Audit Coverage

Audit Coverage

Check MVP component, platform mechanism, and advanced mechanism coverage for analyzed agents. Identifies which demos exist, which are missing, and recommends next steps. Supports both coding agents and platform agents.

Trigger

/audit-coverage [agent-name]

• With agent-name: audit that specific agent
• Without arguments: audit all agents with status in-progress or done

Prerequisites

1. The agent must have an analysis doc at docs/<agent-name>.md
2. Ideally the doc includes section 7.5 (MVP Component Map); if not, components are inferred from D2-D6

Workflow

Step 1: Read Analysis Doc

Read docs/<agent-name>.md and extract:

• D2-D7 findings (architecture, tools, prompt, context, error handling, innovations)
• D7.5 MVP component table (if present)
• D9-D12 platform dimensions (if present — indicates a platform agent)

Step 2: Determine MVP Components

• If 7.5 section exists: use the table directly as the MVP component list
• If 7.5 section is missing: infer MVP components from D2-D6:
  1. 主循环 (from D2)
  2. Tool 注册与分发 (from D3, if applicable)
  3. Prompt 组装 (from D4)
  4. LLM 调用与响应解析 (from D2/D6)
  5. 编辑应用 (from D3, if applicable) Adjust based on agent architecture.

Step 2.5: Determine Platform Mechanisms (platform agents only)

If D9-D12 sections exist and contain content (not "不适用"):

• Extract platform mechanisms worth reproducing from D9-D12
• Check agents.yaml for type: agent-platform confirmation

Step 3: Check Demo Coverage

Read demos/<agent-name>/README.md (overview):

• Parse MVP section: which components have [x] (done) vs [ ] (missing)
• Parse 平台机制 section (if present): which mechanisms have [x] vs [ ]
• Parse 进阶 section: which mechanisms have [x] vs [ ]
• Check if mini-<agent> directory exists (complete integration)

If overview uses old format (flat list without MVP/进阶 sections):

• Map existing demos to MVP vs 进阶 categories
• Note that overview needs migration

Step 4: Check Document Freshness

Read agents.yaml for analyzed_date:

• If more than 30 days old → warn and suggest running /check-updates first
• If missing → warn about no timestamp

Step 5: Generate Report

Output a structured report:

## <agent> — Coverage Audit

### MVP 组件覆盖
| 组件 | Demo 状态 | 建议语言 | 建议操作 |
|------|-----------|----------|----------|
| core-loop | ❌ 缺失 | Python | /create-demo <agent> core-loop |
| edit-apply | ✅ 已覆盖 (search-replace) | — | — |

### 平台机制覆盖 _(仅平台型 agent)_
| 机制 | 来源维度 | 状态 | 建议操作 |
|------|----------|------|----------|
| channel-router | D9 | ❌ | /create-demo <agent> channel-router |
| memory-retrieval | D10 | ❌ | /create-demo <agent> memory-retrieval |

### 进阶机制覆盖
| 机制 | 状态 |
|------|------|
| repomap | ✅ |
| fuzzy-match | ❌ |

### 完整串联
- mini-<agent>: ❌ 未创建（需先完成所有 MVP 组件 + 平台机制）

### 文档状态
- analyzed_date: YYYY-MM-DD (N 天前)
- 7.5 节: ✅ 存在 / ❌ 缺失（建议补充）
- D9-D12: ✅ 已填写 / ⬜ 不适用（纯 coding agent）

### 总结
- MVP: 2/5 (40%) | 平台: 0/3 (0%) | 进阶: 4/8 (50%) | 串联: 0/1 | 总计: 6/17
- 下一步: 创建 core-loop demo

Relationship with /check-updates

These two skills are complementary:

• /audit-coverage → 查覆盖缺口（有什么没有什么）
• /check-updates → 查内容时效（现有的还准不准）

When /audit-coverage discovers freshness issues (analyzed_date > 30 days), it recommends running /check-updates first.