service-digital-engagement-messaging-site-integrate

Pass

Audited by Gen Agent Trust Hub on Aug 28, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes site configuration metadata (such as content.json and homeGuestLayout.json) retrieved from a Salesforce org. This represents an indirect prompt injection surface as malicious content within these files could potentially influence the agent's behavior during patching or verification.
  • Ingestion points: Site metadata files retrieved via sf project retrieve (specified in SKILL.md).
  • Boundary markers: Absent in the instructions.
  • Capability inventory: Includes shell execution for patching scripts, metadata deployment via the Salesforce CLI (sf project deploy), and network requests via curl for verification.
  • Sanitization: The skill uses jq with variable binding (--arg) to safely modify JSON content, which prevents shell-level injection during the patching process.
  • [DYNAMIC_EXECUTION]: The helper scripts patch_lwr_bundle.sh and patch_aura_bundle.sh include a fallback to dynamically execute a small Python snippet to generate UUIDs if the uuidgen system utility is not available.
  • Evidence: python3 -c 'import uuid; print(uuid.uuid4())' in scripts/patch_lwr_bundle.sh and scripts/patch_aura_bundle.sh.
  • [COMMAND_EXECUTION]: The skill relies on executing shell scripts and various CLI tools (sf, jq, curl) to perform its primary function of metadata manipulation and deployment. This is the intended purpose of the skill and is carried out using standard, authorized tools.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 28, 2026, 01:32 PM
Security Audit — agent-trust-hub — service-digital-engagement-messaging-site-integrate