Menu
Find a person, logo, object, landmark, or visual reference across local clips or captured media with timestamped Overcast evidence.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
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-visual-target-search |
| description | Find a person, logo, object, landmark, or visual reference across local clips or captured media with timestamped Overcast evidence. |
Use this skill when the task is to locate a visual target across videos, images,
or captured case media. Use the broad overcast skill and
overcast/reference/verbs.md for exact flags.
For a person with a reference image:
overcast doctor --json
overcast case init --json
overcast face ./clip.mp4 --match ./person.jpg --json # a match ≥75% auto-suggests a finding
overcast finding list --state triage --json # triage the auto-suggested lead(s)…
overcast finding accept <id> --json # …accept a real match (or `dismiss <id>`)
overcast crop <face-record-id> --all --class face --json
overcast ask "where does the reference person appear, with timestamps and confidence?" --json
overcast brief --export ./visual-search.md --json # short by default; --full for the per-match timeline
For an object or open-vocabulary target (--detect needs a bound OWLv2 detector —
build it once with scripts/visual-db-uv.sh --detect (it prints DETECT_PY), then
export DETECT_PY=… and bind via the preset: overcast provider setup apply --preset owl-local --yes, which persists a portable shipped: ref for detect.py and uses the
venv python, NOT system python3 which lacks torch/transformers):
overcast see ./clip.mp4 --detect "red backpack" --json
overcast crop <see-record-id> --all --class "red backpack" --json
overcast ask "list target detections with timestamps, confidence, and crop paths" --json
For logos, landmarks, or near-duplicate visual references:
overcast index create refs --type image-ransac --local --json
overcast index add ./reference-logo.png --to <index-id> --json
overcast image match ./clip.mp4 --index <index-id> --json # a RANSAC hit auto-suggests a finding
overcast finding list --state triage --json # then accept/dismiss the lead
Return timestamped matches, similarity or confidence where available, source
record.id, media.at, and cropped evidence paths created by crop.
face --match / image match auto-suggest findings — resolve them with
finding list --state triage → accept/dismiss so a run doesn't leave an
un-triaged queue; the default brief is short, --full for the per-match timeline.
State whether the match came from face --match, see --detect, or local
image-ransac matching.
Face detections are sampled-frame detections, not unique-person counts. Use
face --match <image> for a specific person and include confidence caveats.
For exact evidence, use crop to materialize local image records, then
synthesize with ask and brief.