agent-collab AGENTS.md

Development instructions for agent-collab, a Node-based marketplace that lets one AI coding agent delegate work to other agents. It includes commands, project rules, platform details, and testing guidance.

In plain words
What is it for?
Use it when developing agent-collab, running its build or test commands, working on Codex or Cursor integrations, or checking changes before merging.
Why use it?
It gives contributors the project-specific rules needed to change the code safely and verify that changes work across its supported environments.

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/zebbern/agent-collab/agents-md
Clone the repo
git clone --depth 1 https://github.com/zebbern/agent-collab

Made for: Codex, OpenCode.

Per session 2,030 This file is loaded in full into every session.
When invoked 2,030 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 $0.02030 $0.02030
Opus 5 $0.01015 $0.01015
Sonnet 5 $0.00406 $0.00406
Haiku 4.5 $0.00203 $0.00203

Measured 2d ago against content hash 57cf639f71a6, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

agent-collab 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.

AGENTS.md · 131 lines

How it starts

The opening of the file, as written. The whole thing — 131 lines — stays where its author put it; the contents beside it link to each section on GitHub.

agent-collab — Agent Guide

Claude Code marketplace with two plugins that delegate work to other AI coding agents: plugins/codex (community fork of openai/codex-plugin-cc) and plugins/cursor (drives Cursor's cursor-agent CLI). Plain Node .mjs with JSDoc types — no build step for runtime code, no dependencies beyond dev tools.

Commands

npm run verify                          # THE pre-merge gate: versions + build + native suite + Linux suite in docker
npm test                                # full suite (~2 min; caps file concurrency at 8)
node --test tests/<file>.test.mjs       # one file
node --test --test-name-pattern="..." tests/<file>.test.mjs   # one test
npm run build                           # regenerates app-server types + tsc checkJs
  • Verify test results by exit code, never by grepping output.
  • tests/runtime-*.test.mjs and the cancel e2e tests spawn real detached processes; prefer targeted patterns while iterating, but run the full suite (or rely on CI) before claiming a semantics change is safe — "targeted" has missed pinned contracts before.
  • npm test runs scripts/run-tests.mjs, which caps file concurrency at min(cores, 8). The process-spawning e2e tests starve their detached workers under unbounded parallelism on high-core machines (a 32-core box would run 32 files at once), so a ceiling is required — but a hardcoded --test-concurrency=8 is an absolute value, not a max, and would raise concurrency on a 2-core CI runner, oversubscribing it and blowing the state lock's warn-and-proceed timeout (lost updates in the concurrency test). The runner's min(cores, 8) caps high-core boxes without ever inflating low-core ones. Run a single file directly with node --test tests/<f>.
  • Linux behavior is reproduced exactly by docker run node:22 with the repo copied in; ubuntu CI will agree with it.
  • npm run verify is the local merge gate; Required CI is the complementary remote gate and must be required in branch protection. The local gate runs version metadata, build, the native suite, and the full suite on Linux in docker (the Linux leg runs the ~40 win32-guarded tests the Windows run skips, so the two legs together cover more than either alone). A leg that cannot run is reported UNVERIFIED, the gate prints INCOMPLETE, and the process exits 2 — never silently "passed" to a human or automation. Run npm run verify:no-linux only when deliberately skipping docker.
  • Do not stream the docker leg's TAP output back over inherited stdio. Doing so reproducibly failed 7 timing-sensitive shared-broker e2e tests that pass every time otherwise — the pipe backpressure was being measured, not the code. scripts/verify.mjs keeps that output inside the container and echoes only a summary.

Read the full file on GitHub · 131 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. 2d ago First seen · 131 lines · 2,030 tokens per session scan A 57cf639f71a6

Subscribe to this mod's changes

agent-collab AGENTS.md is an instructions file published in the GitHub repository zebbern/agent-collab (32 stars, last pushed 17d ago), licensed Apache-2.0. It adds 2,030 tokens to every session, about $0.0102 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.