菜单
Localize the exact instant of a one-way state change (light on, door opens, poster removed) by binary-searching frames — ~log2(N) VLM calls.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
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-event-bisect |
| description | Localize the exact instant of a one-way state change (light on, door opens, poster removed) by binary-searching frames — ~log2(N) VLM calls. |
Use this skill when a video contains a single MONOTONE transition — a condition
that is false before some instant and true after it (or vice versa), and does not
flip back. Binary search finds it in about log2(window/precision) vision calls
(~12 calls localizes to ~1s inside an hour). Use the broad overcast skill and
overcast/reference/verbs.md for exact flags.
see frame:// needs media on disk):overcast doctor --json
overcast case init --json
overcast watch ./clip.mp4 --json # -> record id REC (also gives shot context)
overcast see frame://REC@<lo> --prompt "Is <predicate> true? answer only yes or no" --json
overcast see frame://REC@<hi> --prompt "Is <predicate> true? answer only yes or no" --json
If both give the same answer, the flip isn't in [lo,hi]. If the predicate
toggles more than once, it isn't monotone — use overcast-pinpoint instead.
hi - lo is within your precision:overcast see frame://REC@<mid> --prompt "Is <predicate> true? answer only yes or no" --json
# keep the straddling half: if mid's answer == lo's answer, set lo=mid; else hi=mid
# (correct whichever way it flips — false->true OR true->false)
overcast note "<predicate> flips" --ref REC --at <lo-hi> --confidence high --json
overcast view REC --at <lo-hi> --json
overcast brief --export ./transition.md --json
The converged [lo, hi] bracket — the two adjacent see frames that straddle
the flip (their record.ids + media.at) — and the call count. That bracket IS
the answer window — don't over-claim a single frame.
Bisection is only valid for a one-way change; a light that blinks or a person who
comes and goes will mislead it — verify monotonicity at step 2, and fall back to
overcast-pinpoint (peak search) or overcast-presence-window for recurring or
interval events. Needs the video local.