doc-drafter
Pass
Audited by Gen Agent Trust Hub on Oct 5, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from user-provided briefs and external reference documents which could contain malicious instructions.
- Ingestion points: The skill reads reference documents (old contracts, templates) via a Google Drive connector and processes text-based briefs provided by the user in the conversation as mentioned in
SKILL.mdandreferences/doc-types.md. - Boundary markers: Safety Rule 7 in
SKILL.mdexplicitly instructs the agent to treat these materials as untrusted and to prioritize user instructions over any text found within the source material. - Capability inventory: The skill possesses the ability to write files to Google Drive and create document objects within a PandaDoc account, potentially enabling automated data movement based on injected instructions if safety rules are bypassed.
- Sanitization: While the skill uses bracketed placeholders (e.g.,
[NEEDS: ...]) for missing information, there is no explicit instruction to sanitize or escape the content of the generated Markdown draft to prevent it from influencing downstream processing.
Audit Metadata