The Agent Skills Directory

[PERSISTENCE]: The skill provides instructions to modify user shell configuration files to maintain state across sessions.
Evidence: The setup section explicitly commands the agent to append export statements to ~/.zshrc or ~/.bashrc to store API keys.
[CREDENTIALS_UNSAFE]: The skill manages sensitive API keys by writing them into a local YAML configuration file and shell profile scripts.
Evidence: Instructions for creating ~/.config/subfinder/config.yaml include fields for VirusTotal, SecurityTrails, Censys, and Shodan keys. The CHAOS_API_KEY is also stored in shell profile files.
[INDIRECT_PROMPT_INJECTION]: The reconnaissance pipeline ingests and processes large volumes of untrusted data from third-party sources and target web servers.
Ingestion points: Data is pulled from certificate logs (crt.sh), historical archives (waybackurls), and active crawls (katana) of external target websites.
Boundary markers: The skill does not define clear delimiters or "ignore embedded instructions" warnings when processing these external data streams.
Capability inventory: The skill has extensive shell command execution capabilities, including curl, subfinder, nuclei, katana, and custom Python scripts for JS analysis.
Sanitization: The scripts use basic filtering tools like jq, sed, and grep, but they do not implement rigorous sanitization of content before processing it through the toolchain.
[TIME_DELAYED_CONDITIONAL]: The skill includes logic for continuous, scheduled execution of monitoring tasks.
Evidence: Includes Bash scripts for monitoring new subdomains and GitHub commits, with instructions to schedule them using crontab at various intervals.

web2-recon