automate-me
Pass
Audited by Gen Agent Trust Hub on Jul 21, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill creates a surface for indirect prompt injection by ingesting and processing external workspace data to generate durable agent instructions.
- Ingestion points: The skill explicitly reads workspace-scoped transcripts, repository instructions,
.context/CONTINUITY.md, and handoff files to gather evidence of user preferences (SKILL.md, Section 2). - Boundary markers: There are no explicit instructions to the agent to distinguish between data-carrying preferences and embedded malicious instructions within these ingested files.
- Capability inventory: The skill uses the
skill-creatortool to write new executable instructions to the file system and supports Git operations (commit, push, PR) to persist these changes (SKILL.md, Section 5 and 6). - Sanitization: While the skill includes guardrails against including secrets or private excerpts, it does not explicitly sanitize the source text for adversarial instructions intended to subvert the resulting agent mode.
Audit Metadata