メニュー
Find the interval a person or object is present — anchor one appearance, then sweep outward with face-match / detect until it drops off both sides.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Give any agent senses (video/audio/image understanding) and OSINT reach (search/capture/monitor) organized around an investigation case. Use when the user wants to analyze a video/audio/image, scan or monitor sources for a target, or ask/brief over accumulated findings. Drives the `overcast` CLI (built on pi + the tinycloud/Cloudglue perception backend); see reference/verbs.md for the full verb surface.
Install and configure overcast for this harness: install the CLI, verify the system ffmpeg, and configure reusable provider profiles. Use once per machine/profile before driving the `overcast` skill.
Turn a case into a visual evidence board — materialize face and object crops, link the same person across clips, connect themes, and render the corkboard as a CSI brief plus a live monitor wall.
Make unreadable footage legible — denoise/upscale a marked moment, re-run OCR and detection on the enhanced output, and record what was recovered with honest provenance (interpolation is a lead, not proof).
Work out where a photo or clip was taken — check embedded EXIF/GPS first, then pull signage, landmarks, and terrain clues, reverse-image-search the strongest ones, and corroborate to a location with cited evidence.
Run a standing surveillance watch on public sources for a target — auto-sense new media, auto-suggest matches into a triage queue, and keep a live control-room wall — so new evidence surfaces itself over time.
| name | overcast-presence-window |
| description | Find the interval a person or object is present — anchor one appearance, then sweep outward with face-match / detect until it drops off both sides. |
Use this skill to answer "from when to when is on screen?" — a first/last
appearance interval, not a single instant. It anchors one confirmed appearance and
expands until the target stops appearing, the presence-tracking analog of a
temporal search. Use the broad overcast skill and
overcast/reference/verbs.md for exact flags.
overcast doctor --json
overcast case init --json
overcast watch ./clip.mp4 --json # -> record id REC
overcast face ./clip.mp4 --match ./person.jpg --json # a specific person (similarity 0-100)
overcast grid ./clip.mp4 --count 16 --json # then see the montage for an object
# person: widen the window; --fps controls sample density (precision vs cost)
overcast face ./clip.mp4 --match ./person.jpg --start <a> --end <b> --fps 1 --min-similarity 55 --json
# object: step frames outward and check presence
overcast see frame://REC@<t> --prompt "Is <target> present? answer only yes or no" --json
overcast note "<target> present" --ref REC --at <first-last> --confidence medium --json
overcast view REC --at <first-last> --json
overcast brief --export ./presence.md --json
One or more [first-last] intervals with the per-hit citations (record.id +
media.at) that bound them, and the sample density used. If the target leaves
and returns, report each interval separately rather than one span covering the
gap.
Sampled detections are per-frame, not continuous — presence between samples is
inferred; raise --fps to tighten boundaries at higher cost. Occlusion or an
off-camera moment splits one presence into several intervals — that's a real
result, not noise. Face similarity is 0-100. Needs the video local.