AI Engineering

Ralph processes tasks from a work queue, spawning isolated subagents for each phase to keep context fresh. Use it to run multi-phase knowledge base pipelines in serial, parallel, batch, or dry-run modes.

Reduce extracts structured knowledge from source material into atomic, retrievable notes. Use it when triggered by /reduce, "extract insights", or when processing inbox items and handoff queues to externalize reasoning and evidence.

Reflect finds connections between notes and updates MOCs using semantic judgment. Use it after /reduce creates notes, when exploring connections, or when a topic needs synthesis.

Remember captures friction as methodology notes for Claude Code agents. Use /remember with an explicit description, empty for contextual review of recent corrections, or --mine-sessions to scan transcripts for patterns and prevent repeating mistakes.

Rethink challenges system assumptions against accumulated evidence, triaging observations and tensions, detecting patterns, and generating proposals. Use it when a knowledge system needs to review what it has learned and revise its own rules.

Codebase Wiki guides agents in generating and refreshing a source-grounded, version-controlled wiki of a codebase, with per-folder rules, freshness and log discipline, and audience/depth knobs. Use when the project has a wiki/ knowledge base or when asked to generate or refresh a wiki of the codebase.

Open Knowledge defines the agent-runtime contract for working inside an OpenKnowledge project, a markdown-CRDT knowledge base exposed over MCP. Use it when reading, listing, searching, editing, or linting any .md or .mdx file, and before any mcp__open-knowledge__* tool call.

Open Knowledge Format provides guidance for creating, reading, reviewing, and maintaining OKF v0.2 bundles. Use it when responding to OpenKnowledge okf plugin warnings or when choosing types, provenance, links, indexes, or logs.

OpenKnowledge Skill Authoring guides the creation of Agent Skills (SKILL.md files) by checking existing skills, resolving scope, and structuring progressive-disclosure instructions. Use it when writing a new skill for OpenKnowledge or an editor.

Auto-Empirical Research routes empirical-research requests to the right vendored skill from a catalog of 1,096 skills across 76 collections, covering causal inference, econometrics, data acquisition, writing, and peer review.

Empirical Research Navigator routes agents to the correct empirical research skill by goal, method, and language across a library of 1,052 skills. Use it before starting an empirical pipeline, paper, or replication package to select Python, Stata, or R workflows and linked AER skills.

Ask delegates work to or requests information from another CCB-managed agent. Use it when the user asks to delegate, hand off, consult, or send work to a named CCB agent.