github-repo-monitor

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 1.00). The automation intentionally fetches and forwards plaintext user secrets (including the GITHUB PAT and LLM/api keys) into spawned agent conversations and grants the agent terminal/file_editor tools, enabling credential access, exfiltration, and remote code execution by the spawned agent.

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

• Third-party content exposure detected (high risk: 0.95). Yes: at runtime the automation polls GitHub issue/PR comments (outsider-authored free text in comment["body"] and issue/PR body/recent comment bodies) and embeds them into the LLM prompt via _build_initial_prompt(...) → create_conversation(..., initial_message=...).

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

• Potentially malicious external URL detected (high risk: 0.90). The automation calls the agent server at runtime (e.g. f"{agent_url}/api/settings" and f"{url}/api/settings/secrets/{name}") to fetch agent settings and user secrets which are then injected into spawned conversations (LLM settings, tools, and secret lookups), meaning remote content from the agent server directly controls the agent configuration/behavior.