oma CLAUDE.md

Repository guidance for oma, an open-source service that recreates Claude’s managed-agent API. It explains the project’s Cloudflare and self-hosted versions, including its pnpm monorepo—a repository containing multiple related packages.

In plain words
What is it for?
Use it when installing dependencies, running development servers, checking types, or testing oma across its Cloudflare and Node environments.
Why use it?
It gives Claude Code the project context, commands, and warnings needed to work safely in this codebase.

Instructions file

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/duyet/oma/claude-md
Clone the repo
git clone --depth 1 https://github.com/duyet/oma
Per session 1,471 This file is loaded in full into every session.
When invoked 1,471 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 $0.01471 $0.01471
Opus 5 $0.00736 $0.00736
Sonnet 5 $0.00294 $0.00294
Haiku 4.5 $0.00147 $0.00147

Measured 3d ago against content hash 68144bbc3d3b, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

oma CLAUDE.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.

CLAUDE.md · 52 lines

How it starts

The opening of the file, as written. The whole thing — 52 lines — stays where its author put it; the contents beside it link to each section on GitHub.

CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

What this is

Open-source, self-hostable reimplementation of the Claude Managed Agents API. A meta-harness: the platform (a SessionDO Durable Object) prepares what an agent has (tools, skills, history, sandbox, credentials); a harness decides how to drive the model loop. The same business logic runs two ways — Cloudflare (Workers + Durable Objects + Containers) and self-hosted Node (docker compose). pnpm monorepo (apps/*, packages/*).

The domain model, tool catalog, event types, harness API, and release process live in the imported @AGENTS.md below — don't restate them here.

Commands

  • Install: pnpm install (pnpm 11, Node ≥20). pnpm 11 hard-fails install if a native dependency isn't listed under allowBuilds: in pnpm-workspace.yaml.
  • Dev (Cloudflare simulators): pnpm dev — API on :8787, Console on :5173.
  • Typecheck: pnpm typecheck — root tsc --noEmit plus a separate node-only pass (typecheck:node); node packages are excluded from the root tsc.
  • Test: pnpm test — runs three suites in sequence (all required): root vitest run (Cloudflare Workers pool) → test:packagestest:console.
  • Single test (workers pool): pnpm vitest run path/to/file.test.ts -t "case name".
  • Single test (node-pool package): pnpm --filter @duyet/oma-session-runtime test -- run <file> -t "name".

Gotchas that will bite you

  • typecheck and test both gate PRs. .github/workflows/ci.yml runs both jobs on every PR and both block merge — a red test job is not advisory. Run pnpm typecheck && pnpm test locally before every push (or use /preflight) so CI doesn't catch it for you.
  • The vitest suite is split by pool. Root vitest run uses @cloudflare/vitest-pool-workers. Node-native packages ship their own vitest.config.ts with pool: "threads" and run only via pnpm --filter <pkg> test, excluded from the root run via its test.exclude list. test:packages wires up 10 of them: session-runtime, main-node, cap, integrations-adapters-node, sandbox, k8s-bridge, browser-harness, observability, vault-forward, and @getoma/cli. That is why pnpm test chains three commands. Adding a new node-pool package means wiring it into test:packages and the root exclude list — miss the second and its tests run twice, once in the wrong pool.
  • Internal @duyet/oma-* packages have no build step — they ship raw .ts (main/exports point at src/*.ts), resolved via workspace links plus a hand-maintained alias wall in vitest.config.ts. Adding a new package subpath may need a new alias entry there, or workers-pool tests won't resolve it. Only @getoma/cli and @getoma/sdk have a build.
  • Harnesses register by name in apps/agent/src/index.ts (registerHarness("default", …)); an agent selects one via harness: "<name>". The seam is apps/agent/src/harness/interface.ts. Use /new-harness to add one.
  • The prompt cache is byte-sensitive. deriveModelContext output must be byte-deterministic and <system-reminder> injections must sit in the cached prefix — any drift silently invalidates Anthropic's cache. See the contract comments in harness/interface.ts.
  • Model provider resolution is in apps/agent/src/harness/provider.ts — branches on ApiCompat (ant / ant-compatible / oai / oai-compatible); the OpenAI-compat path uses /chat/completions, never the Responses API.
  • SessionDO (apps/agent/src/runtime/session-do.ts) is the per-session Durable Object: append-only event log (DO-SQLite), streaming lifecycle, crash recovery. Streaming chunk/thinking/tool-input events are broadcast-only; the final agent.* event is the persisted record.
  • Credentials never enter the sandbox — injected by an outbound proxy; MCP calls proxy through the main worker. Don't add code paths that hand plaintext credentials to the agent DO.
  • Pagination: every paginated table exposes created_at + id, ordered (created_at, id) DESC; cursors are opaque and a stale cursor silently restarts from page 1. Helpers in packages/shared/src/pagination.ts.

Read the full file on GitHub · 52 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 · 52 lines · 1,471 tokens per session scan A 68144bbc3d3b

Subscribe to this mod's changes

oma CLAUDE.md is an instructions file published in the GitHub repository duyet/oma (5 stars, last pushed 14d ago), licensed Apache-2.0. It adds 1,471 tokens to every session, about $0.0074 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 2026-08-31.

Related

Other instructions, from other repositories

vscode buildNext.instructions.md

Working notes and architecture documentation for the new esbuild-based build system in build/next. Use when making changes to the new build pipeline (transpile/bundle commands, NLS plugin, source-map handling, resource copying, or self-hosting watch tasks).

microsoft/vscode · 6,785 tokens

spec-kit AGENTS.md

AGENTS.md instructions for github/spec-kit, covering agents.md, about spec kit and specify, quickstart — add a new integration in 5 steps, integration architecture and integrationmanifest — file tracking.

github/spec-kit · 7,104 tokens

codex AGENTS.md

AGENTS.md instructions for openai/codex, covering rust/codex-rs, the codex-core crate, code review rules, crate api surface and model visible context.

openai/codex · 5,182 tokens

langchain AGENTS.md

AGENTS.md instructions for langchain-ai/langchain, covering global development guidelines for the langchain monorepo, corridor security analysis, project architecture and context, monorepo structure and development tools & commands.

langchain-ai/langchain · 4,345 tokens

vscode oss-third-party-notices.instructions.md

Instructions for microsoft/vscode, covering vs code oss third-party-notices pipeline, architecture, pipeline flow in ci, applying the notice (cutover) and fallback chain (never fail the build).

microsoft/vscode · 5,001 tokens

next.js AGENTS.md

Instructions for vercel/next.js, covering next.js development guide, codebase structure, monorepo overview, core package: packages/next and other important packages.

vercel/next.js · 7,296 tokens