code-graph
by alinaqicode-graph is an AST-based code graph skill for fast symbol lookup, dependency analysis, and blast-radius checks via codebase-memory-mcp. Use it for code editing when you want the graph first and file reads second.
This skill scores 68/100, which means it is worth listing for users who want graph-first code navigation, but they should expect some operational gaps and moderate setup ambiguity. The repository clearly explains the intended workflow and when to use it, yet it lacks install-time guidance and supporting assets that would make adoption more frictionless.
- Strong workflow intent: 'graph first, file second' with explicit use cases like symbol lookup, dependency analysis, and blast radius
- Good triggerability from metadata and frontmatter, including a clear when-to-use instruction and a non-user-invocable flag
- Substantive skill content with headings, code fences, and repo/file references that suggest real operational guidance rather than a placeholder
- No install command, support files, or companion references, so users may need to infer setup and MCP integration steps
- Contains placeholder markers ('todo'), which lowers trust that all edge cases and workflows are fully finished
Overview of code-graph skill
What code-graph does
The code-graph skill helps Claude use an AST-based code graph through codebase-memory-mcp to find symbols, trace dependencies, and estimate blast radius before opening files. It is best for code editing work where fast structure-aware navigation matters more than raw text search.
Who should install it
Install code-graph if you regularly make changes in medium or large codebases, debug cross-file behavior, or need a better way to answer “what depends on this?” before editing. It is especially useful when grep alone is too noisy and you want the code-graph skill to reduce guesswork.
Why it is different
The main advantage of code-graph for Code Editing is that it pushes the agent to query structure first and read files second. That makes it stronger for symbol lookup, call tracing, refactor planning, and impact analysis than a generic prompt that starts by scanning folders manually.
How to Use code-graph skill
Install and activate it
Use the repository’s skill installation flow for code-graph install, then verify the skill is available in your Claude environment and that the MCP server can be reached. The when-to-use note in the skill says to query the graph before reading files, so activation matters only if the graph backend is actually connected.
Start with the right input
A good code-graph usage request names the target symbol, file area, or change goal instead of asking for “help me understand the repo.” For example: “Use code-graph to find where AuthService is called, map dependencies, and tell me the blast radius before I change token refresh.” That gives the skill enough structure to search the graph effectively.
A practical workflow
Use this code-graph guide flow: first ask for the relevant symbols or dependency chain, then ask for impacted files, then open only the files needed to confirm behavior or edit code. Prefer graph queries for architecture questions and direct file reads for implementation details, config values, or strings that are not well represented in code structure.
Read these files first
Start with SKILL.md to understand the graph-first rule and the supported MCP tools. Then inspect README.md, AGENTS.md, and any rules/, resources/, references/, or scripts/ folders if they exist. In this repo, SKILL.md is the only support file, so it is the main source of truth for code-graph install and usage boundaries.
code-graph skill FAQ
Is code-graph better than grep?
It is better when you need structural answers such as “what depends on this method?” or “what is the blast radius if I rename this symbol?” Grep is still better for literal strings, log messages, and content outside code structure, so the skill is a complement, not a replacement.
When should I not use it?
Do not rely on code-graph when the question is mostly about prose, documentation, or searching for exact text in non-code files. It is also a weaker fit if the MCP server is unavailable or if you only need a one-off answer from a tiny repository where file reading is already trivial.
Is it beginner-friendly?
Yes, if you can describe your goal clearly. Beginners get the most value by asking for a symbol map, dependency path, or impact check before they edit anything, rather than trying to use the skill as a broad repo tour.
Does it replace reading files?
No. The skill is designed to reduce unnecessary file reads, not eliminate them. Use the graph to narrow scope, then read the minimal files needed to verify behavior and make the change safely.
How to Improve code-graph skill
Give the graph a concrete target
The strongest code-graph skill inputs name one symbol, one feature area, or one change outcome. “Find all references to paymentIntent, show callers, and identify files likely affected by a retry change” is far better than “analyze payments,” because the graph can return usable paths instead of a broad summary.
Ask for scope before solution
A common failure mode is jumping straight to implementation. With code-graph, first request the dependency chain, affected modules, and likely edge cases; only then ask for edit recommendations. That sequence improves code-graph usage because the skill is built to answer structure questions before code rewriting.
Resolve uncertainty with targeted file reads
If the graph suggests multiple candidate paths, ask for the smallest set of files that would confirm the right one. This is where code-graph for Code Editing becomes most useful: the graph narrows the search, and short follow-up reads validate the exact behavior before you patch anything.
Iterate on the first result
If the first output is too broad, tighten the query by adding exact identifiers, package names, or the change type: rename, move, refactor, bug fix, or dependency audit. The best code-graph guide prompts are specific enough that the skill can report concrete callers, dependencies, and blast radius without guessing your intent.