メニュー
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.
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
| name | overcast-init |
| description | 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. |
One-time setup for overcast.
pi install npm:@kdrrr/overcast (inside pi) or
npm i -g @kdrrr/overcast for the standalone binary.overcast skills install --dest ~/.codex/skills
overcast skills install remains the Claude Code default
(~/.claude/skills), and --harness claude-code is the explicit Claude
target.npm i -g @cloudglue/tinycloud@0.3.12 then tinycloud install --latest, or
tinycloud update). overcast needs tinycloud ≥ 0.3.12 — its watch
envelope inlines the verbatim watch/listen transcripts
(watch.speech.v1); the face + index verbs need ≥ 0.3.4, and the image
see/extract verbs behind the opt-in see:tinycloud provider need
≥ 0.3.7. Override the invocation with OVERCAST_TINYCLOUD_CMD if it isn't
on PATH.overcast doctor --json (pi pinned, ffmpeg/ffprobe runnable,
Cloudglue key, tinycloud CLI + version, optional uv/visual-db readiness).watch/listen/face/index providers
reach Cloudglue via the tinycloud CLI; configure it (tinycloud setup cloudglue)
or export CLOUDGLUE_API_KEY.overcast provider setup show --profile default --json
overcast provider setup plan --preset cloudglue --profile default --json
overcast provider setup apply --preset cloudglue --profile default --yes --json
overcast doctor --profile default --json
cloudglue for tinycloud watch/listen/face plus built-in ffmpeg enhance.fal for see/enhance with FAL_KEY.hf for see/enhance with HF_TOKEN.elevenlabs for listen/enhance with ELEVENLABS_API_KEY.owl-local for OWLv2 open-vocabulary object detection.see:tinycloud (choice, --verb see --choice tinycloud) for Cloudglue
file-level image analysis via tinycloud see/extract (needs tinycloud
≥ 0.3.7; --detect facts are boxless — no crop).deepface-local for local face detect/match through DeepFace.face verb, while image/face DBs are still case-owned local indexes:
scripts/visual-db-uv.sh --face
overcast doctor --json
overcast provider setup apply --verb face --choice deepface-local --profile default --yes --json
overcast index create logos --type image-ransac --local --json
overcast index create localfaces --type deepface-local --local --json
overcast skill per investigation to run
case setup, select targets/sources/indexes, and optionally set case-level
automation such as --auto-sense, --auto-index-new, and --findings
(defaults to suggest — score/text triggers auto-suggest leads; review is
the legacy text-only mode, off disables).Then use the overcast skill to drive the verbs.
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.
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.
Is this real? — triage whether an image or video was altered by accumulating independent leads: C2PA / Content Credentials provenance, EXIF capture metadata, ELA/noise forensic overlays, and a sun/shadow time check — never a single-signal call.