Skip to main content
rule CX Codex

pentest-agents-mistakes — Claude Code rule: testing pitfalls

pentest-agents-mistakes is a rule that collects 11 mistakes agents make in security testing, from hallucinated paths to fake PoCs. Read it before a vulnerability hunt.

no license ★ 1,298 updated 2026-06-12

Install into GitHub Copilot · Manually: put the files in place All methods and platforms ↓
.github/instructions/pentest-agents-mistakes.instructions.md
Open on GitHub

What pentest-agents-mistakes does

This is a registry of real failures. Each rule came from a session where an agent wasted time, got corrected, or shipped a weak finding. The top 11 cover the essentials: save artifacts to disk, never invent file paths, validate before reporting, use a real browser for WAF and JavaScript checks, prove impact with actual data not theory, test sibling endpoints and fields, match CVSS version to the platform, read policy before the first request, only call a finding CONFIRMED when you have a working PoC on the live target, recalibrate after correction.

One rule stops a common trap: satisfying gate checkers with markers instead of evidence. If writing a string clears a gate but no real work happens, the gate fails. Other rules flag sneakier problems: re-asking for credentials or tokens that are already in session memory, listing findings when asked to rank them, using placeholder usernames like researcher when the real identity is in policy.md.

What it helps with

  • Distinguish a real proof of impact from a fingerprint or chain of inference
  • Keep findings persisted so the next session doesn't re-discover everything
  • Pick the right CVSS scoring system for each bug bounty platform
  • Break the pattern when the same mistake repeats
  • Stop agents from gaming checklists with fake markers instead of work

Example prompts for the agent

  • Read pentest-agents-mistakes at the start of /hunt on a new scope
  • If an agent repeats the same mistake twice, point it to the relevant rule
  • Verify credentials and test accounts are loaded into session memory before kicking off a recon phase

How to install pentest-agents-mistakes

Manually: put the files in place

.github/instructions/pentest-agents-mistakes.instructions.md

The applyTo field sets which files it applies to

Step by step, with commands for each platform: AGENTS.md vs CLAUDE.md: what they are and which one to use

Questions about pentest-agents-mistakes

How do I install pentest-agents-mistakes in Claude Code?

Copy the file .claude/rules/pentest-agents-mistakes.md from the repository into your local .claude/rules/ folder. Claude Code will load the rule automatically in the next session.

What is pentest-agents-mistakes?

A collection of 11 behavioral rules for agents hunting bugs in web applications and APIs. Each rule documents a real failure and how to avoid it.

Why does CVSS version matter when choosing a bounty platform?

HackerOne requires CVSS 3.1, while Bugcrowd, Intigriti, and Immunefi use CVSS 4.0. Submitting the wrong version forces a triage rework and delays payment.

Similar IDE rules

All IDE rules for Agent setup: memory, context, orchestration →

From the same set43

The whole set →
hunt hunt — a prompt for systematic vulnerability detection on target hosts. Reads scope, detects tech stack, auto-selects vuln classes by likelihood, and runs targeted tests with concrete payloads and WAF bypass rules. CX Codex +1 ★ 1,298repo stars hunt Conducts targeted vulnerability hunting by validating scope, fingerprinting stacks, and dispatching exploit tests. CX Codex +1 ★ 1,298repo stars autopilot autopilot coordinates security testing for a target by checking scope and program rules before dispatching specialized subagents. Autonomous mode runs through its completion checks before reporting. CX Codex +1 ★ 1,298repo stars autopilot autopilot — a Claude Code skill that runs a hunt across all vulnerability classes for a target and keeps the agent from calling it done too early. CX Codex +1 ★ 1,298repo stars hunt-idor hunt-idor guides your agent in auditing IDOR and BOLA flaws across APIs, GraphQL schemas, and multi-tenant SaaS platforms. CX Codex +1 ★ 1,298repo stars hunt-xss hunt-xss — a skill for Claude Code, Cursor, Codex, GitHub Copilot, Gemini CLI, and Windsurf that teaches the agent to find XSS vulnerabilities — DOM-based, stored, reflected, mXSS, and newer variants. CX Codex +1 ★ 1,298repo stars hunt-info-disclosure hunt-info-disclosure — a skill for tracking down information disclosure bugs: exposed .env and .git files, debug endpoints, Spring Boot actuator paths and leaked repo secrets. CX Codex +1 ★ 1,298repo stars hunt-business-logic hunt-business-logic finds business logic flaws like price tampering, race conditions, and 2FA bypasses. Useful for testing checkout workflows and subscription logic. CX Codex +1 ★ 1,298repo stars

Project sponsors

Companies that keep this analytics open