warren AGENTS.md

The main instruction guide for Warren, a system that runs AI coding agents as isolated workloads on infrastructure controlled by an operator. It explains the runtime, sandbox options, project settings, and Git delivery.

In plain words
What is it for?
Use it to understand agent adapters, run environments such as containers or Kubernetes pods, workspace branches, project configuration, and optional pull-request or tracker actions.
Why use it?
It gives agents the project's architecture and operating rules, including how runs are isolated, managed, recovered, and connected to Git workflows.

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/jayminwest/warren/agents-md
Clone the repo
git clone --depth 1 https://github.com/jayminwest/warren

Made for: Codex, OpenCode.

Per session 10,174 This file is loaded in full into every session.
When invoked 10,174 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 $0.10174 $0.10174
Opus 5 $0.05087 $0.05087
Sonnet 5 $0.02035 $0.02035
Haiku 4.5 $0.01017 $0.01017

Measured yesterday against content hash c0cc40ae2db7, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

warren 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 yesterday.

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 · 822 lines

How it starts

The opening of the file, as written. The whole thing — 822 lines — stays where its author put it; the contents beside it link to each section on GitHub.

AGENTS.md

This file is the canonical instruction doc for AI coding agents working in this repo, following the agents.md convention. CLAUDE.md is a symlink to this file, so one document serves every harness and there is nothing to drift.

What this project is

Coding agents are tools. Warren turns them into infrastructure.

It runs agent harnesses as isolated workloads on infrastructure the operator controls. Warren owns the workspace, lifecycle, limits, events, intervention, recovery, and Git delivery.

AgentRuntimeAdapter keeps the kernel harness-neutral. A harness needs an adapter before the runtime can drive it. The current distribution ships Pi and Claude Code adapters.

Runtime providers place each run in a local bwrap sandbox on Linux or sandbox-exec sandbox on macOS, a sibling Docker container, or a Kubernetes pod.

The kernel guarantees a pushed workspace branch. Project settings can add PR creation and tracker reactions around that boundary.

The fresh-install path is standalone. A user with a GitHub URL and the model credential for a shipped harness can dispatch a run end-to-end with no other os-eco tooling. Around that kernel, two os-eco data-plane tools integrate as opt-in features, not required infrastructure.

  • mulch — persistent agent memory across runs. Activated by a .mulch/ directory in the project.
  • seeds — the integrated issue queue agents read from and write to. Activated by a .seeds/ directory. Seeds is implementation #1 of the IssueTracker seam (src/tracker/contract.ts), with external trackers reachable through the RemoteTracker bridge — see docs/design/issue-tracker.md.

The agent registry is entirely inline. BUILTIN_AGENTS (src/registry/builtins/) ships seven agents and boot seeds them into the agents table on every start: claude-code, pi, planner, nightwatch, bugwatch, pr-fixer, and healer. There is no external agent library. GET /agents still reports source: "builtin" | "library" provenance. The library arm survives only for legacy rows.

Read the full file on GitHub · 822 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. yesterday First seen · 822 lines · 10,174 tokens per session scan A c0cc40ae2db7

Subscribe to this mod's changes

warren AGENTS.md is an instructions file published in the GitHub repository jayminwest/warren (346 stars, last pushed 3d ago), licensed MIT. It adds 10,174 tokens to every session, about $0.0509 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.

Related

Other instructions, from other repositories