Back to skills
SKILL.md
02 Accepted
ASecurity| Status | Date | Note | |--------|------|------| | proposed | 2026-05-06 | Created | | accepted | 2026-05-06 | Accepted |
- 3 stars
- 0 votes
- 0 copies
- 0 views
- Added October 18, 2026
Security analysis
100/100Pro scans all 21 files and shows the line behind each finding
npx -y skills add lythos-labs/lythoskill --skill 02-accepted --agent claude-codeAre you the author of 02 Accepted?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/lythos-labs-02-accepted)# ADR-20260506103209293: Supersede combo-skill-as-orchestration-layer — combo is now a deck-level prompt, not a separate skill
## Status History
| Status | Date | Note |
|--------|------|------|
| proposed | 2026-05-06 | Created |
| accepted | 2026-05-06 | Accepted |
## 背景
ADR-20260424114401090 设计了 `combo-` 前缀命名的专用编排 skill — 作为 orchestrator layer,类似后端的 Manager/BPMN 引擎。该设计假设 combo 需要独立的 SKILL.md + 触发条件 + 路由表 + TypeScript 代码。
2026-05-06 的 deck 重构和 fork 机制验证后,三个新事实推翻了这个假设:
1. **Fork 压缩了 combo 的存在空间**。Fork 提供真正的 remix 能力("借用心智,改成我的场景")。如果 combo 只是"几个 skill 一起用",fork 可以做到更多。
2. **Agent 本身就是自然语言驱动的**。combo 的"编排逻辑"不需要 TypeScript — agent 读到 "Search the web, then generate DOCX with diagrams" 就知道怎么做。这和 agent 读取 SKILL.md description 的机制完全一样。
3. **`max_cards` 预算模型不支持 combo 占位**。如果每个 combo 都创建一个独立 skill,它会和工具 skill 竞争卡位。Combo 是元数据("这些 skill 怎么配合"),不是能力。
**同一模式出现过**:lythoskill BDD 从 Gherkin/Cucumber 退化到 Given/When/Then 在 markdown 里。Agent 不需要框架,只需要自然语言。Combo 遵循同样的退化路径。
## 决策驱动
- Combo 的编排能力已被 fork + agent 自然语言理解覆盖
- `max_cards` 约束下,combo 不应该和 tool skill 争份额
- Deck prompt 就是外挂的 description — agent 扫描 → 匹配 → activate
## 选项
### 方案A:保持 combo- 前缀 skill(ADR-20260424114401090 原方案)
**缺点**: 每个 combo 占一个 max_cards 位;需要独立 SKILL.md + 维护;功能与 fork 重叠。
### 方案B:combo = deck 声明 + prompt(新方案)
```toml
[combo.report-generation]
skills = ["web-search", "docx", "mermaid"]
prompt = "Search for latest info, then generate professional document with diagrams"
```
**优点**:
- 零 runtime 代码,零 SKILL.md
- 不占 max_cards(元数据,不在 working set 计数)
- Agent 读取方式和读 SKILL.md description 一致
- Arena 可以 A/B test 同一个 skill 组 + 不同 prompt
## 决策
**选择**: 方案B。**Supersede ADR-20260424114401090。**
**combo 不是 skill,是 deck 的 prompt** — 和 SKILL.md description 一样的外挂元数据。Agent 扫描 → 匹配 → 按 prompt 描述的方式协调 skills。不需要代码层、不需要命名约定、不需要 package。
## 影响
- **正面**: `max_cards` 只计 tool skill,combo 是免费元数据
- **负面**: 无
- **后续**: `combo-` 前缀 skill 废弃。已有 combo skill(如 `report-generation-combo`)可转为 deck 声明
## 相关
- **Supersedes**: ADR-20260424114401090 (combo skill as orchestration layer)
- ADR-20260423101938000 (thin skill pattern)
- BDD 退化模式:Gherkin → markdown Given/When/Then(同一次简化)
Files in this skill
- ADR-20260423101938000-thin-skill-pattern.md
- ADR-20260423101950000-esm-import-fix.md
- ADR-20260423124812645-should-dist-be-committed-to-git-or-ignored.md
- ADR-20260423130348396-port-skill-manager-into-lythoskill-ecosystem-as-deck-governance.md
- ADR-20260423182606313-skill-md-template-variable-substitution-and-cli-help-delegation.md
- ADR-20260423191001406-deck-npm-package-naming-lythos-deck-vs-skill-specific-naming.md
- ADR-20260424000744041-curator-output-is-personal-environment-scan-not-project-artifact.md
- ADR-20260424013849984-lythoskill-as-anti-corruption-layer-and-meta-governance-boundary.md
- ADR-20260424113352614-project-scribe-remains-independent-with-optional-skill-cooperation.md
- ADR-20260424113917838-red-green-release-heredoc-migration-patch-design.md
- ADR-20260424115621494-virtual-evaluator-swarm-adaptive-concurrency-skill-design.md
- ADR-20260424120936541-player-deck-separation-and-deck-boundary-rationale.md
- ADR-20260424125637347-handoff-format-migration-from-fixed-file-to-daily-first.md
- ADR-20260430174746744-deck-add-command-convenience-download-without-locking-users-into-a-single-package-manager.md
- ADR-20260501090811296-ci-consistency-check-abandoned-in-favor-of-pre-commit-hook-for-skill-build.md
- ADR-20260501092809000-skills-branch-preserves-directory-prefix-to-avoid-dual-locator-standards.md
- ADR-202605011600-desc-preference-learning-arena.md
- ADR-20260501170000000-description-preference-learning-via-arena-pilot-results.md
- ADR-20260502012643244-fq-only-locator-no-bare-name-resolution.md
- ADR-20260502012643344-self-bootstrap-via-localhost-symlink-no-special-case.md
Attribution
Comments
Loading comments…
Newsletter
Get the weekly list of safe new skills, and the ones we pulled.