arc-interactive AGENTS.md

Repository instructions for an agent that designs and implements ARC-AGI-3 games. ARC-AGI is a benchmark that tests whether an AI can solve new visual puzzles under the same rules and interface used by humans.

In plain words
What is it for?
Use it to specify grid-based games, define players, targets, walls, hazards, actions, and win or loss conditions, then implement the game in the required environment.
Why use it?
It gives a structured design process so games have clear rules, fair clues, and goals that people can understand. It also defines the information needed before implementation begins.

Instructions file for CodexOpenCode

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add instructions/theredbluepill/arc-interactive/agents-md
Clone the repo
git clone --depth 1 https://github.com/theredbluepill/arc-interactive

Made for: Codex, OpenCode.

Per session 5,009 This file is loaded in full into every session.
When invoked 5,009 The same file — it is already loaded in full.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.05009 $0.05009
Opus 5 $0.02505 $0.02505
Sonnet 5 $0.01002 $0.01002
Haiku 4.5 $0.00501 $0.00501

Measured 3d ago against content hash 9e15177911f6, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

arc-interactive AGENTS.md scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 3d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

AGENTS.md · 467 lines

How it starts

The opening of the file, as written. The whole thing — 467 lines — stays where its author put it; the contents beside it link to each section on GitHub.

ARCAGI-3 Game Designer Agent

Role

Agent responsible for designing and implementing ARC-AGI-3 games.

Benchmark principle

ARC-AGI-style benchmarks are about whether an AI can solve puzzles that humans can solve when both operate under the same official interface: defined actions, observation format, and stated game rules. The goal is to measure general problem-solving, not which system had the most task-specific training data or memorized solutions.

When you author games in this repo, treat human solvability (given that shared spec) as a design guide: prefer well-posed levels—where goals are clear from layout, mechanics, or fair in-observation cues—over under-specified puzzles that only yield to guessing, huge search, or spoilers absent from the agent’s observation.

Workflow

1. Game Design Phase

Input: Game concept or requirements Output: Game specification

Key Questions:

  • Grid size? (8x8, 16x16, 24x24, 64x64)
  • What entities? (player, targets, walls, hazards)
  • What actions? (define what ACTION1-7 mean for your game)
  • Win/lose conditions?

2. Implementation Phase

Input: Game specification Output: Working game in environment_files/

Steps:

  1. Create directory: environment_files/{stem}/{version}/ (two-letter stem + digits, e.g. ez01; version folder is usually v1 on first landing, then often an 8-char git prefix after CI — see CONTRIBUTING.md)
  2. Implement {stem}.py with:
    • Sprite definitions
    • Static levels (no PCG)
    • Game class extending ARCBaseGame
    • Win/lose conditions
  3. Test with: arc.make using the full game_id from that folder’s metadata.json, or locally uv run python run_game.py --game {stem} --version auto — add --mode human for pygame hand-play (scripts/human_play_pygame.py).

3. Documentation Phase

Input: Completed game Output: Updated tracking files

Steps:

  1. Add entry to GAMES.md with all metadata columns
  2. If you discover a reusable pattern (not stem-specific), add a short bullet under Lessons learned (cross-repo) below; otherwise rely on GAMES.md, {stem}.py, and scripts/render_arc_game_gif.py (see skill generate-arc-game-gif in skills/)
  3. Optional: add assets/{stem}.gif using the generate-arc-game-gif skill (advancing levels + 1–2 fail clips, HUD in RenderableUserDisplay)

Read the full file on GitHub · 467 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 3d ago First seen · 467 lines · 5,009 tokens per session scan A 9e15177911f6

Subscribe to this mod's changes

arc-interactive AGENTS.md is an instructions file published in the GitHub repository theredbluepill/arc-interactive (54 stars, last pushed 2mo ago), licensed MIT. It adds 5,009 tokens to every session, about $0.0250 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.