Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Deeply Understand (codemap) — 持久化代码图谱 + 变更影响分析 + 理解地图。接手遗留项目 Day-1 / PR 影响分析 / 解释陌生代码 / PRD·ADR 前现状理解时用。编排 tools/codemap/ 的 codemap CLI
.claude/skills/pcliangx-agf-code-map/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-10 | ✗→✓ | ▲ Improved | -28% | 0% |
| case-01 | ✗→✓ | ▲ Improved | -44% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 86% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 206% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 58% | 0% |
ADR-021 Deeply Understand 代码理解引擎的 AGF 入口。Python 原生（tools/codemap/），SQLite 持久图谱 + tree-sitter 多语言 + 反向 BFS 影响分析 + 静态 HTML dashboard。替换 agf-understand（M8 验证后删 agf-understand.js）。
codemap build + onboard + dashboardcodemap diff（→ code-reviewer 审查清单）codemap explaincodemap understand（接管 agf-understand 的理解地图）uv run codemap <subcmd>）build [path] — 全量建图 → .agf/code-map.dbupdate — 增量（git diff + 双指纹）diff [--base <ref>] [--hop N] — 变更影响分析（反向 import BFS + 风险评分）explain <target> — 节点深度解释（邻居 + 源码片段）onboard — 项目概览 + 复杂度热点 + 高扇入understand [topic] — 理解地图（目录子系统 + 核心依赖 + 风险点）context "<query>" — 相关子图（给 agent 注入）search "<query>" — 检索（默认 FTS5 关键词；跑过 embed 后自动启用 semantic 余弦，三态降级）embed — 生成节点 embedding 存 .agf/code-map.db（需 uv sync --extra semantic；jina-code-v2，首次下载 ~90MB + torch；未装/未跑则 search 自动降级 FTS，功能不中断）dashboard [--out <path>] — 静态 HTML 图（cytoscape，浏览器 open）首发 7 语言：Python / TS / JS / SQL / YAML / JSON / Java（04 设计）。Swift/WXML 诚实缺口（04 §6）。
docs/specs/ / docs/adr/ / docs/design/DESIGN.md / OpenAPI 契约.agf/* gitignored（派生缓存，可重建）；消费层报告落 docs/reviews/| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-10 | fail→pass | 11,857 | 3,864 | -67% | 1 | 1 | 0% | 1,875 | 1,353 | -28% | 0 | 0 | — |
case-01 | fail→pass | 18,199 | 6,048 | -67% | 1 | 1 | 0% | 2,831 | 1,577 | -44% | 0 | 0 | — |
case-02 | fail→fail | 10,094 | 4,149 | -59% | 1 | 1 | 0% | 1,558 | 1,167 | -25% | 0 | 0 | — |
case-03 | fail→pass | 4,591 | 4,449 | -3% | 1 | 1 | 0% | 738 | 1,369 | +86% | 0 | 0 | — |
case-04 | fail→pass | 3,643 | 7,242 | +99% | 1 | 1 | 0% | 566 | 1,734 | +206% | 0 | 0 | — |
case-05 | fail→pass | 5,845 | 3,942 | -33% | 1 | 1 | 0% | 842 | 1,332 | +58% | 0 | 0 | — |
case-06 | fail→fail | 2,772 | 5,401 | +95% | 1 | 1 | 0% | 264 | 889 | +237% | 0 | 0 | — |
case-07 | fail→pass | 14,353 | 2,297 | -84% | 1 | 1 | 0% | 2,514 | 1,046 | -58% | 0 | 0 | — |
case-08 | pass→pass | 17,800 | 1,886 | -89% | 1 | 1 | 0% | 2,852 | 1,002 | -65% | 0 | 0 | — |
case-09 | fail→pass | 25,143 | 2,692 | -89% | 1 | 1 | 0% | 4,065 | 1,055 | -74% | 0 | 0 | — |
case-11 | fail→pass | 18,750 | 7,529 | -60% | 1 | 1 | 0% | 2,905 | 1,905 | -34% | 0 | 0 | — |
case-12 | pass→pass | 15,777 | 2,819 | -82% | 1 | 1 | 0% | 2,489 | 1,168 | -53% | 0 | 0 | — |
case-13 | fail→pass | 14,009 | 3,631 | -74% | 1 | 1 | 0% | 1,994 | 1,234 | -38% | 0 | 0 | — |
case-14 | fail→pass | 21,991 | 4,782 | -78% | 1 | 1 | 0% | 4,181 | 1,384 | -67% | 0 | 0 | — |
case-15 | fail→pass | 10,831 | 4,327 | -60% | 1 | 1 | 0% | 1,692 | 1,424 | -16% | 0 | 0 | — |
case-16 | fail→pass | 13,482 | 2,899 | -78% | 1 | 1 | 0% | 2,055 | 1,103 | -46% | 0 | 0 | — |
case-17 | fail→pass | 9,958 | 10,643 | +7% | 1 | 1 | 0% | 1,525 | 2,478 | +62% | 0 | 0 | — |
case-18 | fail→pass | 9,904 | 2,807 | -72% | 1 | 1 | 0% | 1,289 | 1,083 | -16% | 0 | 0 | — |
case-19 | fail→pass | 10,185 | 2,961 | -71% | 1 | 1 | 0% | 1,458 | 1,165 | -20% | 0 | 0 | — |
case-20 | pass→pass | 20,143 | 11,392 | -43% | 1 | 1 | 0% | 3,041 | 2,433 | -20% | 0 | 0 | — |
case-21 | fail→fail | 8,807 | 2,292 | -74% | 1 | 1 | 0% | 1,166 | 1,034 | -11% | 0 | 0 | — |
case-22 | fail→pass | 12,105 | 2,425 | -80% | 1 | 1 | 0% | 1,752 | 1,070 | -39% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 21 counted toward the lift figure. The other 1 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +73 percentage points is the difference between those two pass rates over the 21 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.