The Agent Skills Directory

[PROMPT_INJECTION]: The toolkit includes a dedicated 'Sneaky Bits' utility in tools/sneaky_bits.py that uses Unicode characters (U+2062 and U+2064) to hide malicious instructions from human eyes. It also provides pre-built payloads in tools/hai_payload_builder.py designed to force models to ignore previous instructions, extract their system prompts, and escalate severity levels. Several files like skills/web2-vuln-classes/SKILL.md and skills/bug-bounty/SKILL.md were flagged for containing patterns that override system instructions.
[REMOTE_CODE_EXECUTION]: Both the README.md and install_tools.sh scripts instruct users to execute remote code from untrusted sources using the curl | sh pattern. This includes installation scripts for Ollama and Homebrew. Furthermore, the tools/cve_scan.sh script downloads a remote Python script from a GitHub user for CVE-2018-7600 scanning and executes it.
[COMMAND_EXECUTION]: The skill provides the AI agent with extensive shell access through scripts like recon_engine.sh, vuln_scanner.sh, and hunt.py. These scripts wrap dozens of external security tools including nmap, sqlmap, nuclei, and ffuf, allowing the agent to perform broad network operations and local system commands.
[DATA_EXFILTRATION]: tools/secrets_hunter.sh is designed to scan for and extract verified API keys, tokens, and credentials from filesystems, git history, and JavaScript bundles. When combined with the toolkit's automated network capabilities (curl, httpx), this creates a high risk of sensitive data exfiltration.
[CREDENTIALS_UNSAFE]: The demo/app.py and demo/README.md files contain multiple hardcoded sensitive secrets, including fake AWS Access Keys and JWT signing keys. While used for demonstration purposes, these files contribute to a 'high' risk profile and were flagged by reputation scanners.
[EXTERNAL_DOWNLOADS]: The toolkit frequently pulls content from external domains. Automated scanners flagged book.hacktricks.xyz as a blacklisted domain, although it is a known security wiki. The toolkit also pulls various wordlists and nuclei templates from various remote GitHub repositories during initialization.

bug-bounty