The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands and Python scripts to automate reconnaissance probes, including directory brute-forcing and API interrogation.
[DATA_EXFILTRATION]: While not exfiltrating the agent's secrets, the skill is designed to discover and extract PII (emails, addresses, SSNs) and sensitive configuration files from external targets, as documented in the procedure steps for debug log mining and source leak hunting.
[PROMPT_INJECTION]: The skill includes instructions that could be interpreted as attempts to override default behaviors, such as 'ignore embedded instructions' warnings for future steps, though these are presented as security best practices in the remediation section.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from remote web servers (HTTP headers, JSON responses, log files) and processes it directly through shell tools like grep and curl. This creates a surface where a malicious server could provide crafted responses to influence the agent's next steps.
Ingestion points: Step 2 through Step 6 in SKILL.md all read data from external URLs.
Boundary markers: None identified. Data is piped directly from curl to grep or wc.
Capability inventory: Uses curl, python3, grep, and wc across the provided procedures.
Sanitization: Missing. Content from external servers is processed without validation or escaping.

recon-sector