AI Engineering

Research Deep reads a research outline and launches independent agents for each item to perform deep research, writing structured JSON results with validation and resume support.

Research Deep reads a research outline (outline.yaml) and launches independent background agents to investigate each item, writing structured Chinese JSON results with validation and checkpoint resumption. Use it to run multi-item deep research pipelines in Claude Code.

ACE-Step Music Generation creates background music, vocal tracks, covers, and stems using the ACE-Step 1.5 model. Use it to produce soundtracks, jingles, or audio edits for video projects.

ElevenLabs Audio generates voiceovers, sound effects, and music using the ElevenLabs APIs. Use when creating audio content for videos, podcasts, or games.

Ideogram 4 generates images via Ideogram's hosted v4 API using structured JSON captions for legible in-image text, exact brand colors, and controlled layout. Use it when producing title cards, thumbnails, quote cards, or CTAs that need baked-in text.

LTX2 generates ~5 second video clips from text prompts or images using the LTX-2.3 22B DiT model. Use it when you need b-roll, animated slide backgrounds, animated portraits, or other motion content for video production.

Qwen Image Edit provides prompting patterns, parameter tuning, and examples for AI image editing with Qwen-Image-Edit, used to preserve identity in photo edits, reframe cropped images, change clothing or accessories, adjust poses, apply style transfers, or transform characters.

Blue Ocean Strategy creates uncontested market space using value innovation instead of competing head-to-head. Use it to apply the Four Actions Framework, Six Paths, and buyer utility map when a market is crowded or you need to find underserved non-customers.

Design Code Architecture guides an agent through eight phases of app architecture decisions — boundaries, domain model, data, and resilience — recording choices in docs/ so the journey resumes across sessions. Use it to design a new app's architecture before building, choose structure, or decide monolith vs microservices.

Karpathy LLM Wiki builds and maintains a personal LLM-powered knowledge base by ingesting sources into raw/ and compiling durable wiki articles with citations. Use it to query wiki knowledge, lint consistency, and compound a knowledge base over time.

Fleet Auditor detects installed agent systems, collects token usage data, identifies waste patterns, and recommends fixes with dollar savings estimates. Use it when running multiple agent systems, spending $2-5/day on agents, or suspecting idle heartbeats are burning tokens.

Resume Checkpoint fetches a prior session's checkpoint on demand when continuing prior work, returning fenced, source-labeled recovery context. Call it when a user says continue or resume and the SessionStart pointer was not enough.