메뉴
Use when you need structural graph navigation — trace call paths, detect blast radius, find dead code, or query architecture — before reading files. Structural queries only: not for semantic meaning (use cocoindex-code for that).
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Auto-configure AGENTS.md and brain/ knowledge docs for a new project. Use when setting up Brain Bootstrap in a new repository. Discovers the codebase, fills in templates, and writes project-specific configuration.
Use before any new feature, component, or behavioral change — when you need to explore requirements and design before writing code. Hard gate — no code until design is approved.
Build the project and verify it compiles cleanly. Use after making changes to confirm nothing is broken before running tests. Reads the build command from brain/build.md.
Activate safety mode — review every destructive command before running. Use before working on production infrastructure, migration scripts, or any task where a typo could be catastrophic.
Generate a user-facing changelog from git commits — categorize, filter noise, translate to user language. Pass a tag or date as argument (e.g. "$changelog v1.0.0" or "$changelog 2026-01-01").
Save session state before context gets full or before ending. Writes current task state, branch, and loaded docs to brain/tasks/todo.md so the next session can resume cleanly.
Zero-file-read structural analysis via codebase-memory-mcp. Always try these tools BEFORE reading files for structural questions.
Not for semantic search (use $cocoindex-code). Not for rename/refactor (use $serena).
| Question | Tool |
|---|---|
What calls foo()? | trace_path(function_name="foo", direction="inbound") |
What does foo() call? | trace_path(function_name="foo", direction="outbound") |
| Full call chain | trace_path(direction="both", depth=3) |
| Impact of my change | detect_changes(base_branch="main") |
| Architecture overview | get_architecture |
| Find by name/pattern | search_graph(name_pattern="Auth.*") |
| Dead code | search_graph(max_degree=0, label="Function") |
| Custom Cypher query | query_graph(query="MATCH (n:Function)-[:CALLS]->(m) RETURN n.name LIMIT 20") |
| Execution flows | list_flows → get_affected_flows |
# Index the repository (one-time setup, ~6s for 500 files)
index_repository(repo_path=".")
# Verify
list_projects
# View architecture map
get_architecture
.codex/config.toml[mcp_servers.codebase-memory]
command = "uvx"
args = ["codebase-memory-mcp@latest", "--data-dir", ".codex/codebase-memory"]
Or if using the binary release directly:
[mcp_servers.codebase-memory]
command = "codebase-memory-mcp"
args = []
1. detect_changes(base_branch="main") → risk score + blast radius
2. get_affected_flows → which execution paths break
3. trace_path on highest-risk functions → full call chain context
1. get_architecture → community map, entry points, god nodes
2. search_graph(name_pattern="<keyword>") → find relevant nodes
3. trace_path → traverse from found nodes
4. get_code_snippet(qualified_name="<name>") → source with context
trace_path requires exact qualified names — use search_graph(name_pattern=...) first to find themhas_more and paginate with offsetquery_graph has a 200-row cap — use search_graph with degree filters for countsdirection="outbound" misses cross-service callers — use direction="both" to see allindex_repository(mode="fast") for immediate refresh120× fewer tokens than file exploration for structural questions. A call chain that would require reading 15 files can be answered in 1-2 MCP calls.