lum1104/understand-anything
8 skills6.5k installs439k starsGitHub
Install
npx skills add https://github.com/lum1104/understand-anythingSkills in this repo
1UnderstandUnderstand is part of the understand-anything agent skill family that helps solo builders and small teams build durable structural knowledge of a repository so coding agents do not re-explore from scratch every session. The documented build-fingerprints step replaces a broken inline SKILL.md example that never produced a baseline and silently broke auto-update (issue #152). During a full /understand rebuild, the skill invokes a Node script with your project root, selected source paths, and current git hash, then materializes fingerprints.json under .understand-anything for downstream incremental updates. That baseline powers change detection when the codebase evolves, which matters for indie projects where you iterate fast across frontend, backend, and scripts. Use it when onboarding to an unfamiliar repo, refreshing docs after a large refactor, or keeping agent context aligned with git reality.1.1kinstalls2Understand Dashboardunderstand-dashboard is a slash-style agent skill that starts the Understand Anything interactive web UI for whichever project directory you pass—or the current working directory when you omit a path. Before launch it verifies that `.understand-anything/knowledge-graph.json` exists; if not, it tells you to run /understand first so the graph is materialized. The skill then locates dashboard code under `packages/dashboard/` using a ordered search: Claude plugin root, a universal symlink under ~/.understand-anything-plugin, self-relative paths from installed skill dirs, and common clone locations for Codex, OpenCode, and Pi. Solo builders benefit when onboarding to a unfamiliar monorepo or revisiting architecture after a break—you get a browser view of entities and relationships instead of ad-hoc file tree spelunking. It complements analysis skills rather than replacing them; no graph file means no dashboard. Pair with your agent’s long-running dev server habits and keep the project path explicit in multi-root workspaces.852installs3Understand Explainunderstand-explain is an agent skill for solo builders who inherit or grow a codebase and need trustworthy explanations of one component at a time. It assumes a pre-built knowledge graph JSON with rich node metadata—summaries, tags, complexity—and typed edges for imports, calls, configuration, and documentation links, plus optional layers and a narrative tour. The workflow tells the agent to search the graph with Grep for relevant IDs and summaries instead of dumping the entire graph into context, then synthesize a thorough explanation of the requested file, function, or module. That makes it valuable when onboarding to your own older code, reviewing a partner’s module before changing it, or tracing how a service connects to tables and endpoints. It is not a substitute for generating the graph itself; you run it when the graph already exists and you want human-grade clarity on a narrow slice of the system.824installs4Understand Onboardunderstand-onboard is an agent skill for solo builders and small teams who already maintain a structured knowledge graph of their repository and need newcomer-ready documentation without rewriting the same architecture story by hand. Invoke it when a teammate, contractor, or future-you must ramp on imports, services, configs, and documented flows without spelunking every folder. The skill expects JSON with project metadata, typed nodes with summaries and tags, relationship edges, optional layers, and a tour sequence—then applies a read-efficient workflow (search first, load only needed slices) so the agent does not dump the entire graph into context. It fits the Build phase as living docs, Operate when the graph is refreshed after refactors, and Grow/Support when onboarding scales beyond the original author. Pair it with whatever skill in your stack produces or updates the graph; this command focuses on packaging that intelligence into an onboarding guide.819installs5Understand ChatUnderstand Chat is a journey-wide agent skill for solo builders who inherit a repo, fork a template, or return to a side project and need trustworthy answers grounded in a structured map of the system. It expects an analyzed project with a knowledge graph at `.understand-anything/knowledge-graph.json`, including project metadata, typed nodes with summaries and tags, weighted edges, optional layers, and a guided tour. The workflow prioritizes efficient reads: search within the JSON for relevant node IDs and summaries before pulling large sections, then explain behavior, dependencies, and documentation links in plain language. Use it whenever you must scope a feature, debug a flow, or brief an agent without manually opening dozens of files. It pairs naturally with upstream graph-generation tooling in the Understand Anything family and keeps procedural knowledge in SKILL.md so Claude Code, Cursor, or Codex can reuse the same graph-reading discipline every session.809installs6Understand Diffunderstand-diff is an agent skill that connects raw git changes to a structured project knowledge graph stored at `.understand-anything/knowledge-graph.json`. Solo builders use it when a diff or pull request is too large to reason about from filenames alone: the skill explains how to read nodes (code, config, services, domains, articles) and weighted edges so you can trace imports, calls, deployments, and documentation links to what actually moved. It emphasizes efficient context use—search the graph with Grep and pull only the sections you need rather than loading the entire JSON. The workflow fits pre-merge review, release notes prep, and post-incident diffs when you already maintain an Understand Anything graph from prior analysis. It does not replace human judgment on security or product scope; it augments review with graph-grounded impact hints. Install alongside the broader understand-anything setup so the graph exists before invoking diff analysis.752installs7Understand DomainUnderstand-domain is an agent skill packaged with a Python codebase scanner that walks a project root and writes structured domain context for downstream analysis. Solo builders use it when they inherit or fork a large app and need a grounded picture of modules, entry points, and representative source before planning changes or writing docs. The script enforces practical limits on depth, file count, and output size so results fit typical agent context windows. It targets source extensions across TypeScript, Python, Go, Rust, Java, and more while ignoring dependency and build artifact folders. The workflow pairs mechanical extraction with agent-side interpretation to name domains and flows rather than guessing from README alone. It is intermediate complexity because you must run the extractor from the correct project root and interpret JSON alongside the repo. Best after clone or before major refactors; not a substitute for runtime debugging or product discovery interviews.685installs8Understand KnowledgeUnderstand-knowledge is the merge step in the understand-anything pipeline for solo builders who want a navigable knowledge graph instead of scattered markdown and batch JSON. After deterministic scanning and LLM batch analysis, this skill combines manifests and batches, deduplicates entities, normalizes edges against shared type enums, and assembles layers and guided tours aligned with index.md structure. It is aimed at indie developers documenting codebases, research wikis, or agent-readable corpora who already run the upstream scan and analysis steps. Use it when intermediate batch files exist and you need a single assembled-graph.json for downstream tooling or browsing. It matters because it turns partial analyses into one consistent graph without hand-merging JSON or breaking type contracts shared with core TypeScript types.653installs