty.
Record durable architectural or policy choices with `mt decision`. Use when choosing between alternatives, setting lasting constraints, superseding an accepted decision, or capturing significant rationale while context is fresh.
Use when a workstream needs a reviewable deliverable, especially a GitHub PR. Covers `mt pr` for the normal PR-backed path, `mt artifact create` for existing deliverables, and the workstream delivery contract that hands PRs into review and merge gates.
Use when asked to write, migrate, or organize long-form Mighty docs, or decide whether something should be a doc or a spec/task/decision. Covers `mt doc` authoring, doc types, and when guides or runbooks should live in the Mighty graph instead of standalone repo markdown.
Use Mighty-native eval primitives from external repos. Covers the native eval record model, local or hosted write-back flows, and when to reach for the deeper `external-evals` skill for framework-specific mapping.