What is agentguard AGENTS.md?

A set of project instructions for AgentGuard, a tool for reviewing the security of software that uses AI models or agents. It covers threats such as unsafe tool use, prompt injection, memory, retrieval, and multi-agent interactions.

Use it when threat-modeling or auditing agentic and large-language-model software, including MCP tools, tool calling, memory, retrieval, and secure-by-design changes.

It gives coding agents a defined process for security reviews and connects findings to OWASP and CWE security categories. This makes risks easier to organize and address.

agentguard AGENTS.md — Instructions file by Ethonik

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/ethonik/agentguard/agents-md

Clone the repo

git clone --depth 1 https://github.com/Ethonik/agentguard

Made for: Codex, OpenCode.

Per session 1,461 This file is loaded in full into every session.

When invoked 1,461 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.

Model	Per session	Once invoked
Fable 5	$0.01461	$0.01461
Opus 5	$0.00731	$0.00731
Sonnet 5	$0.00292	$0.00292
Haiku 4.5	$0.00146	$0.00146

Measured 2d ago against content hash 1b44483cfdbf, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

agentguard 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 · 106 lines

How it starts

The opening of the file, as written. The whole thing — 106 lines — stays where its author put it; the contents beside it link to each section on GitHub.

AGENTS.md — AgentGuard

Universal, agent-neutral instructions. This file is read natively by most coding agents (OpenAI Codex, Cursor, Google Antigravity/Jules/Gemini CLI, opencode, GitHub Copilot, Windsurf, Zed, Aider, Devin, and 30+ others via the open AGENTS.md standard). Claude Code additionally uses the richer skills/agentguard/SKILL.md.

What AgentGuard is

AgentGuard secures software that uses AI/LLMs and agents as components (tool-calling, memory/RAG, MCP, multi-agent). It has two workflows and maps every finding to OWASP Top 10 for Agentic Applications 2026 (ASI01–ASI10), OWASP Top 10 for LLM Applications 2025 (LLM01–LLM10) and CWE (MITRE).

When to act as AgentGuard

Act as AgentGuard when the user asks to review / audit the security of agentic or LLM code, or for secure-by-design advice on an agentic feature (threat model, controls, checklist). Triggers: "agentic security", "seguridad agéntica", "OWASP LLM/ASI review", "prompt injection", "tool misuse", "MCP security", "threat model de agentes".

Engine location

The engine is a self-contained directory holding scripts/ and references/. Locate it before running anything:

In this repository: skills/agentguard/.
When installed into another project by install.sh: .agentguard/.
Claude Code install: ~/.claude/skills/agentguard/ or .claude/skills/agentguard/.

Below, <engine> means whichever of these exists. The scripts need only Python 3 (stdlib, no dependencies).

Hard rules (do not break)

Read-only on the target project. Never modify the code you audit. The only file you write is the report.
The scripts read files as text and never import or execute target code. Run them with python3 directly.
Every finding maps to an OWASP ID (ASI/LLM) and a CWE when one applies, and carries actionable remediation. If a purely-agentic failure has no native CWE (goal hijack, memory poisoning, behavioral drift), report it with the ASI ID alone — never force a CWE.
Defensive/authorized use only. No exploits, no offensive tooling.

Read the full file on GitHub · 106 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.

2d ago First seen · 106 lines · 1,461 tokens per session scan A 1b44483cfdbf

Subscribe to this mod's changes

agentguard AGENTS.md is an instructions file published in the GitHub repository Ethonik/agentguard (2 stars, last pushed 1mo ago), licensed MIT. It adds 1,461 tokens to every session, about $0.0073 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.

Building agents Security

agentguard AGENTS.md

Getting it into your agent

What it costs to keep this loaded

Grade A, and why

Nothing flagged

How it starts

AGENTS.md — AgentGuard

What AgentGuard is

When to act as AgentGuard

Engine location

Hard rules (do not break)

What this file has done since we first saw it

Other instructions, from other repositories

spellbook AGENTS.md

open-supermarkets AGENTS.md

helm copilot-instructions.md

flyto-core CLAUDE.md

fusion-skills AGENTS.md

aeon CLAUDE.md