AI Engineering

Agentic Workflow defines a multi-agent pipeline for implementation tasks in Claude Code, using background agents, file-based handoffs, and TDD. Use it to orchestrate research, planning, validation, implementation, and review without flooding the main context.

Agentica Claude Proxy connects Agentica SDK to Claude Code CLI through a local proxy so agents can use Read, Write, Edit, and Bash tools. Use it when setting up or debugging agent integrations, fixing permission errors, or preventing hallucinations.

Agentica Infrastructure documents the Agentica multi-agent coordination APIs, including 11 pattern classes, primitives, and the coordination database schema. Use it when building multi-agent workflows, implementing custom patterns, or debugging agent tracking.

Agentica Prompts provides templates and patterns for writing reliable prompts for Agentica/REPL agents, reducing instruction ambiguity and improving multi-agent orchestration reliability.

Agentica SDK builds Python agents using the Agentica framework with @agentic decorator, spawn(), persistence, and MCP integration. Use it to add agentic capabilities to Python code or orchestrate multiple agents.

Agentica Server configures and runs the Agentica SDK with a local Claude proxy so Python agents can use Claude CLI as their inference backend. Use it when setting up a fresh environment, debugging connection issues, or troubleshooting tool access and hallucinations.

Agentica Spawn executes multi-agent patterns such as swarms, hierarchical coordination, generator/critic loops, and juries after a pattern is selected. Use it to implement parallel research, coordinated building, iterative refinement, or high-stakes validation with spawned agents.

Background Agent Pings instructs Claude Code agents to trust system reminders for background task progress rather than polling. Use it when you want to avoid wasted tokens and context flooding while background agents run.

Braintrust Tracing explains how to capture Claude Code session traces in Braintrust, including hook event flow, state management, and sub-agent correlation. Use it to debug agent runs, inspect token usage, or link sub-agent traces to parent sessions.

Compound Learnings analyzes session learning files to extract recurring patterns, then proposes and creates permanent capabilities such as rules, skills, hooks, and agent updates. Use it when asked to improve setup, compound learnings, or turn session patterns into reusable artifacts.

Create Handoff generates a YAML handoff document that summarizes a Claude Code session's goal, completed work, decisions, blockers, and next steps, so another agent can pick up where it left off. Use it when ending a session or transferring context to a new session.

Debug Hooks is a systematic workflow for diagnosing Claude Code hooks that aren't firing, producing wrong output, or behaving unexpectedly. Use it to check outputs, verify registration, test hooks manually, and catch silent failures.