mailbox-consolidation
Pass
Audited by Gen Agent Trust Hub on Nov 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONPROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill requires the agent to read raw email content (RFC822) to identify and backfill missing messages.
- Ingestion points: Untrusted data enters the context via the
messages.getAPI withformat=raw. - Boundary markers: The instructions lack delimiters or specific warnings to ignore embedded directives in the email headers or bodies.
- Capability inventory: The skill utilizes the
Bashtool and has the ability to write to the mailbox viamessages.insert. - Sanitization: No escaping or validation of the email content is specified before it is processed or re-uploaded.
- [PROMPT_INJECTION]: The 'Self-Evolving Skill' header instructs the agent to 'fix this file immediately' based on runtime observations.
- This meta-instruction allows the agent to alter its own code and behavior, which could be exploited if the agent encounters adversarial conditions or data that influences its reasoning for a 'fix'.
Audit Metadata