The Agent Skills Directory

[PROMPT_INJECTION]: The skill implements 'Agent Obedience Engineering' patterns across multiple files (e.g., RULES.md, references/README.md). It uses mandatory execution contracts ('CRITICAL: Route Execution Contract', 'ACTION REQUIRED') and an 'Excuse Rebuttal Table' specifically designed to override the agent's internal safety reasoning and force compliance with high-risk offensive tasks.
[CREDENTIALS_UNSAFE]: Multiple modules, particularly references/attack-chain/README.md, provide explicit workflows for harvesting hardcoded credentials, including cloud provider keys (AWS AK/SK), database connection strings, JWT keys, and VPN credentials.
[DATA_EXFILTRATION]: The references/attack-chain/ module contains instructions for exfiltrating sensitive data from targets using curl and grep to collect API keys, tokens, and passwords from source code and configuration files.
[PRIVILEGE_ESCALATION]: The skill provides detailed, actionable instructions for escalating privileges on Windows (using the 'Potato' family of exploits) and Linux (via SUID and sudo abuse), as documented in references/attack-chain/README.md and references/pwn-chain/.
[PERSISTENCE]: Dedicated workflows exist for establishing persistence on compromised hosts, including the implantation of SSH keys, creation of crontab backdoors, LD_PRELOAD hijacking, and modification of Windows Registry Run keys.
[COMMAND_EXECUTION]: The repository relies on numerous PowerShell and Bash wrapper scripts (e.g., references/apk-reverse/scripts/frida-run.ps1, references/diagram-generator/scripts/render_diagram.py) that execute arbitrary system commands and spawn subprocesses.
[EXTERNAL_DOWNLOADS]: The bootstrap-reverse scripts automate the download and installation of security tools from external sources like GitHub and NPM. While the author claims to verify hashes, this remains a significant supply chain risk when executed by an autonomous agent.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process highly untrusted external data, such as malicious APKs, binaries, and PCAP files, which can contain hidden instructions targeting the analysis agent.
[DYNAMIC_EXECUTION]: Several scripts use dynamic execution patterns, such as python -c to execute strings generated at runtime (found in frida-run.ps1) and subprocess.run to call dynamically resolved binaries in the diagram-generator module.

reverse-engineer