AI Assisted Speedruns — AI skill for Claude Code
An open standard and plugin framework for AI Assisted Speedruns: an LLM agent plays a game through three hardened tools, every decision lands on one timeline, the run is recorded and published as a bu.
How to install AI Assisted Speedruns
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open moeilijk/ai-assisted-speedruns and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What AI Assisted Speedruns does
An open standard and plugin framework for AI Assisted Speedruns: an LLM agent plays a game through three hardened tools, every decision lands on one timeline, the run is recorded and published as a bundle anyone can verify.
Alternatives in AI
- Orkas — Open-source multi-agent AI desktop client — build and command your AI agent team through conversation 1.5k ★
- StereOS — A Linux based operating system hardened and purpose built for AI agents 489 ★
- Upstream Watch — Review pydantic-ai and pydantic-ai-harness releases published since the last watch entry, and record a verdict 361 ★
README
ai-assisted-speedruns
A plugin framework and an open standard for **AI Assisted Speedruns**: an LLM agent plays a game through a small, hardened tool interface, every decision is logged on one timeline, the run is recorded, and the result is published in a form anyone can verify and compare.
Three kinds of plugins, one core:
| Plugin | Role | First implementations |
|---|---|---|
| game | exposes one game as a controller object plus API documentation | Portal (via cozyblaze/portal-agent), Slay the Spire (via Communication Mod); Kerbal Space Program (kRPC) planned |
| runtime | starts the agent against the broker with a hardened config | Codex, Claude Code |
| recorder | records the run and reacts to run events | OBS (obs-websocket v5), source-demo (in-game demo via SPT), null |
| timer | speedrun timer on screen and splits | LiveSplit (LiveSplit Server) |
The agent only ever sees three tools per game: `_documentation`, `_screenshot`, `_exec`. That shape, the broker, the process hardening and the log/export format come from [cozyblaze/portal-agent](https://github.com/cozyblaze/portal-agent) (MIT), which is credited as their origin and is not affiliated with this project.
Disclaimer: anti-cheat, bans, your own risk
This framework drives a game with injected tooling (for Portal: SourcePauseTool's DLL and an IPC patch, with the game started outside the normal launcher). Anti-cheat and monitoring systems such as VAC, EAC or BattlEye can treat that as cheating and **ban the account, possibly permanently**. Use it only with games and builds that have no anti-cheat and no online component, never with online or competitive games, and preferably with a separate offline copy of the game (Portal here runs from Source Unpack, not from the Steam install). Do your own research on the game's terms and anti-cheat before you connect anything, use common sense, and accept that everything you do with this repository is at your own risk: the
Related Skills
Apex Formula
APEX FORMULA — open-source Formula 1 racing game for the browser. Three.js, 100% procedural assets, 120 Hz veh
AI Gaming Agent
A step by step implementation of building an AI agent that plays 3d shooting game
Agent Prompt Observatory
Visual history of Claude Code's system prompt: stream-graph timeline across every published version, inline co
Claude Drives Gemini
Let your AI agent drive the Gemini web app through your existing Google AI Pro subscription - generate images
Operator Cmo
An AI marketing team that has to show its plan. Claude Code plugin: eight marketing skills in three layers (de
Dayz AI Survivor
Autonomous DayZ survivor driven by Claude Code — multi-agent NPCs with memory, voice (Discord + in-game), and
Related Agents
Player Classic
Clash Royale 1v1 player agent. Plays cards at maximum speed during battles. Use for standard ladder matches.
Docs Drift Review
Given a session's commits or diff, finds the docs that the change made wrong — stale paths and symbols, checkl
ADR Decision Check
Use to check whether a set of pending code changes represents an architectural decision that should be recorde