goose

Warn

Audited by Gen Agent Trust Hub on Oct 8, 2026

Risk Level: MEDIUMMETADATA_POISONINGEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [METADATA_POISONING]: The skill's description identifies 'Goose' as a social media marketing and automation tool, but the provided official documentation link (https://pkg.go.dev/github.com/kevinburke/goose/v3) refers to a database migration tool. This mismatch creates a risk where an agent might perform database modifications under the impression it is managing social media content.
  • [EXTERNAL_DOWNLOADS]: The skill instructs users to install the @membranehq/cli package via NPM. While this is a vendor-provided tool, it introduces a dependency on external code that is executed locally.
  • [COMMAND_EXECUTION]: The integration relies on shell commands executed via the Membrane CLI to manage authentication and interact with the Goose API.
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from the external Goose API, which could contain malicious instructions.
  • Ingestion points: API responses captured from the terminal output of membrane action run and membrane request commands (found in SKILL.md).
  • Boundary markers: No explicit delimiters or instructions are provided to the agent to distinguish between legitimate data and potential instructions within the API output.
  • Capability inventory: The agent has the ability to execute system commands via the Membrane CLI.
  • Sanitization: The instructions do not define any sanitization or validation protocols for the external data ingested from the API.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Oct 8, 2026, 11:34 PM