audit-bot

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

• Third-party content exposure detected (high risk: 0.75). Required workflow Step 1 ingests outsider-authored free text/prose into the LLM context via curl fetching trades_broker.csv from a Pi API endpoint and via scp of ffm_diag.log/bot_ml.log* logs that are not authored by the operating user; these files are then read by the audit scripts and can be included in generated report text.

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

• Attempt to modify system services in skill instructions detected (high risk: 0.70). The skill explicitly instructs using sudo on the Raspberry Pi to copy root-owned files and change their permissions (sudo cp + sudo chmod), which performs privileged filesystem modifications and thus pushes the agent to perform elevated operations that alter machine state.