Generates a ranked, citation-backed focus map of load-bearing symbols in a codebase.
npx skills add https://github.com/Ar9av/obsidian-wiki --skill code-understand
Please help me install this Agent Skill. Skill: code-understand Source: https://github.com/Ar9av/obsidian-wiki/tree/main/.skills/code-understand Command: npx skills add https://github.com/Ar9av/obsidian-wiki --skill code-understand
When working in an unfamiliar or large codebase, developers waste time grepping and reading files at random to figure out which files and symbols actually matter before making a change. This Skill produces a ranked focus map of the symbols an architecture hangs on, each with a file:line citation, so you read only what matters.
Ask your agent to map out this project's architecture and show which files are load-bearing before making changes.
High-intent search queries and answers about installing and using this skill.
Run obsidian-wiki code-understand --project <dir> to get a ranked focus map of load-bearing symbols with file:line citations. Open only the cited locations instead of reading the whole tree, then form your opinion from the actual source.
Use the --changed flag with specific files or --since with a git ref like HEAD~5 to seed the focus map from your diff. The output ranks symbols connected to those changes so you can see what would break before finishing the edit.
Yes, the auto and builtin backends work with zero setup using a regex AST extractor plus ripgrep cross-file references. CodeGraph adds real call-graph edges, but rg-reference entries from the builtin backend are text matches that should be confirmed by opening the file.
The CodeGraph backend produces structural facts from real call-graph edges that can be cited directly. The builtin backend's defines, imports, and changed-file entries are facts, but its rg-reference entries are text-match evidence only and need manual confirmation.
Do not use it to summarize what a project does for a wiki; that is handled by wiki-update Step 3b which calls the same extractor internally. Also treat the output as a disposable cache, never committing it to docs or PR descriptions.
hamsurangQuery GitHub repository wiki documentation for architecture and API summaries.
kangig94Propose and create structured knowledge pages in an Obsidian LLM-Wiki.
vTRKASearch codebases via CodeGraph queries for symbols, callers, and dependencies.
JerryGanstSearch code structure with tree-sitter AST parsing to locate functions and symbols.
