skills/pluginagentmarketplace/custom-plugin-ai-red-teaming/responsible-disclosure/Gen Agent Trust Hub
responsible-disclosure
Pass
Audited by Gen Agent Trust Hub on Oct 16, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes vulnerability data (such as descriptions, reproduction steps, and proof of concept code) to generate professional reports and track timelines. This represents an ingestion point for untrusted data that could potentially contain malicious instructions intended to influence the agent's behavior.
- Ingestion points: Data fields within the
VulnerabilityReportclass (SKILL.md) and input parameters for theDisclosureTracker(scripts/disclosure-tracker.py). - Boundary markers: No explicit delimiters or instructions are provided to the agent to disregard embedded content within the vulnerability data fields.
- Capability inventory: The accompanying script (
scripts/disclosure-tracker.py) performs local file system writes to save disclosure reports (disclosure_report.json). - Sanitization: The provided logic does not include sanitization or validation of the input data before it is interpolated into templates or written to disk.
- [COMMAND_EXECUTION]: The included tracking script (
scripts/disclosure-tracker.py) performs standard local file operations (writing a JSON file). This is a functional capability necessary for the skill's stated purpose and does not involve arbitrary command execution or network access.
Audit Metadata