OpenParable
Skip to content
← Back to skills

Codeguard

Development
codeguard

Codeguard is an Agent Skill for development published in the cisco-ai-defense/defenseclaw repository on GitHub: a SKILL.md instruction file an AI agent can load to do this task.

Security-aware code generation — teaches the agent CodeGuard rules to write secure code by default

View on GitHub ↗
QUICK START

How to use this skill

Bring this guide into your coding agent with a prompt tailored to the tool you use.

  1. Open your project in Codex.
  2. Copy and paste the prompt below.
  3. Review files and risks before approving any action.
Prompt to paste
Inspect the Agent Skill at https://github.com/cisco-ai-defense/defenseclaw/blob/HEAD/skills/codeguard/SKILL.md and all supporting files. Report provenance, permissions, scripts, dependencies, and security risks. Propose an exact installation into .agents/skills/codeguard, but wait for my approval before writing files or running scripts.

Copying this prompt does not install or run the skill. Review third-party files before use.

CodeGuard: Secure Code Generation Rules

You MUST follow these security rules when writing code. Code that violates these rules will be blocked by the DefenseClaw CodeGuard scanner before it reaches disk. Write it correctly the first time.


Credentials — all languages

CG-CRED-001: Never hardcode API keys or secrets

Assigning API keys, secret keys, access tokens, or private keys directly in source code exposes them in version control and build artifacts.

# BAD — blocked
api_key = "sk-proj-abcdefghij1234567890"
secret_key = "wJalrXUtnFEMI/K7MDENG/bPxRfiCYEXAMPLEKEY"

# GOOD
import os
api_key = os.environ["API_KEY"]
// BAD — blocked
const accessToken = "ghp_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx";

// GOOD
const accessToken = process.env.ACCESS_TOKEN;

CG-CRED-002: Never include AWS access key IDs

AWS keys that start with AKIA followed by 16 alphanumeric characters are detected regardless of context.

# BAD — blocked
aws_key = "AKIAIOSFODNN7EXAMPLE"

# GOOD
import boto3
session = boto3.Session()  # uses ~/.aws/credentials or IAM role

CG-CRED-003: Never embed private keys (CRITICAL)

PEM-encoded private keys (-----BEGIN RSA PRIVATE KEY----- and variants) are the highest severity finding. They grant full authentication as the key holder.

# BAD — blocked (CRITICAL severity)
KEY = """-----BEGIN RSA PRIVATE KEY-----
MIIEpAIBAAKCAQEA...
-----END RSA PRIVATE KEY-----"""

# GOOD — load from file or secrets manager at runtime
with open("/etc/ssl/private/server.key") as f:
    key = f.read()

Command Execution — Python, JavaScript, TypeScript, Ruby, PHP

CG-EXEC-001: Never use os.system(), eval(), exec(), or child_process.exec()

These functions pass strings to a shell interpreter, enabling command injection when any part of the string comes from user input or external data.

# BAD — blocked
import os
os.system(f"grep {user_input} /var/log/app.log")

# GOOD
import subprocess
subprocess.run(["grep", user_input, "/var/log/app.log"], check=True)
// BAD — blocked
const { exec } = require("child_process");
exec(`ls ${userDir}`);

// GOOD
const { execFile } = require("child_process");
execFile("ls", [userDir], callback);
# BAD — blocked
result = eval(user_expression)

# GOOD
import ast
result = ast.literal_eval(user_expression)

CG-EXEC-002: Never use shell=True in subprocess (Python)

Even with subprocess, passing shell=True re-introduces shell injection risk.

# BAD — blocked
subprocess.call(f"convert {infile} {outfile}", shell=True)

# GOOD
subprocess.run(["convert", infile, outfile], check=True)

SQL — Python, JavaScript, TypeScript, Ruby, PHP, Java

CG-SQL-001: Never format strings into SQL queries

String interpolation in SQL enables SQL injection. Always use parameterized queries with bind variables.

