membership-internal-subscription-grant

Pass

Audited by Gen Agent Trust Hub on Sep 9, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied data such as email addresses, subscription IDs, and descriptive reasons which are subsequently used in write-capable tools. Although the workflow mandates manual dry-runs and human confirmation steps that act as significant mitigations against automatic obedience to embedded instructions, it represents an ingestion surface for untrusted data.\n
  • Ingestion points: User-provided email addresses, plan details, and note fields used in the create/modify/delete workflows (SKILL.md).\n
  • Boundary markers: No explicit delimiters are specified in the instructions for the input data; safety is managed via the mandatory dry-run and confirmation workflow.\n
  • Capability inventory: Write access to subscription data via classic_create_internal_subscription, classic_update_internal_subscription, and classic_delete_internal_subscription (SKILL.md).\n
  • Sanitization: Input sanitization is not explicitly mentioned; the logic relies on human operator review of the payload during the dry-run phase.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 9, 2026, 02:57 PM
Security Audit — agent-trust-hub — membership-internal-subscription-grant