The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Automated scanners detected a malicious download link for a ZIP file (claude_bug_bounty_v3.6.zip) and a Windows setup installer flagged as a Trojan (hash 6f45ed67...). The repository owner Mikacr1138 serves these files from GitHub, and several core skill files (README.md, SKILL.md) have been identified as malicious by file reputation scanners.
[REMOTE_CODE_EXECUTION]: The skill facilitates the execution of remote code from unverified sources. Specifically, install_tools.sh downloads and runs a script via curl | bash, and full_hunt.sh pipes dynamic output from the crt.sh certificate database directly into a Python interpreter (python3), which is a critical execution risk.
[OBFUSCATION]: The skill implements ASCII smuggling techniques through scripts like sneaky_bits.py and hai_payload_builder.py. These use Unicode characters (U+2062 and U+2064) to hide malicious instructions or data exfiltration URLs within plaintext, designed to evade AI monitoring and human auditing.
[PROMPT_INJECTION]: A significant portion of the skill's instruction set consists of advanced prompt injection payloads. These are designed to override AI guardrails, extract system prompts, and manipulate the behavior of AI agents (e.g., for severity escalation or data exfiltration).
[DYNAMIC_EXECUTION]: Several Python tools (hunt.py, zero_day_fuzzer.py, cve_hunter.py) use subprocess.run(shell=True) to execute shell commands. This pattern is vulnerable to command injection if the input strings (such as target URLs or recon data) are not perfectly sanitized.
[COMMAND_EXECUTION]: The skill relies on and encourages the use of numerous powerful shell-based tools for network interaction, file manipulation, and system-level auditing without sufficient oversight.

bug-bounty