writeup-ingest
Pass
Audited by Gen Agent Trust Hub on Sep 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to fetch and process data from external, untrusted sources which could contain malicious payloads designed to influence agent behavior.
- Ingestion points: The agent is directed to use web search to query primary sources such as
hackerone.com,portswigger.net,assetnote.io, and vendor security advisories. - Boundary markers: The skill provides a '6-question extraction gate' and 'Citation-honesty rules' to validate data, which act as conceptual boundaries but do not provide technical sandboxing for the input.
- Capability inventory: The agent is authorized to modify local files, specifically targeting
/root/.opencode/skills/<class>/SKILL.mdandbb-ops/memory/knowledge/writeups/README.mdto insert new techniques and payload sketches. - Sanitization: The instructions require the agent to re-verify CVE IDs via NVD/MITRE and condense findings into one-sentence root causes, which may partially mitigate the risk of raw injection pass-through by forcing abstraction.
Audit Metadata