create-avatar

Fail

Audited by Snyk on Nov 19, 2026

Risk Level: CRITICAL
Full Analysis

CRITICAL E004: Prompt injection detected in skill instructions.

  • Potential prompt injection detected (medium risk: 0.30). The skill is a legitimate Atmee avatar builder with its own vendor domains and safeguards, but it has two disclosed-yet-sketchy items: a Codex sandbox change that applies to every session on the machine, and an instruction to withhold tool limitations from the creator.

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

  • Third-party content exposure detected (low risk: 0.10). The required workflow reads the creator's own uploaded documents and any links the creator supplies for the character draft (drafting.md, Step 3), which are arbitrary pages the creator selected, so an outsider can publish poison but cannot force delivery.

MEDIUM W009: Direct money access capability detected (payment gateways, crypto, banking).

  • Direct money access detected (medium risk: 0.30). Decisive route: the sponsoring tools (grant_sponsor, update_sponsor, withdraw_sponsor) move the creator's credits to sponsor an avatar. Rule 3 says credits move "only to sponsor an avatar when the creator asks and confirms." Moving credits to fund a sponsor pool is treated as a direct financial transfer, not a service cost. The paid generation calls (portraits, voices, seeds) are service costs for the agent's own generation and are not counted. Direct operation, so the control test applies. Control evidence: the sponsoring route is documented as working "signed in," meaning it requires an authenticated session to the Atmee service, which is a service-enforced access check. The confirm: true flag and the "ask first" rule are agent-followed and do not count as the control, but the sign-in requirement does. Because the route is direct with documented authentication, the tier is medium, not high. The evidence is thin (the skill does not explicitly describe how the service enforces sign-in for the sponsor route), so this is a borderline call; the recall-first override was considered.

Issues (3)

E004
CRITICAL

Prompt injection detected in skill instructions.

W011
MEDIUM

Third-party content exposure detected (indirect prompt injection risk).

W009
MEDIUM

Direct money access capability detected (payment gateways, crypto, banking).

Audit Metadata
Risk Level
CRITICAL
Analyzed
Nov 19, 2026, 10:58 PM
Issues
3