The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The file references/threat-intelligence.md contains multiple functional reverse shell payloads and remote code execution patterns for Bash, Python, PowerShell, PHP, Ruby, and Node.js. This file was flagged as malicious malware (FileRepMalware) by automated scanners.
[COMMAND_EXECUTION]: The skill provides instructions to use the Bash tool and includes reference files containing dangerous system commands, such as rm -rf / and pipe-to-shell execution patterns (e.g., curl | bash).
[DATA_EXFILTRATION]: Reference documentation includes detailed code snippets and techniques for exfiltrating sensitive data—including environment variables, SSH keys, AWS credentials, and browser data—to external Command and Control (C2) servers.
[OBFUSCATION]: The skill provides numerous examples and templates for obfuscating malicious intent through techniques like multi-layer Base64 encoding, Hex escapes, XOR encryption, and string reversal.
[CREDENTIALS_UNSAFE]: The reference files contain regex patterns and logic for systematically harvesting secrets, API keys, and connection strings from source code, configuration files, and environment variables.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and analyze untrusted codebases, which represents a massive attack surface. A malicious codebase could attempt to trigger the skill's shell execution or agent-spawning capabilities to perform unauthorized operations on the host system.
Ingestion points: Read, Grep, and Glob tools used to systematically scan all files in a target repository.
Boundary markers: Uses instruction-based 'safety rules' to tell the agent to treat input as data, which is insufficient to prevent sophisticated injection.
Capability inventory: Full access to shell (Bash), file modification (Write, Edit), and sub-agent creation (Agent).
Sanitization: No programmatic sanitization or filtering of the untrusted data is implemented prior to analysis.

cybersecurity