How do I install the Cgh Use Codegraph skill?

Run npx skillmds@latest add altikva/cgh-use-codegraph in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Cgh Use Codegraph safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Cgh Use Codegraph?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Cgh Use Codegraph free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Cgh Use Codegraph?

altikva (@altikva) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Cgh Use Codegraph by altikva — Agent Skill

Name: Cgh Use Codegraph
Author: altikva

Cgh Use Codegraph

Use codegraph MCP tools BEFORE reading files when navigating code. Triggers on symbol lookups ("where is X defined", "what calls Y"), multi-file exploration ("how does X work"), and task kickoff. Saves 60-90% of exploration tokens.

altikva 2855b25 2.8 KB Updated Sep 17, 2026

File contents

Use codegraph before reading files

Codegraph is a local code graph (symbols, calls, imports, docs) exposed over MCP. Using it for navigation returns exact file paths + line ranges so you only read the lines you actually need.

Token economics: MCP tool execution runs server-side and costs zero model tokens. The only token cost is the JSON response, which is capped and truncated. Read/Grep, in contrast, pulls the entire file into the transcript. Call codegraph tools liberally: they're almost always cheaper than reading. Default bias: call the tool, don't hesitate.

Triggers → tool

Situation	Tool to call first
"Where is `X` defined?"	`mcp__codegraph__symbol_lookup(name="X")`
"What calls `fn`?" / "Who uses `fn`?"	`mcp__codegraph__find_callers(fn_name="fn")`
"What does `fn` call?"	`mcp__codegraph__find_callees(fn_name="fn")`
"Trace the flow from `fn`" (multi-hop)	`mcp__codegraph__find_callees(fn_name="fn", max_depth=N)`: one call returns the ordered chain, don't chain single hops
"What breaks if I change `X`?"	`mcp__codegraph__impact_of(symbol_or_file="X", max_depth=N)`
Fuzzy / partial name	`mcp__codegraph__search_symbols(query="...")`
Full-text (docstrings, names)	`mcp__codegraph__fts_search(query="...")`
Regex / substring over files	`mcp__codegraph__pattern_search(pattern, glob?)`: INSTEAD of Grep
"How does [feature] work?": starting a non-trivial task	`mcp__codegraph__context_for_task(task="...")`
"What docs mention `X`?"	`mcp__codegraph__search_docs` / `doc_refs`
"Blast radius of `file.py`"	`mcp__codegraph__subgraph(file_path="...")`
"Is `fn` dead / unused?"	`mcp__codegraph__find_dead_code`

Workflow

Identify the target (symbol name, file, or task description).
Call the relevant codegraph tool.
Use the returned file_path + start_line/end_line to Read only the needed range: not the whole file.

Don't

Don't Grep / Read a file "to find where X is" if symbol_lookup would return it directly.
Don't read an entire file just to list its functions: use doc_outline for markdown or walk symbols via search_symbols.
Don't run manual cgh CLI commands during a session; use MCP tools so results are logged and cached.

Staleness

The graph reflects the last scan, not the live working tree. If the user mentions recent git pull, checkout, rebase, or major edits, call mcp__codegraph__scan_status first. If fresh=false, call mcp__codegraph__scan_repo before trusting symbol results.

---name: cgh-use-codegraphdescription: Use codegraph MCP tools BEFORE reading files when navigating code. Triggers on symbol lookups ("where is X defined", "what calls Y"), multi-file exploration ("how does X work"), and task kickoff. Saves 60-90% of exploration tokens.---# Use codegraph before reading filesCodegraph is a local code graph (symbols, calls, imports, docs) exposed overMCP. Using it for navigation returns exact file paths + line ranges so youonly read the lines you actually need.**Token economics**: MCP tool execution runs server-side and costs zeromodel tokens. The only token cost is the JSON response, which is cappedand truncated. Read/Grep, in contrast, pulls the entire file into thetranscript. Call codegraph tools liberally: they're almost alwayscheaper than reading. Default bias: call the tool, don't hesitate.## Triggers → tool| Situation | Tool to call first ||---|---|| "Where is `X` defined?" | `mcp__codegraph__symbol_lookup(name="X")` || "What calls `fn`?" / "Who uses `fn`?" | `mcp__codegraph__find_callers(fn_name="fn")` || "What does `fn` call?" | `mcp__codegraph__find_callees(fn_name="fn")` || "Trace the flow from `fn`" (multi-hop) | `mcp__codegraph__find_callees(fn_name="fn", max_depth=N)`: one call returns the ordered chain, don't chain single hops || "What breaks if I change `X`?" | `mcp__codegraph__impact_of(symbol_or_file="X", max_depth=N)` || Fuzzy / partial name | `mcp__codegraph__search_symbols(query="...")` || Full-text (docstrings, names) | `mcp__codegraph__fts_search(query="...")` || **Regex / substring over files** | `mcp__codegraph__pattern_search(pattern, glob?)`: INSTEAD of Grep || "How does [feature] work?": starting a non-trivial task | `mcp__codegraph__context_for_task(task="...")` || "What docs mention `X`?" | `mcp__codegraph__search_docs` / `doc_refs` || "Blast radius of `file.py`" | `mcp__codegraph__subgraph(file_path="...")` || "Is `fn` dead / unused?" | `mcp__codegraph__find_dead_code` |## Workflow1. Identify the target (symbol name, file, or task description).2. Call the relevant codegraph tool.3. Use the returned `file_path` + `start_line`/`end_line` to `Read` only the needed range: not the whole file.## Don't- Don't Grep / Read a file "to find where X is" if `symbol_lookup` would return it directly.- Don't read an entire file just to list its functions: use `doc_outline` for markdown or walk symbols via `search_symbols`.- Don't run manual `cgh` CLI commands during a session; use MCP tools so results are logged and cached.## StalenessThe graph reflects the last scan, not the live working tree. If the usermentions recent `git pull`, `checkout`, `rebase`, or major edits, call`mcp__codegraph__scan_status` first. If `fresh=false`, call`mcp__codegraph__scan_repo` before trusting symbol results.

Frequently asked questions