dx-packaging-push-upgrade-configure

Pass

Audited by Gen Agent Trust Hub on Nov 7, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the Salesforce Dev Hub, such as subscriber organization names (OrgName) and package version metadata. This content is used to build rollout plans and display reports. If an attacker controls a subscriber organization's name or other metadata within the Dev Hub, they could attempt to inject malicious instructions that influence the agent's behavior during the plan generation phase.
  • Ingestion points: Data retrieved via sf data query from the PackageSubscriber, MetadataPackage, and MetadataPackageVersion objects (identified in SKILL.md Phases 2 and 5).
  • Boundary markers: The skill explicitly implements a 'generate → preview → confirm → run' workflow to ensure that the agent does not execute push upgrades or aborts without human oversight.
  • Capability inventory: The skill uses sf package push-upgrade schedule to initiate upgrades in remote subscriber environments and performs local file writes to create batch CSV files (identified in SKILL.md Phase 4).
  • Sanitization: The skill relies on human-in-the-loop validation, requiring the user to verify the inline CSV contents and the command parameters before execution.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 7, 2026, 01:06 AM