run-pre-merge-checks

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

• Third-party content exposure detected (high risk: 0.70). The required workflow runs the live e2e tests in tests/e2e/ against real upstream APIs and then parses their textual output/logs (e.g., via tee into /tmp/pre-merge-checks/...log and grep), which can include outsider-authored free text returned by external providers at runtime (HTTP/API responses or error strings) and then be injected into the LLM context when the skill renders the screenshot-ready summary from that output.

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

• Attempt to modify system services in skill instructions detected (high risk: 1.00). The skill explicitly instructs the agent to run every Bash invocation with dangerouslyDisableSandbox: true and to write to system paths (e.g. /tmp, ~/.cache/uv, the git working tree), i.e. it directs the agent to bypass sandbox/security protections and modify machine state.