菜单
Analyze a codebase and build a world-model — traces behaviors, records entities with evidence, creates boundaries by concern. Use when the user says analyze, understand, or map a codebase or specific area.
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
Run the self-serve verification loop — boot the current source stack against a golden fixture, assert every layer (MCP write → store → HTTP API → projection → rendered browser UI), and inspect the map visually via headless Playwright. Use when you changed Cartographer's service, store, projection, or UI and need to verify it yourself end-to-end without a human watching the browser.
Deep autonomous codebase exploration — traces behaviors end-to-end, creates boundaries by concern, builds navigable depth with sub-boundaries, enforces quality via cartographer_check_depth. Use when the user says explore, map, understand, or analyze deeply.
Visualize a pull request on the Cartographer map — see which entities are affected, the blast radius, and the review path. Use when user says "review pr", "review pull request", or "/cartographer review-pr".
Query the Cartographer world-model to answer questions about a codebase using stored evidence. Use when the user asks how something works, what calls what, or wants to understand structure.
| name | analyze |
| description | Analyze a codebase and build a world-model — traces behaviors, records entities with evidence, creates boundaries by concern. Use when the user says analyze, understand, or map a codebase or specific area. |
| model | opus |
| effort | high |
Verify tools: call cartographer_get_summary. If it fails, stop — MCP
tools are not connected.
Set the project root: cartographer_set_project.
If analyzing a specific concern (e.g., "analyze auth"), create a perspective
first with cartographer_create_perspective — it auto-switches, so entities
you write next join it. No separate cartographer_switch_perspective needed.
Behaviors first. Identify the most important things this system does. Trace each behavior from entry point to completion. Record every entity encountered along the flow. Record the path as a behavior slice.
Boundaries emerge from behaviors. Cluster entities that participate in the same flows. Name by concern, not by directory.
Evidence is mandatory. Every cartographer_write_entity and
cartographer_write_relationship call must include source anchors
(filePath, lineStart, lineEnd, snippet). Set confidence: proven for
direct observations, lower for inferences. Include reasoning for
anything below proven.
Descriptions are for everyone. Name entities by what they do, not how they're implemented. Behavior flows read like stories. Technical detail belongs in evidence anchors, not descriptions.
Call cartographer_get_summary and report what was found. Open the map
with cartographer_open_map.
