Communitygithub.com

TIKAZI/TIKAZ-Codex-Video-Intelligence

Evidence-graded video reading with timestamps, keyframes, source cards, and cross-video synthesis.

TIKAZ-Codex-Video-Intelligence とは?

TIKAZ-Codex-Video-Intelligence is a Codex agent skill that evidence-graded video reading with timestamps, keyframes, source cards, and cross-video synthesis.

対応~Claude CodeCodex CLI~Cursor
npx skills add TIKAZI/TIKAZ-Codex-Video-Intelligence

お気に入りのAIに質問する

このエージェントスキルを事前に読み込んだ状態で新しいチャットを開きます。

ドキュメント

Video Intelligence

Designed, integrated, refactored, and continuously maintained by TIKAZ.

TIKAZ method

  1. Create a source ledger before analysis: platform, stable ID, author, date, duration, access status, and evidence available.
  2. Route acquisition to video-platform-reader; never treat a title, description, or search snippet as watched content.
  3. Assign an evidence level: L0 metadata, L1 native transcript, L2 timestamped ASR, L3 transcript + keyframes, or L4 media evidence + primary-source verification.
  4. Separate five statement types: source claim, transcript evidence, visual observation, external verification, and TIKAZ inference/recommendation.
  5. Produce one card per source before cross-source synthesis.
  6. Merge only after recording agreements, conflicts, dependencies, missing evidence, and confidence.

Handoff contract

Pass structured source records, not prose-only summaries. Knowledge capture may begin only after every requested source has an access and evidence status.

Completion gate

Finish when every source is accounted for, important visual claims are checked or explicitly unverified, timestamps support key claims where available, and synthesized recommendations cite their evidence path.

Read references/routing.md for routing and references/output-contract.md for the record schema.

関連スキル