Graphify architecture review
Skill rarestg/rarestg-skills/skills/graphify-architecture-review
Run reusable Graphify-led architecture analysis for codebases using semantic graphs, optional subagent extraction, graph synthesis, source-search validation, graph-shape review, and follow-up refactor planning. Use when asked to analyze repo architecture, god nodes, surprising edges, topology, module boundaries, or graph-derived cleanup/refactor opportunities.From its SKILL.md
npx -y skills add rarestg/rarestg-skills --skill graphify-architecture-reviewAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
- 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
5.1 KB, as published. Nobody here has run it
Graphify-Led Architecture Review
Use this skill to turn a codebase graph into source-validated architecture insights and concrete refactor plans.
Read references/SOP.md before running the workflow. It is the detailed
operating model for this skill.
Procedure
- Read repo guidance, architecture docs, planning docs, and search guidance.
- Record repo path, branch, commit SHA, date/time, Graphify version, included paths, and excluded paths.
- Identify generated, vendor, asset, raw-log, build, and binary paths to exclude or downrank.
- Ensure
graphify-out/and.graphify_*are ignored unless the user wants generated graph artifacts versioned. - Run Graphify on the repo or scoped subtree.
- Generate and inspect
GRAPH_REPORT.md,graph.json, andgraph.html. - Label communities in plain domain language.
- Filter or discount generic nodes before interpreting architecture.
- Package first-pass findings into concise review docs.
- Validate graph claims with source search and file reads.
- Dispatch investigation agents for god nodes, surprising edges, generated questions, and topology anomalies when available.
- Dispatch synthesis agents to identify top refactor opportunities from the curated docs when available.
- Write plans only for source-validated, high-leverage opportunities.
- Ask fresh agents to review plans before finalizing when available.
Subagents
Use subagents or parallel investigators for extraction chunks, investigation, synthesis, and plan review when available. Give each one exact files or docs, one bounded question, and a required output shape.
Do not delegate final acceptance. Validate important conclusions locally against source.
Noise Filtering
Before judging god nodes, bridges, or topology, filter or discount generic labels such as:
.new().default().drop().clone().from().into()
Prefer normalization over deletion. Qualify methods by receiver/type when
possible, remove unresolved generic nodes and incident inferred edges, preserve
domain-specific short methods, and treat main(), run(), .load(),
.open(), .save(), .build(), and .parse() as suspicious but not
automatically noisy.
Insight Extraction
Process GRAPH_REPORT.md in this order:
- Corpus summary and graph size
- Community hubs and labels
- God nodes
- Surprising connections
- Hyperedges
- Ambiguous edges
- Knowledge gaps and thin communities
- Suggested questions
For each important finding, record graph evidence, source evidence, current
behavior, why the connection exists, verdict, and recommended action. Use
verdicts: good, neutral, bad, artifact, or watchpoint.
Use Graphify for discovery:
graphify explain "<node>" --graph graphify-out/graph.json
graphify path "<node-a>" "<node-b>" --graph graphify-out/graph.json
graphify query "<question>" --graph graphify-out/graph.json --budget 2000
graphify query "<question>" --dfs --graph graphify-out/graph.json --budget 2000
Use source search for proof. Rewrite graph output as observable source behavior and search the likely owner path before accepting a graph claim.
Output Artifacts
Create concise review docs using the repo's preferred docs location, or an ignored scratch folder until the user chooses a durable location:
index.mdoverview.mdgod-nodes.mdsurprising-connections.mdquestion-answers.mdgraph-shape-findings.md- one refactor plan per validated opportunity
Keep generated graph artifacts uncommitted unless explicitly requested.
Graph Shape
Treat graph topology as triage, not proof. Do not trust visual position unless
coordinates are persisted. Filter contains edges and generic labels first.
Validate stragglers, bridge nodes, and community boundaries in source.
Classify topology findings as real boundary, watchpoint, fixture noise, file-node artifact, generic-label artifact, or not actionable.
Quality Gates
- Graph has nonzero nodes and edges.
- Expected output files exist.
- Chunk failures are below threshold.
- Generated artifacts are ignored or intentionally versioned.
- Strong claims have graph and source evidence.
- Generic-node artifacts are filtered or explicitly classified.
- Refactor plans include acceptance criteria and verification commands.
- Tracked docs pass lint.
Stop Conditions
Stop when graph findings are classified, high-value questions are source-validated, weak topology findings are downranked or promoted with evidence, strong opportunities are captured as plans, and no new validated opportunity outranks the current plan set.