メニュー
WR3 editorial-bench skill cortex — operational heuristics, on-tone examples, lessons growth surface. Loaded by wr3-editorial-bench agent at every dispatch.
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 | wr3-editorial-bench |
| description | WR3 editorial-bench skill cortex — operational heuristics, on-tone examples, lessons growth surface. Loaded by wr3-editorial-bench agent at every dispatch. |
| model | opus |
| lifecycle_tier | scheduled |
| cost_class | reasoning |
| ceiling | $0.5 |
| contract_version | 1.0.0 |
Monthly cron 1st Monday 07:00 WITA. Researches SOTA video editorial from 12 reference brands (NYT Opinion Video, Bloomberg Originals, Reuters, Wired, Pudding, Rest of World, ProPublica, Vox, Quartz, Drift, Vice News, Al Jazeera English) + 3 Bali Zero competitor video accounts (Lets Move Indonesia, Emerhub, Flado IG Reels). Multi-LLM: Gemini 3.1 Pro for long-context ingestion, Claude Opus for synthesis, DeepSeek for pattern extraction.
Law 8 (Passato/Presente/Futuro) — read by yt-metrics-analyst (weekly) and critic (every run via skill load)
(seeded empty — populated by wr3-reflexion-synth weekly after first 3 episodes pass critic gate)
TBD-2026-05-18 — first 3 pilot episodes feed examples here.
(growth surface — wr3-reflexion-synth appends max 10 lessons/week here)
Skill cortex created at S7.3 step of WR3 genesis. No operational lessons yet.
_external-bench-video-YYYY-MM.md)| Tier | LLM | Cost | Quota source |
|---|---|---|---|
| 2 | Gemini 3.1 Pro via agy CLI | $0.00 | Google AI Ultra subscription |
| 3 | Claude Opus 4.7 via claude-cascade.sh | ~$0.30 | Claude MAX OAuth (no SDK) |
| 4 | DeepSeek V4 Pro reasoning_effort=high | ~$0.05 | DeepSeek API (Keychain key) |
| Total | ~$0.35 | within $0.50 ceiling |
ceiling_usd: $0.5 (declared in docs/wr3/contracts/editorial-bench.yaml, hard_halt_on_exceed: true)BudgetExceededError → cascade decision per Law 7 (Numeri prima):
~/.claude/agents/wr3-editorial-bench.md~/Desktop/nuzantara/docs/wr3/contracts/editorial-bench.yaml~/.claude/skills/bali-zero-brand/~/Desktop/nuzantara/docs/wr3/symbiosis-precedence.mdUSE FOR EVERY non-trivial mandate — feature, fix, refactor, research, audit, ops, content — coding or not. The master operating loop of the organism: TRIAGE the mandate into a gear (1 liscio / 2 standard / 3 profondo), then drive GROUND → DESIGN → BUILD → VERIFY → SHIP+ARM → PROVE-LIVE → ALIGN-FLEET → CLEAN → CAPTURE, routing the full arsenal at maximum-without-waste (Fable 5 architect+final-gate, Sonnet 5 implementers, Codex red-team+sandbox, Gemini agy width, DeepSeek/GLM refuters, Ollama local for PII, NotebookLM ground-truth). Supersedes opus-mythos (2026-07-02): Fable is native again — the width-surrogate retires; its deep/wide TAC patterns live on as Gear 3. SKIP only true one-liners — and declare it: "GEAR 1: <why>".
Bali Zero brand cortex — palette tokens, typography, voice registers, taboo phrases, layout families, past-carousel reference. Loaded by wr2-design-architect at start of every carousel run. Progressive disclosure: this entry point is small; deep resources (constitution, tokens, voice examples, layouts) are loaded only when needed.
WR3 audio-asset-producer skill cortex — operational heuristics, on-tone examples, lessons growth surface. Loaded by wr3-audio-asset-producer agent at every dispatch.
WR3 b-roll-curator skill cortex — operational heuristics, on-tone examples, lessons growth surface. Loaded by wr3-b-roll-curator agent at every dispatch.
WR3 brief-interpreter skill cortex — operational heuristics, on-tone examples, lessons growth surface. Loaded by wr3-brief-interpreter agent at every dispatch.
WR3 clip-renderer skill cortex — operational heuristics, on-tone examples, lessons growth surface. Loaded by wr3-clip-renderer agent at every dispatch.