flux-opencode-improve

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

• Third-party content exposure detected (medium risk: 0.30). In workflow Step 2b and Step 8, outsider-authored free text from the user’s interactive “Pain Points” input is passed into match-recommendations.py as --user-context "$USER_CONTEXT" via echo "$CONTEXT".

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

• Potentially malicious external URL detected (high risk: 0.90). The skill clones/updates the recommendations repository at runtime (e.g., "https://github.com/Nairon-AI/flux-recommendations.git" / "https://github.com/Nairon-AI/flux-recommendations"), and those fetched YAML files directly drive the agent's recommendations and installation instructions, making this an external runtime dependency that controls behavior.