Jacob Magar 251979edfe perf(gemini-cli): skip MCP server startup via workspace settings override ... The user has 6 MCP servers configured in ~/.gemini/settings.json. Without mitigation, the gemini CLI spawns all of them on every headless call, adding 10-60+ seconds of startup latency. Two flags reduce this: - cmd.current_dir(workdir): workspace .gemini/settings.json with {"mcpServers":{}} overrides ~/.gemini/settings.json, blocking all 6 MCP servers from spawning. The workdir is /tmp/noxa-gemini/ and is created once at GeminiCliProvider::new(). - --extensions "": prevents extension loading (~3s saved) Per geminicli.com/docs: workspace settings override user settings. The --allowed-mcp-server-names flag was tested but hangs with a fake name and exits without response for empty string — not usable. Result: consistent 13-17s per call vs >60s baseline with MCP servers.		2026-04-11 20:23:33 -04:00
..
src	perf(gemini-cli): skip MCP server startup via workspace settings override	2026-04-11 20:23:33 -04:00
Cargo.toml	feat(noxa-9fw.3): validate structured extraction output with one retry	2026-04-11 07:34:58 -04:00