菜单
Use when the user asks what GitNexus is, when to use it (vs sem), what the web UI offers, or how the index/embeddings work. Conceptual reference for the manual human GitNexus setup. Examples: "what is gitnexus for?", "should I use gitnexus or sem?", "what can the gitnexus UI show me?".
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。
Use when the user wants to run GitNexus to explore or document a codebase: build/refresh its index, open the web UI, generate a wiki, or check status. GitNexus is a manual human tool here — help the user run the `gnx` commands. Examples: "index this repo with gitnexus", "open the gitnexus UI", "generate a wiki for this project", "gnx".
Expert in Claude prompting, skill creation, hooks management, MCP configuration, sub-agents, memory management, and Claude Desktop app features. Use when writing prompts, creating Claude Code skills, configuring hooks, setting up MCP servers, creating custom sub-agents, managing memory (auto memory, CLAUDE.md, rules), asking about Claude Code architecture, or discussing Claude Desktop features (scheduled tasks, cowork, etc.).
Analyzes past Claude Code conversations to roast your prompting habits and compute efficiency. Reads user prompts, cross-references with tool errors and corrections, analyzes model/reasoning choices (Fable vs Opus vs Sonnet vs Haiku), then generates dual scores (prompt quality + compute efficiency), worst habits, techniques, and a personalized model selection cheat sheet. Tracks scores over time so you can see improvement. Use when you want honest feedback on your prompting skills.
Analyzes past Claude Code conversation logs to learn from mistakes. Extracts tool errors, user corrections, and repeated failures, then writes actionable learnings to auto memory. Use when you want Claude to stop repeating the same mistakes across sessions.
Interview the user relentlessly about a plan or design until reaching shared understanding, resolving each branch of the decision tree one question at a time. Use proactively while in plan mode and before finalizing or presenting any implementation plan, design, or architecture decision — and whenever the user wants to stress-test a plan, get grilled on their design, or says "grill me".
Generates personalized learning plans for any topic (tech or non-tech). Assesses current knowledge, researches official documentation and resources, creates structured learning paths with theory and practice modules. Stores the plan in Obsidian for interactive follow-up with the learning-tutor agent. Use when: learn, study plan, learning path, curriculum, teach me, how to learn, formation, apprendre, study, training.
| name | gitnexus-guide |
| description | Use when the user asks what GitNexus is, when to use it (vs sem), what the web UI offers, or how the index/embeddings work. Conceptual reference for the manual human GitNexus setup. Examples: "what is gitnexus for?", "should I use gitnexus or sem?", "what can the gitnexus UI show me?". |
GitNexus indexes a repository into a knowledge graph (symbols, call edges, execution
flows, Leiden clusters) and surfaces it through a web UI, a wiki generator, and
architecture maps. In this environment it is a manual human tool: there is no MCP
server, no autonomous agent use, and no "must run before editing" mandates. To run it, see
the gitnexus-cli skill (the gnx commands).
sem| Need | Use | Why |
|---|---|---|
| "What does this change break? What depends on this symbol?" | sem | Always-fresh (no index), zero-footprint, git-native, already in the pre-commit hook. |
| "Help me understand / document this codebase — search flows, browse the call graph, generate a wiki." | GitNexus (gnx) | Whole-repo graph, execution-flow tracing, a browsable UI, and wiki/architecture output sem doesn't have. |
Rule of thumb: sem for per-change impact (agentic, fast), GitNexus for whole-system
exploration (manual, visual).
gnx serve → http://127.0.0.1:4747)The underlying gitnexus binary still exposes graph queries directly, if you prefer the
terminal over the web UI:
gitnexus query "auth flow" # execution flows related to a concept
gitnexus context <symbolName> # 360° view of a symbol
gitnexus impact <symbolName> # blast radius of a symbol
gitnexus cypher '<query>' # raw graph query (see graph schema below)
<repo>/.gitnexus/lbug (LadybugDB: graph + full-text
search + vector embeddings), gitignored. Plus parse caches for fast re-index.~/.gitnexus/registry.json (which repos are indexed) and
~/.config/gitnexus/config.json (settings). No global embedding store.gnx index <path> before
browsing (idempotent; skips repos already current with git HEAD).gnx index handles that automatically.VECTOR extension installs and loads fine, but
CREATE_VECTOR_INDEX(... metric := 'cosine') throws an opaque native error, so semantic
search always falls back to exact-scan (brute-force cosine, full accuracy). Exact-scan
is capped at 10k chunks by default; gnx raises it to 50k via
GITNEXUS_SEMANTIC_EXACT_SCAN_LIMIT so even the largest repos (~20-30k symbols) get
complete coverage. doctor shows VECTOR index: available / Semantic mode: vector-index,
but that only means the platform is supported — not that HNSW actually builds. Revisit
HNSW on a future GitNexus release; until then exact-scan is the working path.gitnexus cypher)Nodes: File, Function, Class, Interface, Method, Community, Process
Edges (via CodeRelation.type): CALLS, IMPORTS, EXTENDS, IMPLEMENTS, DEFINES, MEMBER_OF, STEP_IN_PROCESS
MATCH (caller)-[:CodeRelation {type: 'CALLS'}]->(f:Function {name: "myFunc"})
RETURN caller.name, caller.filePath