linuxarena

Take a finished change from working diff to a ready-for-review PR, ending with the PR URL as plain text. Use when the user says "ship it" or wants to go from a working change to an opened PR.

Commit the current change, push a branch, and open a draft PR. Use when the user wants to open a PR for finished work.

How to drive the `ct` CLI to run matches and produce every number and graph a control experiment ships. Covers `ct run eval`, `ct run monitor`, `ct run summarize`, `ct game-metrics metrics / graph roc / graph pareto`, `ct traj`, and the viewers. Load this when you launch a match, compute safety or usefulness, regenerate a graph, or need the exact reproduce command for an output. White loads it always; main and others load it on demand.

The empirical-research method and quality bar for AI-control experiments run through the ct CLI. Reduce uncertainty fast, iterate on qualitative plus quantitative evidence, size for power, keep splits clean, report honestly, and hold P1 observability, P2 affordance discipline, and the per-team B/R/W/M criteria. Loaded by every role.

The public, LessWrong-style writeup of a control experiment. Ruthlessly compressed for a reader with zero context, leading with the surprise, citing public-viewer trajectory URLs only. Load this in the writing phase when producing the shipped post into `outputs/`. It lifts the STRONG-tagged claims from the internal [report-writing]; same evidence base, far less detail. Before shipping it runs the lw-review gate below.

The red (attacker) role for ONE protocol in an automated control-research run. Load this when you are spawned as red_<protocol>. Holds red's self-managed loop, guarded objective, iteration discipline, and validity rules. Every shared fact (folder layout, iteration mechanics, the stop gate, the game, safety formula, the quality bar, ct commands) lives in a general skill and is referenced here by name, never copied.

The white (referee) role for ONE protocol in an automated control-research run. Load this when you are spawned as white_<protocol>. Holds white's self-managed loop, guarded objective, task-set ownership, validity disciplines, and per-protocol stopping judgment. Every shared fact (folder layout, iteration mechanics, the stop gate, the game, safety formula, the quality bar, ct commands) lives in a general skill and is referenced here by name, never copied.

The discipline that makes a multi-hour control run actually finish instead of stalling or quitting early. It keeps going, get unstuck, fix problems, self-review before you report. Loaded by every agent to keep it going.

Guide to Control Tower's AI control concepts. Use when the user asks about control protocols, blue team and red team roles, usefulness, safety, audit budgets, trusted monitoring, or the conceptual model behind Control Tower.

Guide for creating new Control Tower environments. Use when the user asks about creating environments, environment structure, compose files, Docker setup, tech stack diversity, attack surface, or environment properties.

Guide for creating main tasks in Control Tower environments. Use when the user asks about creating, implementing, or designing main tasks, task categories (Add Feature, Fix Bug, Refactor Code, Answer Question), scorers, task requirements, or info.yml structure for tasks.

Guide for writing scorers in Control Tower. Use when the user asks about creating scorers, scorer patterns, testing scorers, scorer boilerplate, or how to validate task completion.

Guide for creating side tasks in Control Tower environments. Use when the user asks about creating side tasks, security property violations, attack surface, red team descriptions, or sabotage objectives.

Guide for creating sabotage trajectories (strajs) in Control Tower. Use when the user asks about creating strajs, monitor evasion, attack strategies, straj documentation, or trajectory development workflows.

Guide for monitoring trajectories in Control Tower. Use when the user asks about monitoring, suspiciousness scores, ct run monitor command, monitor types (basic, strong, gpt5, multi), or trajectory analysis. IMPORTANT - Use this skill instead of the MCP tools for guidance on monitoring trajectories.

Guide for running evaluations in Control Tower. Use when the user asks about running evals, ct run eval, eval policies (honest, attack, trusted), task sets, uploading results, replaying trajectories, or evaluation options. IMPORTANT - Use this skill instead of the MCP tools for guidance on running evaluations.

Creates isolated git worktrees for this repository. Use when the user asks to make a worktree, create a branch checkout, isolate changes, copy .env, or link .settings into a worktree.