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/the-open-engine/zeroshot/agents-mdgit clone --depth 1 https://github.com/the-open-engine/zeroshotWhat 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 | $0.22816 | $0.22816 |
| Opus 5 | $0.11408 | $0.11408 |
| Sonnet 5 | $0.04563 | $0.04563 |
| Haiku 4.5 | $0.02282 | $0.02282 |
Grade C, and why
zeroshot AGENTS.md scanned grade C with 2 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.
Asks for rootmediumPrivilege escalation
A mod that escalates privileges can change anything on the machine, not only the project.
OMP's Docker isolation is env/broker-only, zero-automatic-mount, and sessionless — never same-shaped as the mount-based providers. Its registry `docker` entry has no `mount`: `~/.omp`, `agent.db`, WAL/SHM files, and host Reads agent configuration directoriesmediumAgent snooping
.claude/, .codex/, .gemini/ hold keys, settings and other credentials a mod has no legitimate need for.
OMP's Docker isolation is env/broker-only, zero-automatic-mount, and sessionless — never same-shaped as the mount-based providers. Its registry `docker` entry has no `mount`: `~/.omp`, `agent.db`, WAL/SHM files, and host How it starts
The opening of the file, as written. The whole thing — 1,356 lines — stays where its author put it; the contents beside it link to each section on GitHub.
UPDATE THIS FILE when making architectural changes, adding patterns, or changing conventions.
Zeroshot: Multi-Agent Coordination Engine
Operational rules and references for automated agents working on this repo. Install:
npm i -g @the-open-engine/zeroshot or npm link (dev).
CRITICAL RULES
- Never spawn without permission. Do not run
zeroshot run <id>unless the user explicitly asks to run it. - Never use git in validator prompts. Validate files directly.
- Never ask questions. Agents run non-interactively; make autonomous decisions.
- Never edit
CLAUDE.mdunless explicitly asked to update docs. - Detached (
-d) runs must forward allzeroshot runoptions viaZEROSHOT_RUN_OPTIONS(seebuildDaemonEnv+buildStartOptions) so PR/worktree config cannot be dropped. mainis the single development and release trunk. Target normal PRs atmain; never recreate a long-liveddev -> mainrelease-promotion flow.- Pull request titles are Conventional Commit headers because squash merge makes the title the
released commit. For Node-owned changes,
fix:/perf:publish patches,feat:publishes minors, breaking syntax publishes majors, anddocs:/chore:intentionally publish nothing. - Node, Zeroshot Rust, and the Python SDK release independently from
main. Node ownsvX.Y.Zand automatic semantic releases. Rust uses explicitzeroshot-rust-vX.Y.Zreleases. Python useszeroshot-python-vRUST_SDKtags and PEP 440RUST.postSDKpackage versions; Rust releases automatically publish SDK revision1, while later SDK revisions remain separately triggerable. Its PyPI distribution iszeroshot-rust, while its import package remainszeroshot. Rust and Python releases must never affect the next Node version or generated Node notes. - Checked-in publication manifests are non-authoritative development versions. Release tags, npm
metadata, and GitHub Releases are authoritative; automation must never commit versions to
main. - Curated notes live at
docs/releases/vX.Y.Z.md. Recovery may operate only from an immutablevX.Y.Ztag whose exact commit is an ancestor ofmain, and it must never overwrite an existing npm version or GitHub Release. - Provider output-silence liveness checks are opt-in (
enableLivenessCheck: true). Recovery tests that exercise stale-agent termination must enable the watchdog explicitly. - Isolation copies must reuse the shared pinned-root boundary in
src/copy-containment.tsfor traversal, directory creation, synchronous copies, and worker copies. Revalidate the source and destination immediately before every filesystem effect; never reconstruct unchecked effect paths.
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 · 1,356 lines · 22,816 tokens per session scan C 85f410ec9c20
zeroshot AGENTS.md is an instructions file published in the GitHub repository the-open-engine/zeroshot (1,815 stars, last pushed 2d ago), licensed MIT. It adds 22,816 tokens to every session, about $0.1141 per session on Opus 5. A static security scan graded it C with 2 findings (asks for root, reads agent configuration directories). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other instructions, from other repositories
VisionClaw-Agent-Public-Release AGENTS.md
Instructions for Huskyauto/VisionClaw-Agent-Public-Release, covering agents.md — working in the visionclaw codebase, identity, the cardinal discipline: search before reading, hard rules (non-negotiable) and architecture map (entrypoints).
sample-autonomous-cloud-coding-agents AGENTS.md
AGENTS.md instructions for aws-samples/sample-autonomous-cloud-coding-agents, covering agents.md, your role, commands (run these), git workflow and where to make changes.
babysitter CLAUDE.md
Claude Code instructions for a5c-ai/babysitter, covering claude.md, repo-specific overrides, read these first, quick commands and claude code local notes.
swarm-357 CLAUDE.md
Instructions for TechTideOhio/swarm-357, covering claude.md, what this repo is, python package (packages/techtide-swarm/), or from pypi and key classes.
harness CLAUDE.md
Instructions for puristajs/harness, covering claude guidance and subagents v1.0.
COORD-Harness AGENTS.md
AGENTS.md instructions for 0marm0/COORD-Harness, covering coordharness agent entrypoint, start and finish and context hierarchy.