AI Engineering

DeepSeek Harness delegates code review, development, and research tasks from the OpenCode host to an external DeepSeek Harness agent. Use it to get an isolated review, a candidate patch, or offload work while keeping credentials out of prompts.

iPolloWork Template Generation helps create, repair, validate, and save reusable templates for Design, Slides/PPT, and HyperFrames Video. Use it when building template packages, manifests, variables, covers, or surface-specific structure.

Voiceover turns a feature request into an approved demo narration before any code, then builds on a fresh worktree and opens a PR with frame-by-frame proof. Use when a feature request arrives or when the user runs /voiceover.

Caveman keeps technical substance while removing communication waste, defaulting to concise direct answers in AI coding tools. Use it for terse, low-token responses plus built-in code review and commit message patterns.

Lazy Senior Dev guides coding agents to write minimal, efficient code by reusing existing utilities, standard libraries, platform features, and installed dependencies before adding new code. Enable it for bug fixes and feature work where the shortest correct diff wins, but not for exploratory research or redesigns.

Obsidian Second Brain operates any Obsidian vault as a self-rewriting knowledge base for AI agents, saving notes, updating pages, and reconciling contradictions automatically. Use it when Claude needs to read, write, search, or manage an Obsidian vault.

Agent Topology Audit evaluates whether a multi-agent setup earns its coordination cost. Use it before adding an agent or when a workflow feels slow or agents agree without adding signal.

AI Debate Hub runs structured multi-provider AI debates with Claude and available advisors to help make critical decisions. Use it when you want multiple AI perspectives synthesized before choosing a technical or strategic direction.

Context Detection automatically identifies whether the current task is development or knowledge work, so agent workflows can tailor research, build, and review behavior. It checks explicit overrides, prompt indicators, and project structure before returning a confidence-scored context.

Copilot Provider integrates GitHub Copilot CLI as an optional zero-cost provider in the Claude Octopus multi-LLM ecosystem, using programmatic mode for research and exploration. Use it to supplement other providers with Copilot responses, with graceful degradation if unavailable.

Cost Projections estimates remaining workflow costs from per-phase averages and warns when projections exceed a budget ceiling. Use it when you need to track and forecast spend across multi-step AI workflows.

Council orchestrates a configurable council of multiple LLMs with personas, budget caps, synthesis, veto gates, and optional implementation handoff. Use it when you need adversarial review, decision support, or implementation planning across models.