Skip to main content
ClaudeWave
Skill482 repo starsupdated 3d ago

plans

Install in Claude Code
Copy
git clone --depth 1 https://github.com/aqm857886159/Nomi /tmp/plans && cp -r /tmp/plans/docs/superpowers/plans/2026-08-27-release-media-pack- ~/.claude/skills/plans
Then start a new Claude Code session; the skill loads automatically.

2026-08-27-release-media-pack-skill.md

# Nomi Release Media Pack Skill Implementation Plan

> 🚧 进行中/待实施(状态标记由 `check:doc-status` 门岗要求;本行不改动原方案内容)

> **For agentic workers:** REQUIRED SUB-SKILL: Use superpowers:subagent-driven-development (recommended) or superpowers:executing-plans to implement this plan task-by-task. Steps use checkbox (`- [ ]`) syntax for tracking.

**Goal:** Ship a built-in Nomi playbook that turns verified desktop-release evidence into a story-first, bilingual release media package with bounded TikHub research and an honest Nomi-to-local-post handoff.

**Architecture:** Add one new `skills/release-media-pack` Skill Pack v2. Its manifest reuses the existing playbook orchestrator and current document/canvas tools; its Markdown defines the release-evidence, TikHub, story, generation, post-production, localization, and status contracts. No new network, audio, filesystem, or publishing tool is introduced, so unavailable execution is explicitly handed off instead of simulated.

**Tech Stack:** Skill Pack v2 JSON + Markdown, existing Zod manifest loader, Vitest built-in-skill regression tests.

---

## File map

- Create `skills/release-media-pack/SKILL.md`: runtime-injected workflow, taste rules, hard gates, and honest capability boundary.
- Create `skills/release-media-pack/skill.json`: seven-stage playbook, current tool whitelist, provider requirements, permissions, inputs, examples, and stage references.
- Create `skills/release-media-pack/README.md`: operator-facing material-pack layout, TikHub/runtime preflight, local post-production handoff, and status meanings.
- Create `skills/release-media-pack/examples/release-evidence.template.md`: source-of-truth checklist for current and previous releases.
- Create `skills/release-media-pack/examples/material-pack-manifest.template.json`: reusable artifact/state manifest without fake output paths.
- Create `skills/release-media-pack/evals/evals.json`: three pressure scenarios and objective assertions derived from the baseline run.
- Modify `electron/skills/builtinSkills.test.ts`: require the new built-in, validate its stage order, tool containment, pause gates, and no nonexistent runtime tool claims.

### Task 1: Add a failing built-in contract test

**Files:**
- Modify: `electron/skills/builtinSkills.test.ts`

- [ ] **Step 1: Write the failing test**

Add a test that loads `skills/release-media-pack/skill.json`, expects the stage order `evidence → research → story → build → generate → assemble → handoff`, verifies the release pack has separate story and handoff pauses, and asserts every stage tool is present in the top-level whitelist.

- [ ] **Step 2: Run the focused test and verify RED**

Run: `pnpm vitest run electron/skills/builtinSkills.test.ts`

Expected: FAIL because `skills/release-media-pack/skill.json` does not exist.

### Task 2: Add the minimal valid playbook manifest

**Files:**
- Create: `skills/release-media-pack/skill.json`

- [ ] **Step 1: Declare the current executable boundary**

Use only `read_full_text`, `read_selection`, `read_canvas_state`, `propose_storyboard_plan`, `create_canvas_nodes`, `connect_canvas_edges`, `set_node_prompt`, `run_generation_batch`, and `arrange_storyboard_to_timeline`. Do not declare TikHub, audio, filesystem, FFmpeg, HyperFrames, or upload tools that do not exist.

- [ ] **Step 2: Declare seven paused stages**

Set dependencies linearly from `evidence` through `handoff`; give generation only `image`/`video` model preferences; reuse existing writer/director skill references by directory handle.

- [ ] **Step 3: Run the focused test and verify remaining failures are about missing Markdown/content only**

Run: `pnpm vitest run electron/skills/builtinSkills.test.ts`

Expected: manifest schema and stage-contract assertions PASS.