# BAD — blocked
cursor.execute(f"SELECT * FROM users WHERE name = '{username}'")
cursor.execute("SELECT * FROM users WHERE id = %s" % user_id)

# GOOD
cursor.execute("SELECT * FROM users WHERE name = ?", (username,))
cursor.execute("SELECT * FROM users WHERE name = %s", (username,))
// BAD — blocked
db.query(`SELECT * FROM users WHERE id = ${userId}`);

// GOOD
db.query("SELECT * FROM users WHERE id = ?", [userId]);
// BAD — blocked
stmt.execute("SELECT * FROM users WHERE id = " + userId);

// GOOD
PreparedStatement ps = conn.prepareStatement("SELECT * FROM users WHERE id = ?");
ps.setInt(1, userId);

Deserialization — Python

CG-DESER-001: Never use pickle or yaml.load on untrusted data

pickle.load/pickle.loads can execute arbitrary code during deserialization. yaml.load without a safe Loader is equally dangerous.

# BAD — blocked
import pickle
obj = pickle.loads(request.data)

import yaml
config = yaml.load(user_yaml)

# GOOD
import json
obj = json.loads(request.data)

import yaml
config = yaml.safe_load(user_yaml)

Cryptography — Python, JavaScript, TypeScript, Java, Go, Ruby

CG-CRYPTO-001: Never use MD5 or SHA1

MD5 and SHA1 are cryptographically broken. Use SHA-256 or stronger.

# BAD — blocked
import hashlib
h = hashlib.md5(data)
h = hashlib.sha1(data)

# GOOD
import hashlib
h = hashlib.sha256(data)
// BAD — blocked
crypto.createHash("md5").update(data);
crypto.createHash("sha1").update(data);

// GOOD
crypto.createHash("sha256").update(data);

Network — Python, JavaScript, TypeScript, Go

CG-NET-001: Validate outbound URLs

HTTP requests to URLs constructed from variables can enable SSRF (Server-Side Request Forgery). Validate and allowlist target URLs.

# CAUTION — flagged for review
response = requests.get(user_url)

# GOOD — validate against allowlist
ALLOWED_HOSTS = {"api.example.com", "cdn.example.com"}
parsed = urllib.parse.urlparse(user_url)
if parsed.hostname not in ALLOWED_HOSTS:
    raise ValueError("URL not allowed")
response = requests.get(user_url)

Path Safety — all languages

CG-PATH-001: Never construct paths with ../

Path traversal sequences (../) allow escaping intended directories to read or write arbitrary files.

# BAD — blocked
path = os.path.join(upload_dir, "../../etc/passwd")

# GOOD — canonicalize and validate
real = os.path.realpath(os.path.join(upload_dir, filename))
if not real.startswith(os.path.realpath(upload_dir)):
    raise ValueError("path traversal detected")

Quick Reference

RuleSeverityLanguagesInstead ofUse
CG-CRED-001HIGHallapi_key = "sk-..."os.environ["API_KEY"]
CG-CRED-002HIGHallAKIA... in sourceIAM roles / ~/.aws/credentials
CG-CRED-003CRITICALall-----BEGIN PRIVATE KEY-----Secrets manager / file at runtime
CG-EXEC-001HIGHpy,js,ts,rb,phpos.system(), eval()subprocess.run([...])
CG-EXEC-002MEDIUMpyshell=Truesubprocess.run([...])
CG-SQL-001HIGHpy,js,ts,rb,php,javaf-strings in SQLParameterized queries
CG-DESER-001HIGHpypickle.loads()json.loads()
CG-CRYPTO-001MEDIUMpy,js,ts,java,go,rbhashlib.md5()hashlib.sha256()
CG-NET-001MEDIUMpy,js,ts,gorequests.get(var)URL allowlist validation
CG-PATH-001MEDIUMall../../etc/passwdos.path.realpath() + prefix check