d3x293

Creates, lists, and removes custom project-specific agents. Custom agents are stored in .claude/crew-agents/ and registered in crew-team.json for routing.

Manages CodeCrew configuration. Allows users to customize default mode, model tiers, agent activation, token budgets, and auto-index settings without editing JSON files.

Smart hybrid execution engine that dispatches agents in parallel or sequential order based on task dependencies. Manages the Agent tool calls, result collection, conflict detection, and integration review.

Lite task router that bypasses Opus CEO entirely. Classifies tasks and routes directly to Sonnet and Haiku agents only. Use when user runs /crew lite or /crew task --lite.

Reads and displays task execution history from crew-history.json with filtering by type, agent, cost, and configurable limits.

Displays CodeCrew status including project profile, agent team configuration, codebase index health, skill catalog stats, and recent task history.

3-layer codebase indexing system for token-efficient code navigation. Builds compact index + symbol map so agents read targeted lines instead of entire files. Achieves ~5-10x token savings.

Analyzes a project codebase, builds 3-layer index, generates project profile, and auto-configures the agent team with relevant skills. The main initialization skill for CodeCrew.

Keeps the 3-layer codebase index fresh after file edits. Performs incremental updates by recomputing hashes and re-extracting metadata for only changed files.

Auto-selects and injects relevant skills from the 125+ ECC catalog into agent prompts based on task type, project stack, and agent role. Prevents skill overload by loading only what's needed.

Reverts the last task's file changes using git. Reads task history for modified files and git reference, shows a preview, requires user confirmation, then reverts and updates the index.

CEO's task decomposition and routing engine. Classifies tasks by type and complexity, selects agents with appropriate model tiers, decides parallel vs sequential execution, and auto-injects relevant skills.