follow-up

Pass

Audited by Gen Agent Trust Hub on Sep 2, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill's documentation instructs the user to install a helper package from the vendor using the command npx -y @coldiq/mcp@latest. This is an expected dependency for utilizing the skill's full features.
  • [INDIRECT_PROMPT_INJECTION]: The skill utilizes data from external sources (social media engagement and signals) to inform the content of generated emails, creating a potential path for untrusted data to influence agent output.
  • Ingestion points: Data is ingested through tools like mcp__coldiq__find_signals and mcp__coldiq__extract_post_engagement (referenced in SKILL.md).
  • Boundary markers: The skill instructions do not currently include delimiters or specific prompts to compartmentalize or ignore instructions that might be embedded in the retrieved social data.
  • Capability inventory: The skill is primarily designed to generate text for emails and read internal reference templates; it does not have high-risk capabilities like arbitrary command execution or file system modification.
  • Sanitization: There is no evidence of sanitization or filtering applied to the external data before it is used to draft email content.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 2, 2026, 08:55 AM
Security Audit — agent-trust-hub — follow-up