flux-publish

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

• Third-party content exposure detected (high risk: 0.80). The skill calls gh pr view to read PR title, branch and comments from GitHub (user-generated/untrusted content) and uses that title to populate the card and to infer the "Área" (changing field selection/behavior), and it also fetches form options from the Flux GraphQL endpoint (isengard.fluxcontrol.com.br) — together these external inputs are read and interpreted as part of the workflow and can change actions like which fields are set and whether to post a PR comment.

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

• Potentially malicious external URL detected (high risk: 1.00). The skill posts to the GraphQL endpoint https://isengard.fluxcontrol.com.br/api/graphql at runtime to fetch form options that directly determine the prompts/options presented and are required to create cards, so this external URL controls agent prompts during execution.