AI Engineering

Research synthesizes web and GitHub findings into an answer with patterns, gotchas, and references. Use it when research files have already been gathered and you need a concise recommendation.

Advisor Consultation uses a tool-free Opus second opinion to evaluate high-risk design trade-offs, unresolved failures, and cross-boundary decisions. Use it after gathering repository evidence when the main agent needs a careful review.

AI Second Brain walks users through building an Obsidian-based AI knowledge base from ChatGPT and Claude exports, research sources, and daily inputs. It also covers Karpathy-style wiki setup and phone access via iMessage Channels and slash commands.

Auto Converge refines rules, prompts, and SKILL.md files until agents can follow them reliably. It uses external references and adversarial review loops when a spec is written but not being obeyed.

Goal Mode tracks a single long-running objective across multiple turns. Use it for complex tasks that need continuous progress until completion, blocking, or reset.

GPT Image generates or edits raster images through a ChatGPT-backed bridge and saves PNGs in the workspace. Use it for text-to-image, reference-based edits, revisions, variants, and transparent PNGs.

Langfuse interacts with Langfuse data through the CLI and helps retrieve documentation, concepts, and SDK guidance. Use it to query traces, prompts, datasets, scores, sessions, and other API resources or to understand Langfuse features.

Learn From History audits recent conversation history to find repeated failures, successful patterns, stable rules, and skill improvements. Use it when you need to summarize prior agent behavior or learn from recent runs.

LLM Log Analyzer analyzes LLM gateway request and response logs under ./data/. Use it to inspect sessions, compare requests, check token usage, and diagnose cache behavior or failures.

Programmatic Tool Calling discovers extra tools and uses ESM code to compose dependent calls into one bounded run. Use it to filter, join, or summarize structured tool results without returning large intermediates.

ROS 2 Engineering provides guidance for ROS 2 development, review, and debugging across nodes, QoS, launch, tf2, control, navigation, simulation, and migration. Use it when diagnosing behavior or making changes in a ROS 2 workspace.

Ultra Task supervises non-trivial work by delegating bounded investigation, implementation, testing, and review to subagents while the main agent coordinates scope, approvals, and verification. Use it when a task needs orchestration rather than a single direct step.