菜单
Writes field-map.md, the flagship overview article that connects all taxonomy areas into one system. A 3000-5000 word systems map showing integration points, paradigm routing logic, and a concrete practitioner flow. Use when writing the field map or systems overview.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Show pending changes and LLM-powered impact analysis without compiling. Detects new/modified/deleted sources via content hashing, maps them to dirty buckets, reads changed sources to assess impact and detect contradictions with existing claims. Use to preview what incremental compilation would do before running it.
Incrementally recompiles changed parts of the wiki after new sources are ingested. The skill handles change detection, LLM-powered impact analysis, and user confirmation. All compilation is delegated to the script pipeline (bun run compile --incremental), which handles entity extraction, resolution with anchoring, synthesis with evidence registry, reference cards, claims, self-eval, and state saving.
Extracts atomic verifiable claims from synthesis articles and runs self-evaluation by verifying a sample against raw sources. Outputs build/claims.json and build/eval-report.json. Use when extracting claims, running self-eval, or verifying wiki accuracy.
Compiles the wiki from raw markdown sources. Read `config/domain.ts` for the domain topic and taxonomy. Orchestrates a multi-phase pipeline using subagents: read sources, write synthesis articles, write reference cards, generate field map and indexes, extract claims, and run self-eval. Use when asked to compile, build, or generate the wiki from raw sources.
Deep-researches a project or paper for the knowledge base. Clones repos to read source code, fetches full paper text, and produces structured analysis files in raw/deep/. Use when asked to deep-research, deep-dive, analyze source code, or enrich a source with implementation details.
Writes a single synthesis article (3000-5000 words) for one taxonomy bucket. Produces landscape analysis with approach categories, convergence, failure modes, and selection guide. Use when writing a synthesis article for a bucket.
| name | compile-field-map |
| description | Writes field-map.md, the flagship overview article that connects all taxonomy areas into one system. A 3000-5000 word systems map showing integration points, paradigm routing logic, and a concrete practitioner flow. Use when writing the field map or systems overview. |
Read config/domain.ts for bucket definitions, the fieldMapTitle, stack order, and crossCuttingThemes.
Write wiki/field-map.md, a systems map that treats the taxonomy buckets as
layers of one stack, not a taxonomy. Use the fieldMapTitle from config/domain.ts as the H1.
Read the synthesis article for each bucket defined in config/domain.ts (one file per bucket at wiki/{bucket-id}.md).
Name the central engineering problem each bucket solves. Link to the synthesis
article. Frame them as layers of a single stack in the order defined in config/domain.ts (first bucket is bottom, last is top).
For each pair of adjacent layers: how does layer N feed layer N+1? What breaks when the integration is missing? Name specific projects that bridge each gap.
Routing logic for retrieval approaches: when to use vector, graph, or file-based retrieval. Name the tradeoff each optimizes for (latency, temporal validity, simplicity). Include decision criteria a practitioner can apply in under 60 seconds.
Two columns: production-ready and research-grade. Name specific projects in each column. What distinguishes them (docs, adoption, API stability)?
One major assumption that practitioners held 12 months ago and have since corrected. Provide evidence from sources. Name what replaced it.
Concrete end-to-end trace: a developer building X uses tool A for step 1, tool B for step 2, etc. Name specific tools at each step. Walk through one realistic workflow that touches all 6 layers.
Use the crossCuttingThemes from config/domain.ts. One paragraph per theme. Cite specific examples.
What to read next based on what the reader is building. 4-5 personas with a recommended path through the wiki articles.
Include a Mermaid graph LR diagram showing the 6-layer stack, not individual projects. Each node is a bucket labeled with its central engineering problem. Edges show integration points between adjacent layers. Include a dotted feedback loop from the top layer back to the bottom.
Do NOT generate a project-level graph (random projects with sparse edges are meaningless). The diagram represents the article's thesis: these areas form one stack.
---
title: "<fieldMapTitle from config/domain.ts>"
type: field-map
last_compiled: "YYYY-MM-DD"
---
Read and follow all rules in ../stop-slop/SKILL.md. No filler, no passive voice, no em dashes, no AI tells.
Every claim cites a source: [Source](../raw/type/file.md). Every project
links to its reference card on first mention: [Mem0](projects/mem0.md).