doc-coauthoring
Pass
Audited by Gen Agent Trust Hub on Aug 26, 2026
Risk Level: SAFEPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from various sources which could contain malicious instructions.
- Ingestion points: Processes content from user-provided templates, files, and external integrations such as Slack, Google Drive, and SharePoint (referenced in SKILL.md under Stage 1).
- Boundary markers: No specific delimiters or instructions to ignore embedded commands within the ingested data are defined.
- Capability inventory: The skill utilizes
create_file,str_replace, and sub-agent invocation to draft and test documents (referenced in SKILL.md under Stage 2 and Stage 3). - Sanitization: There is no mention of filtering or sanitizing content pulled from external integrations before it is passed to the drafting engine or sub-agents.
- [DATA_EXPOSURE]: The workflow actively encourages users to provide sensitive organizational context, including team dynamics, technical dependencies, and internal politics (referenced in SKILL.md under 'Info Dumping'). While this is functional for co-authoring, it results in the ingestion of highly sensitive data into the agent's operational context.
Audit Metadata