trending/codebase-memory-mcp
1
0
Fork 0
mirror of https://github.com/DeusData/codebase-memory-mcp.git synced 2026-10-29 07:13:44 +08:00
Code Issues Packages Projects Releases Wiki Activity

docs(config): rewrite context_injection description for clarity

Browse Source 
Lead with what it does, then why to enable/disable with multiple examples.
Use colons not em-dashes. Cover non-code sessions, token-metered environments,
CI pipelines, and scripted use as disable reasons. Include exact commands.

Signed-off-by: Andrew Hundt <ATHundt@gmail.com>
This commit is contained in:
Andrew Hundt
2026-04-06 04:46:21 -04:00
parent 12c22e201b
commit ee8e6bbad4
1 changed files with 12 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/cli/cli.c
+12 -17
View File
@@ -1892,24 +1892,19 @@ const cbm_config_entry_t CBM_CONFIG_REGISTRY[] = {
"You can also enable individual classic tools without switching modes: "
"config set tool_index_repository true"},
{"context_injection", "true", "CBM_CONTEXT_INJECTION", "Tools",
"Inject _context (schema/architecture) into the first tool response — enable for AI, disable for scripting",
"Inject codebase schema and stats into the first tool response so the AI starts informed",
"true|false",
"WHAT IT DOES: On the first search_code_graph/search_graph call of a session, "
"embeds a _context object containing node/edge counts, all node labels with counts, "
"all edge types with counts, PageRank stats, and detected ecosystem (Go/Python/Rust/etc.). "
"This gives the AI model automatic situational awareness without requiring it to call "
"get_architecture or get_graph_schema first. Subsequent calls omit _context (one-shot "
"delivery, controlled by the context_injected session flag). "
"WHEN TO DISABLE (context_injection=false or CBM_CONTEXT_INJECTION=false): "
"(1) Scripted/programmatic use where you parse tool output and the extra JSON adds noise. "
"(2) CI pipelines where token cost is metered and the model does not need schema context. "
"(3) When the AI is given explicit codebase instructions via system prompt instead. "
"(4) Benchmarking raw tool latency without the schema query overhead. "
"WHEN TO KEEP ENABLED (default): Interactive AI sessions where the model benefits from "
"knowing the codebase structure before it starts querying. Saves 2-3 round-trips that "
"would otherwise be needed to fetch schema and architecture separately. "
"Per-session override: set CBM_CONTEXT_INJECTION=false in the environment. "
"Per-install default: codebase-memory-mcp config set context_injection false"},
"When true (default), the first search_code_graph/search_graph response includes a "
"_context object: node/edge counts, node labels, edge types, PageRank status, and "
"detected language ecosystem. Delivered once per session; subsequent calls are unaffected. "
"Why enable: the AI gets codebase structure upfront without needing to call "
"get_architecture or get_graph_schema separately. Useful for code exploration, "
"refactoring, debugging, and any session focused on understanding the codebase. "
"Why disable: context window space consumed by _context is wasted when the session "
"involves non-code tasks, scripted/programmatic tool use, CI pipelines, token-metered "
"environments, or when the model already has codebase context from another source. "
"To disable for a session: export CBM_CONTEXT_INJECTION=false "
"To disable by default: codebase-memory-mcp config set context_injection false"},
{"compact", "true", "CBM_COMPACT", "Tools",
"Default compact output for search_code_graph, trace_call_path, and get_code",
"true|false",