luc

Claude Code sub-agents definitions.

Independent code evaluator spawned at the end of a /luc:go run. Reviews the work against phase contracts, acceptance criteria, and quality standards. Provides a structured grade with findings and recommendation. Use proactively after workflow completion to catch issues the generator missed.

Codebase exploration and knowledge extraction agent. Spawned by /luc:go when the codebase needs indexing or when significant changes are detected. Performs deep analysis: indexing, architecture, code flow tracing, dependency mapping, pattern recognition, and cross-cutting concerns. Stores atomic observations in muninndb for future recall by the main workflow agent.

Diagnose and repair Luc tool ecosystem issues. Checks tool availability, health checks, MCP connectivity, hook registration, and .luc/ project directory integrity. Use when something is broken, when /luc:go reports missing tools, or after updating tools. For first-time setup, use /luc:init instead. For dependency updates, use /luc:update.

Start a Luc harness workflow to execute structured, multi-phase tasks. Works for bug fixes, features, refactoring — any task that benefits from understand → plan → implement → validate → commit structure. The harness manages phase transitions, validation, retry loops, and cognitive memory.

First-time setup for Luc on a new project. Audits tool availability (muninndb, rtk, context-mode, codebase-memory), installs missing dependencies, configures RTK hooks, injects Luc instructions into CLAUDE.md, and creates the .luc/ project directory. Run once per project — for diagnostics and repair, use /luc:doctor instead.

Update Luc dependencies by reinstalling all tools from dependencies.json. Runs install commands and verifies health checks pass after installation. For first-time setup, use /luc:init. For diagnostics, use /luc:doctor.

18 hooks across 18 events