04-cold-email-infrastructure

Pass

Audited by Gen Agent Trust Hub on Aug 12, 2026

Risk Level: SAFEPROMPT_INJECTIONNO_CODE
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill's decision-making process for selecting infrastructure providers is driven by external metadata stored in capability.json and README.md files within the providers/ directory. If these metadata files are sourced from untrusted locations or tampered with, they could potentially influence the agent to select malicious or non-compliant providers.\n
  • Ingestion points: Processes provider metadata from the providers/ directory and selection logic from docs/PROVIDER_SELECTION.md.\n
  • Boundary markers: Implements a "fail loud" policy and requires explicit "compliance gates" to be cleared before selecting certain providers.\n
  • Capability inventory: Performs infrastructure provisioning including domain registration, mailbox creation, and DNS configuration.\n
  • Sanitization: Requires the agent to verify warm-up completion using the platform's internal status indicators rather than relying on self-declarations or time estimates.\n- [NO_CODE]: The skill is composed entirely of markdown instructions and does not include any accompanying scripts or executable files, limiting the risk of direct malicious code execution.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 12, 2026, 12:21 PM
Security Audit — agent-trust-hub — 04-cold-email-infrastructure