quasi-coder

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

• Third-party content exposure detected (high risk: 0.85). The skill’s runtime workflow is to interpret collaborator-provided shorthand marked by start-shorthand/end-shorthand and convert ()=> lines into code; that shorthand is outsider-authored free text (e.g., another user’s messages/files) that the agent reads and injects into its LLM context for transformation.

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (medium risk: 0.60). The skill explicitly tells the agent to "execute the requested action (run commands, create files...)" and to perform non-code actions that can modify the filesystem or run shell commands, which can change the machine's state even though it doesn't explicitly request sudo, system-file edits, or user creation.