メニュー
Trace a suspicious or viral clip back to its earliest appearance and originator — reverse-image-search distinctive frames, sweep sources with no recency floor, and return a cited origin verdict.
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-provenance |
| description | Trace a suspicious or viral clip back to its earliest appearance and originator — reverse-image-search distinctive frames, sweep sources with no recency floor, and return a cited origin verdict. |
Use this skill to answer "is this clip real / where did it come from?": given a
suspect video, find the oldest copy and who posted it first. It is the inverse of
overcast-copycat-sweep — searching backward toward the origin rather than forward
for rips — and reuses its geometry-gating and verdict conventions. This skill answers
WHERE a clip came from (origin / earliest copy); for whether it was ALTERED
(manipulation/authenticity — C2PA, EXIF re-save, ELA overlays, a shadow check) use
the complementary overcast-verify-media. Origin and authenticity are distinct,
non-overlapping questions — the leading verify/exif pass below establishes
provenance signals, not a fakery verdict. Use the broad overcast skill and
overcast/reference/verbs.md for exact flags.
verify) and EXIF capture metadata (exif) are the most
direct origin evidence when present — a signed manifest names the creator + edit
history. They're often stripped on re-upload, so also fingerprint distinctive
frames (which survive the re-encodes, crops, and watermarks that defeat exact
hashes AND that strip metadata):overcast doctor --sources --json
overcast case init --json
overcast verify ./suspect.mp4 --json # C2PA / Content Credentials: signer + edit manifest (needs c2patool)
overcast exif ./suspect.mp4 --json # capture device/time, editing software, GPS (needs exiftool)
overcast watch ./suspect.mp4 --json # content + transcript into memory
overcast index create origin --type image-ransac --local --json
overcast image add ./distinctive-frame.png --index <index-id> --json
--since floor, since older results sit closer to the origin:overcast source add "lens:./distinctive-frame.png" --json
overcast source add "x:video:<topic keywords>" --json
overcast source add "youtube:search:<topic keywords>" --json
overcast source add "web:<topic keywords>" --json
overcast scan --limit 20 --json
Triage on metadata alone: sort candidate hits by published ASC — the earliest
dates are your origin candidates; carry author + URL forward.
Capture the oldest few and confirm they are the SAME content (not a look-alike) — frame match plus transcript:
overcast capture <earliest-scan-hit-id> --json
overcast image match <captured-file> --index <index-id> --draw --json # --draw writes the RANSAC overlay proof
overcast listen <captured-file> --json
--ref the image match record so its overlay rides
into the brief; ALWAYS leave a tldr note with the date chain — even
"undetermined":overcast finding create "origin: earliest confirmed copy is @<author> <date> (frame match 88 inliers, transcript identical); the viral <date2> post is a re-upload" --ref <image-match-record-id> --confidence high --json
overcast note "swept lens + x + youtube + web (no recency floor); <n> candidates; earliest confirmed <date> by @<author>; verdict: re-upload" --tag tldr --json
# Wait for the note result before exporting, so the TL;DR is included.
overcast brief --export ./provenance.html --json
An origin verdict — original / re-upload-of- / undetermined — with the
date chain, the earliest confirmed poster and URL, which layers agreed (frame +
transcript), and the strongest record.id + media.at citations. "Undetermined"
is honest when the earliest copy can't be confirmed as the same content.
The earliest date you FIND is a floor, not proof of origin — deleted posts and
platforms you didn't search may predate it; say so. Confirm same-content with the
gated image match (a high inlier count on a degenerate homography is the main
false positive — eyeball the --draw overlay) plus transcript; a shared topic or a
look-alike frame is not the same clip. Face similarity is 0–100; image match is
an inlier count + a 0–1 ratio, never a 0–100 score. Apify sources bill per result.
lens ignores --since.