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.
npx agentmods add instructions/rubys/roundhouse/agents-mdgit clone --depth 1 https://github.com/rubys/roundhouseWhat 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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.01220 | $0.01220 |
| Opus 5 | $0.00610 | $0.00610 |
| Sonnet 5 | $0.00244 | $0.00244 |
| Haiku 4.5 | $0.00122 | $0.00122 |
Grade A, and why
roundhouse 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 2d 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.
How it starts
The opening of the file, as written. The whole thing — 88 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Working on Roundhouse (agents & contributors)
Roundhouse reads Rails source and emits standalone projects in ~12 target languages, plus an inference engine (LSP/MCP/in-browser IDE) that types Rails without annotations. This file is the orientation an AI agent or new contributor needs before touching the code: where to look, and the invariants not to break.
Source of truth for current state is README.md — which
targets are live, benchmark numbers, what works today. It is kept current; the
older docs below are accurate on architecture but may narrate migrations that
have since landed. When a status claim anywhere disagrees with README or CI,
README and CI win.
Start here
| You want… | Read |
|---|---|
| What the project is / current state / the numbers | README.md — fresh, authoritative |
The dev loop, roundhouse-ast, adding an IR variant |
DEVELOPMENT.md |
| Pipeline internals (analyze / lower / emit / runtime / verification) | docs/pipeline/ — architecture, not status |
| Compiler inputs (Ruby+ERB, schema/routes/seeds, method catalog, DB adapter) | docs/data/ |
| Why do this at all (the argument, option value) | WHY.md |
| Why this attempt is different (lineage, the three bets, risks) | BETS.md |
Pipeline shape: Ruby AST → analyze (typed IR) → lower (target-neutral IR) → emit (per-target project + runtime/<target>/ glue). Key files are mapped in
DEVELOPMENT.md § "Pipeline at a glance."
Invariants — do not break these
These are the rules the codebase enforces or depends on. Violating one is a defect even if the build is green.
- Zero error diagnostics is the contract. The subset of Rails we
transpile is defined as "produces no error diagnostics." Warnings are the
modeling-debt ledger and are expected; errors are the invariant. Guarded
by
tests/real_blog.rs(ingests_without_errorsplustype_analysis_coverage, the zero-error + zero-unresolved-type gate).
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.
- 2d ago First seen · 88 lines · 1,220 tokens per session scan A fc82d5562116
roundhouse AGENTS.md is an instructions file published in the GitHub repository rubys/roundhouse (89 stars, last pushed 2d ago), licensed Apache-2.0. It adds 1,220 tokens to every session, about $0.0061 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.
Other instructions, from other repositories
sqlglot AGENTS.md
Instructions for tobymao/sqlglot, covering contributing to sqlglot, about sqlglot, development commands, installation and basic installation.
oxc copilot-instructions.md
Instructions for oxc-project/oxc: When reviewing pull requests in this repository.
compiler-explorer copilot-instructions.md
Instructions for compiler-explorer/compiler-explorer: For each answer to the user, evaluate your level of confidence in the correctness of the answer on a scale from 1 to 10, where 1 is very uncertain and 10 is absolutely certain. If your confidence level is below 8, state so and suggest ways to verify the answer.
neva AGENTS.md
AGENTS.md instructions for nevalang/neva, covering neva engineering guide, project model, repository map, documentation and ai engineering harness.
sky CLAUDE.md
Instructions for anzellai/sky, covering claude.md, claude code — operational rules, 0. goal fidelity in autonomous loops — inviolable, 0.1 remote-push discipline — minimize ci noise and 1. memory safety — scripts/mem-guard.sh must run during dev.
forem AGENTS.md
AGENTS.md instructions for forem/forem, covering forem ai context & rules, meta: ai instruction synchronization, general philosophy, background workers & sidekiq and testing standards.