Gemini CLI integration

Lint-AI can integrate with Gemini CLI through its command hooks. Gemini sends structured JSON to the hook process, so Lint-AI can record session events and inject relevant project memory without wrapping or replacing the Gemini CLI.

Install

Build with the Gemini feature and install the project hooks:

cargo install --path . --features gemini-cli
lint-ai --gemini-cli-install .

The installer updates ~/.gemini/settings.json, preserving unrelated settings and hooks. It registers both the lint-ai MCP server and lifecycle hooks. Use --gemini-cli-config or --gemini-cli-settings to target explicit files. Re-running the installer is safe and replaces only Lint-AI's own entries.

The installed hooks cover SessionStart, BeforeAgent, AfterAgent, BeforeModel, BeforeToolSelection, BeforeTool, AfterTool, PreCompress, and SessionEnd.

Recording and memory

Gemini session events are stored under:

.lint-ai/gemini-cli-sessions/

The Gemini memory index is .lint-ai/gemini-cli-memory/. The shared recording controls work with --session-provider gemini for promotion and replay. Hook execution is fail-open: recording, indexing, or retrieval failures are reported diagnostically and do not block Gemini CLI from continuing its session. Sensitive payloads use the same redaction and size limits as Claude and Codex recordings.

Gemini CLI's hook protocol does not guarantee token usage in every lifecycle payload. Token metrics should therefore be treated as available only when the CLI supplies usage metadata or telemetry for that event.

MCP note

The Gemini MCP server exposes search, info, record_session, enable_lint_ai, disable_lint_ai, and lint_ai_status. It uses the shared JSON-RPC transport and persistent memory/session-control components.