terse-create

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.30). In SKILL.md’s runtime workflow, the job code consumes the event payload at trigger time (via filter and onTrigger, and explicitly via event.formatForAgentRunner() passed into generateText), meaning outsider-authored free text from the triggering system (e.g., GitHub/Linear/Slack content) can be ingested by the LLM without any pre-selection step.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.80). The skill explicitly instructs runtime fetches of external docs (https://docs.useterse.ai/llms.txt and https://docs.useterse.ai/core-concepts/durability) to discover and pull pages used to implement the workflow, meaning remote content fetched at runtime can directly shape prompts/instructions.