implement-issue

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

• Third-party content exposure detected (high risk: 0.85). SKILL.md step 1 runs gh issue view <N> --comments, which ingests the issue body and comments authored by other GitHub users into the agent’s runtime context (outsider-authored free text via authenticated GitHub API).

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 network/git/gh commands "with sandboxing disabled", directing the agent to bypass the sandbox/security mechanism and thus compromise the machine's security posture.