code-review-graph
Use code-review-graph for local-first CLI/MCP code knowledge graphs, graph-backed code review, structural search, and repository-analysis integrations.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- code-review-graph
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Use code-review-graph for local-first CLI/MCP code knowledge graphs, graph-backed code review, structural search, and repository-analysis integrations.
- disable-model-invocation
- true
02 资源文件 15
- 参考文档references/architecture-and-schema.md482 B
- 参考文档references/cli-reference.md2.7 KB
- 参考文档references/custom-languages.md473 B
- 参考文档references/embeddings.md468 B
- 参考文档references/eval.md389 B
- 参考文档references/github-action.md433 B
- 参考文档references/mcp-tools.md2.4 KB
- 参考文档references/registry-and-daemon.md655 B
- 参考文档references/repo-provenance.md2.3 KB
- 参考文档references/repo-routing-metadata.json423 B
- 参考文档references/review-workflows.md982 B
- 参考文档references/search-and-analysis.md760 B
- 参考文档references/start-here.md2.1 KB
- 参考文档references/troubleshooting.md1.7 KB
- 脚本scripts/check_crg_install.py3.7 KB
03 SKILL.md 正文
code-review-graph
Use this repo skill when a task names code-review-graph, CRG, graph-backed code review, local MCP graph tools, blast-radius analysis, structural code search, risk-scored PR review, custom language parsing, or multi-repo code graph operations.
code-review-graph is a Python package and CLI/MCP server that builds a local SQLite code knowledge graph from Tree-sitter parses, updates it incrementally, and exposes token-efficient review/search workflows to coding agents.
Fast route
- For first-time setup, read references/start-here.md, then route to
install-and-setup. - For changed-code review or PR review, route to
review-changes. - For graph queries, architecture, flows, communities, or refactors, route to
graph-exploration. - For optional custom languages, embeddings, wiki, registry/daemon, GitHub Action, or eval, route to
integrations-and-extensions.
Minimal install and smoke check
pip install code-review-graph
code-review-graph install
code-review-graph build
code-review-graph status
For an existing install, the bundled scripts/check_crg_install.py verifies import, package version, CLI discovery, and packaged docs without creating or updating a graph.
Sub-skill routes
| Sub-skill | Read when |
|---|---|
| install-and-setup | Installing CRG, configuring MCP clients, building/updating the graph, checking status, running the MCP server, visualizing, watching, or uninstalling. |
| review-changes | Reviewing a diff or PR, computing blast radius, risk scoring, test gaps, token savings, affected flows, or rendering a PR review comment. |
| graph-exploration | Finding callers/callees/tests/imports, searching graph nodes, inspecting flows/communities/architecture, finding large functions, or previewing refactors. |
| integrations-and-extensions | Custom languages, embeddings/providers, wiki generation, multi-repo registry, daemon watch workflows, GitHub Action setup, or eval/benchmark reproduction. |
Shared references
- references/cli-reference.md maps common CLI commands and flags.
- references/mcp-tools.md maps MCP tools and token-discipline rules.
- references/troubleshooting.md covers cross-cutting install, graph freshness, optional dependency, provider, and SQLite issues.
- references/architecture-and-schema.md summarizes graph node/edge semantics.
- references/repo-provenance.md records the source snapshot used to create this skill and when to refresh it.
- references/repo-routing-metadata.json is structured metadata for managed repo-skill routing.
Operating principles
- Prefer graph tools over broad file reads when the user asks a structural question.
- Start with compact context (
get_minimal_context_tool) when an MCP session is available. - Use
detail_level="minimal"unless the next step requires source snippets or full graph details. - Keep optional extras optional; do not install embeddings, communities, wiki, eval, enrichment, or
allunless the task needs them. - Treat cloud embeddings and external services as opt-in because code-derived text may leave the machine.
- Rebuild or update the graph before trusting review/search answers after a branch switch, rebase, or large edit.
Boundaries and non-fits
Do not use this skill as a general-purpose LSP replacement, full source-code reading strategy, or benchmark runner unless the user explicitly asks for CRG workflows. This skill covers the Python CLI/MCP package and public integrations, not detailed development of the separate VS Code extension package.
Verification expectations
This runtime skill is grounded by package metadata, public docs, installed-package inspection, existing repo-local CRG skills, and native test candidates spanning install, CLI, MCP, review, search, flows, communities, refactor, optional integrations, and CI workflow safety. Final verification artifacts live outside this runtime skill under the review/test artifact directory.
04 同仓库更多 Skills
passnet-triton-opt
Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.
passnet-solve
END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.
passnet-skill
PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.
passnet-pattern-fusion
Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.
passnet-orchestrate
Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.
passnet-feedback
Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).