Agent Workflow

Two-stage retrieval

cgrep agent is optimized for low-token loops with deterministic payloads.

1. locate: return compact candidate set
2. expand: fetch richer context only for selected IDs

# Stage 1: locate (json2-oriented output)
cgrep agent locate "where token validation happens" --changed --budget balanced --compact

# Short alias form:
cgrep a l "where token validation happens" -u -B balanced --compact

# Select first result ID (example)
ID=$(cgrep agent locate "token validation" --compact | jq -r '.results[0].id')

# Stage 2: expand chosen result(s)
cgrep agent expand --id "$ID" -C 8 --compact

Notes: - agent locate/expand use payload minimization defaults - agent locate supports caching for repeated prompts

Integration install

Install instructions into supported agents:

cgrep agent install claude-code
cgrep agent install codex
cgrep agent install copilot
cgrep agent install cursor
cgrep agent install opencode

Uninstall:

cgrep agent uninstall claude-code
cgrep agent uninstall codex
cgrep agent uninstall copilot
cgrep agent uninstall cursor
cgrep agent uninstall opencode

Legacy install-* and uninstall-* commands remain for compatibility.

Cursor note: - agent install cursor writes a project-local rule file: .cursor/rules/cgrep.mdc - MCP is also supported for Cursor via cgrep mcp install cursor