The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides instructions to modify the user's shell configuration (~/.zshrc) and suggests setting up cron jobs to automate reconnaissance tasks, establishing persistence on the system. It also pipes untrusted data from remote targets directly into inline Python scripts for extraction.
[DATA_EXFILTRATION]: The skill reads and writes sensitive configuration files, such as ~/.config/subfinder/config.yaml, to store API keys for third-party services. It also interacts with target APIs that require Authorization headers.
[PROMPT_INJECTION]: The skill has a large ingestion surface, pulling data from sources like crt.sh, the Chaos API, and target JavaScript files. This untrusted data is processed by tools including katana, nuclei, and custom Python scripts without explicit boundary markers or sanitization, creating a surface for indirect prompt injection. Capability inventory includes network operations (curl), file writes, and script execution (python3).
[REMOTE_CODE_EXECUTION]: Untrusted content from target JavaScript bundles is piped into inline Python scripts for extraction. While the scripts use regex for parsing, this pattern of piping remote data directly into an interpreter is a significant security surface that requires user oversight.

web2-recon