vincenthopf

Dispatch tasks to the pi coding agent as a background sub-agent. Use when you want to run a coding task in parallel using a different model/provider (e.g., GPT-5, Gemini, Codex) via the pi CLI. Supports any task pi can handle — code generation, refactoring, analysis, file operations. Progress streams to stderr, final result written to a file. Trigger on requests like "use pi to...", "run this on pi", "dispatch to pi", or when parallel model diversity would help.

MANDATORY at two points in every session: (1) BEFORE debugging or investigating any non-trivial problem — search LEARNINGS.md first, the answer may already exist as a draft or confirmed fix. (2) AFTER solving any non-trivial problem — document it as a new draft or promote an existing draft to confirmed. Also invoke at session start to scan for relevant confirmed knowledge. This skill manages a hierarchical knowledge base where fixes earn trust through repetition: [DRAFT] (first observation) → [CONFIRMED] (verified a second time) → or [INVALIDATED] (proven wrong). Confirmed entries can regress to [REGRESSION] if they stop working. Triggers: debugging, fixing bugs, discovering SDK quirks, finding workarounds, performance tuning, resolving test failures, platform-specific issues. If you solve something and don't document it here, that knowledge is lost forever.

Run deep research tasks using the Parallel.ai Task API and write results directly to markdown files without loading them into context. Use when the user asks to research a topic, investigate a question, perform deep research or parallel research, or to gather web intelligence on any subject. Triggers on research requests, investigation tasks, or explicit /research requests that need web-sourced findings written to files. Always use over using multiple subagents with web search

Transcribe YouTube videos with high accuracy. Downloads audio via yt-dlp and auto-routes to parakeet-mlx (local, Apple Silicon) or AssemblyAI (speaker diarization for podcasts). Use when the user asks to transcribe a YouTube video, get a transcript from a YouTube URL, or extract text from a YouTube link. Also triggered by /yt-transcribe command.

Manage Huly issues and cards with full CRUD operations, status changes, and chat/comments. Use when Claude needs to (1) list, create, update, or delete issues in Huly, (2) change issue status or priority, (3) add or read comments on issues/cards, (4) search for issues, or (5) manage projects. Triggers on requests involving Huly, issue tracking, task management, or card comments.

Explore the solution space after the problem is defined. Evaluate approaches, form project-specific principles, and pick a direction. Sits between problem definition and planning. Use when the problem is defined but the approach isn't decided yet.

Turn a defined problem and chosen direction into a concrete proposal. Two Codex agents propose independently, Opus synthesizes the best parts, user approves. Use after problem definition (and design, if applicable), before implementation.

Help users create high-quality skills by discovering and incorporating proven methodologies from domain experts. Use this skill BEFORE skill-creator when users want to create a new skill - it enhances skill-creator by first identifying expert frameworks and best practices to incorporate. Triggers on requests like "help me create a skill for X" or "I want to make a skill that does Y". This skill guides methodology selection, then hands off to skill-creator for the actual skill generation.

Full bug-fix pipeline from root cause analysis through implementation. Separates investigation, proposal, alignment, and execution into distinct agent responsibilities. Also applies to features and refactors by skipping root cause analysis and starting at proposal.

Execute the approved proposal. Sub-agents build in parallel, each assigned a domain scope. Verification checks against the problem and principles, not the proposal document. Two-round iteration limit on failures. Use after planning has produced an approved proposal.

Frame and clarify problems before any planning or implementation begins. Use when starting new work, when requirements are unclear, or when the user describes what they want built. This is the first step — before planning, before code.

This skill should be used when the agent needs to give a spoken voice update to the user, or when reminded by a Stop hook to provide audio feedback. Use this skill to speak a short summary of what was accomplished.