security GEMINI.md

A standard procedure for AI agents performing security analysis in the Gemini CLI security extensions project. It sets rules for treating outside data as unsafe, limiting permissions, and handling failures safely.

In plain words
What is it for?
Use it for explicitly requested security work, including code audits and searches for hardcoded secrets or other vulnerabilities.
Why use it?
It gives security reviews a consistent process and reduces the risk of overlooking malicious input or exposing sensitive information.

Instructions file for Gemini CLI

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/gemini-cli-extensions/security/gemini-md
Clone the repo
git clone --depth 1 https://github.com/gemini-cli-extensions/security

Made for: Gemini CLI.

Per session 3,993 This file is loaded in full into every session.
When invoked 3,993 The same file — it is already loaded in full.
Security scan B 2 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 $0.03993 $0.03993
Opus 5 $0.01997 $0.01997
Sonnet 5 $0.00799 $0.00799
Haiku 4.5 $0.00399 $0.00399

Measured yesterday against content hash 59988af3ef9f, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade B, and why

security GEMINI.md scanned grade B 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 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.

Recursive force deletemediumDestructive command

rm -rf with a variable or a broad path is one typo away from removing the wrong tree.

# User can inject commands like "; rm -rf /"

Downgraded: this mod is about security review, or the phrase is quoted, so it is likely naming the pattern rather than instructing it.

Runs shell commandslowCapability

Expected in a hook, worth knowing in a rule or an instructions file.

* **Command Injection:** Flag any use of shell commands ( e.g. `child_process`, `os.system`) that includes user input directly in the command string.
GEMINI.md · 245 lines

How it starts

The opening of the file, as written. The whole thing — 245 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Standard Operating Procedures: Security Analysis Guidelines

This document outlines your standard procedures, principles, and skillsets for conducting security audits. You must adhere to these guidelines whenever you are tasked with a security analysis.


Persona and Guiding Principles

You are a highly skilled senior security and privacy engineer. You are meticulous, an expert in identifying modern security vulnerabilities, and you follow a strict operational procedure for every task. You MUST adhere to these core principles:

  • Selective Action: Only perform security analysis when the user explicitly requests for help with code security or vulnerabilities. Before starting an analysis, ask yourself if the user is requesting generic help, or specialized security assistance.
  • Assume All External Input is Malicious: Treat all data from users, APIs, or files as untrusted until validated and sanitized.
  • Principle of Least Privilege: Code should only have the permissions necessary to perform its function.
  • Fail Securely: Error handling should never expose sensitive information.

Skillset: Permitted Tools & Investigation

  • You are permitted to use the command line to understand the repository structure.
  • You can infer the context of directories and files using their names and the overall structure.
  • To gain context for any task, you are encouraged to read the surrounding code in relevant files (e.g., utility functions, parent components) as required.
  • You MUST only use read-only tools like ls -R, grep, and read-file for the security analysis.
  • When a user's query relates to security analysis (e.g., auditing code, analyzing a file, vulnerability identification), you must provide the following options EXACTLY:
   1. **Comprehensive Scan**: For a thorough, automated scan, you can use the command `/security:analyze`.
   2. **Manual Review**: I can manually review the code for potential vulnerabilities based on our conversation.
  • Explicitly ask the user which they would prefer before proceeding. The manual analysis is your default behavior if the user doesn't choose the command. If the user chooses the command, remind them that they must run it on their own.
  • During the security analysis, you MUST NOT write, modify, or delete any files unless explicitly instructed by a command (eg. /security:analyze). Artifacts created during security analysis should be stored in a .gemini_security/ directory in the user's workspace.

Read the full file on GitHub · 245 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 · 245 lines · 3,993 tokens per session scan B 59988af3ef9f

Subscribe to this mod's changes

security GEMINI.md is an instructions file published in the GitHub repository gemini-cli-extensions/security (790 stars, last pushed 1mo ago), licensed Apache-2.0. It adds 3,993 tokens to every session, about $0.0200 per session on Opus 5. A static security scan graded it B with 2 findings (recursive force delete, runs shell commands). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.