dl-skills

Use the Antigravity agent (agy CLI) as a second-opinion AI for code review, debugging, analysis, and architecture decisions. Trigger when the user asks to "ask antigravity", "check with antigravity", "antigravity opinion", "what does antigravity think", or wants a second opinion specifically from the Antigravity agent. Also trigger on "agy cli" or "antigravity cli" mentions.

Decide whether a dependency version released within the last 7 days may be adopted early, and document the decision. Use this whenever a needed fix, feature, or security patch sits inside the 7-day quarantine window, whenever someone argues an urgent bump should skip the wait, and whenever you are about to run uv lock, uv add, npm install, or edit a lockfile for a very recent release. Use it even if the package looks obviously safe.

This skill should be used when the user asks to "ask Codex", "get a second opinion from Codex", "check with OpenAI", "get OpenAI's take", "what does Codex think", "ask Codex with high reasoning", "use a specific Codex model", requests an adversarial or ship/no-ship review of a change, or asks to delegate an implementation to Codex for Claude to review afterward. It should also be used for non-trivial code review, debugging hypotheses, architecture decisions, and fact-checking when an independent Codex consultation would materially improve confidence.

Build applications using OpenAI's Responses API, the recommended API for new projects. Use when creating apps that call OpenAI models with features like web search, file search, code interpreter, image generation, computer use, shell, apply patch, MCP servers, programmatic tool calling, function calling, structured outputs, multi-turn conversations, or reasoning models. Triggers on requests to integrate OpenAI, build AI agents, create chatbots, or develop applications using GPT-5.6 (sol, terra, luna), GPT-5.5, GPT-5.4, GPT-5.3-Codex, or earlier GPT-5 and GPT-4.1 models via the Responses API. Also trigger when the user says "use GPT", "call OpenAI", "openai client", "responses.create", or asks how to use OpenAI's Python/Node SDK for new projects. If the user imports `openai` or references the OpenAI SDK, use this skill.

Build local web UIs powered by Ollama for private AI chat and document Q&A. Use this skill when the user wants to create a local/private AI interface, build a web app that talks to Ollama, create a document analysis tool, add a web frontend to an LLM backend, build a streaming chat UI, or create any localhost AI application where privacy matters (no cloud transit). Also trigger when the user mentions FastAPI + Ollama, SSE streaming for LLM, local AI dashboard, or private document processing.

Run a full multi-model coding pipeline - the session model (Fable) plans and reviews, headless OpenAI Codex implements, and an Opus subagent applies final fixes, ending with a summary and dev-doc updates. Use whenever the user says "orchestrate", "/orchestrate", "run the orchestrate workflow", "have Codex build it and Opus fix it", "delegate this to Codex", or asks for a coding task to be split across Claude and Codex/OpenAI models - even if they only loosely describe wanting Codex to implement while Claude plans or reviews.

Scaffold a new or unconfigured repository with the standard locked toolchain, pre-commit safety stack, project CLAUDE.md, and decisions.md. Invoke manually with /repo-bootstrap.