### Task 3: Write the runtime Skill and operator contract

**Files:**
- Create: `skills/release-media-pack/SKILL.md`
- Create: `skills/release-media-pack/README.md`

- [ ] **Step 1: Write the runtime workflow in no more than 200 lines**

Include: release truth before claims; TikHub 20/10 bounded research and stop conditions; story before features; central title/motion/audio rules; separate ZH and EN outputs; Nomi MCP spend gate; current external handoff boundary; complete/conditional/partial/blocked definitions.

- [ ] **Step 2: Write the operator-facing material-package layout**

Document the version-scoped folder tree, official TikHub documentation URLs, runtime-only credential rule, GitHub Skill verification/safety boundary, local HyperFrames/FFmpeg QA, and actual-file-link handoff requirement.

- [ ] **Step 3: Scan for forbidden claims**

Run: `rg -n 'TBD|TODO|中英混排|自动上传|已完成音频|已接通 TikHub' skills/release-media-pack`

Expected: no placeholder or false-capability claim; references to mixed language appear only as prohibitions.

### Task 4: Add reusable templates and behavior evals

**Files:**
- Create: `skills/release-media-pack/examples/release-evidence.template.md`
- Create: `skills/release-media-pack/examples/material-pack-manifest.template.json`
- Create: `skills/release-media-pack/evals/evals.json`

- [ ] **Step 1: Add the release-evidence template**

Require current/previous exact tag or commit, installer identity/hash/platform/signing, merged change evidence, packaged-build reproduction, user-visible wording, limitations, and official links.

- [ ] **Step 2: Add the manifest template**

Use the package statuses `complete|conditional|partial|blocked` and artifact states `planned|draft|verified|published`; leave paths empty until real artifacts exist.

- [ ] **Step 3: Add the three baseline pressure scenarios**

Store the exact urgency, TikHub-secret/copying, and unavailable-MCP prompts with assertions for evidence, 20/10 bounds, no secret persistence, originality, separate locales, current tool boundary, technical QA, and honest status.

- [ ] **Step 4: Validate JSON syntax**

Run: `node -e "for (const f of ['skills/release-media-pack/skill.json','skills/release-media-pack/examples/materia
root-cause-remediationSkill

Mandatory for every Nomi corrective change: user-reported bugs, regressions, CI-only failures, flaky tests, performance or security defects, review/audit findings, and compatibility failures in any production path. Classify one_off versus recurring before implementation. Recurring and high-risk repairs require a schema-v3 contract, shared enforcement boundary, structural prevention, dependency lifecycle decision, and changed regression evidence.

docxSkill
brand.promoSkill

品牌宣传片 playbook。把产品文案/卖点做成一条「3 秒钩子 → 卖点 → 使用场景 → 行动号召」的短宣传片,分阶段执行(剧本审阅 → 拆镜 → 落画布 → 生成 → 排时间轴),每阶段暂停让用户审阅。

creation-editSkill

编辑创作区文档:读取当前内容,追加或替换文本,维护分镜描述格式。

director.actionSkill

非对抗性动作场景(跑酷/追逐/攀爬/特技/坠落)的动作编排知识——专业动作词汇、环境交互、动作节奏、FPV 追拍、身体力学约束、怎么把动作精确物理化地描述进视频提示词;Nomi 写动作戏 shot 时参考。

director.art-designSkill

服化道——设计「人物设定图」与「场景环境图」的生图提示词,含顶部「风格前缀块」(摄影机/胶片/调色/画幅等烧进画面的统一风格)+ 生图 avoid(多指/文字水印/穿帮等)+ identity DNA(角色跨图一致的关键特征锁定)。Nomi 为角色/场景等视觉 anchor(`carrier: visual`)生成参考图、写它的生图 prompt 时参考。

director.cinematographySkill

镜头语言与摄影技法方法论——统一景别体系/构图规则/运镜的情绪语言/打光方案/景深控制/色温光源/镜头特性,以及这些怎么翻译成视频提示词该怎么写。Nomi 拆镜头或写视频 shot 的 prompt 时参考。

director.consistencySkill

治 AI 视频多镜头「同角色换脸 / 同道具换形 / 同场景换景」三大顽疾的方法论——五维一致性检查、参考图锚定、场内状态表、段间承接。Nomi 拆镜头与跨镜生成时参考:把每个跨镜元素落成画布 anchor 节点(角色/场景/道具/风格),镜头用 anchorIds 引用、系统连参考边,段间用首尾帧承接。