The Agent Skills Directory

Automated Processing of Untrusted Artifacts: The skill is designed to ingest and analyze external content including source code, vulnerability notes, and scanner findings. While these are necessary for the reporting task, there is a potential for indirect prompt injection if the processed artifacts contain instructions meant to influence the agent's behavior. The skill mitigates this by using structured sub-agent prompts and clear instructions for the AI to prioritize source code validation over provided notes.
Execution of Build and Validation Commands: To verify vulnerabilities, the skill instructs the agent to build PoCs using tools like make and run exploratory probes. These operations are conducted within the scope of user-provided lab access or VMs. The skill emphasizes that testing should never occur against live production targets without explicit instruction, and it encourages the use of disposable environments to maintain host security.
Management of Sensitive Technical Information: The skill handles sensitive data including vulnerability details, source code, and potentially lab access credentials. It follows best practices by ensuring produced reports are self-contained and do not include internal metadata, local absolute paths, or sensitive provenance details, ensuring the final output is safe for distribution.

vulnerability-writeup