agent-manager

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

• Third-party content exposure detected (high risk: 0.80). The installer (scripts/install.sh) clones a public GitHub repo (https://github.com/fractalmind-ai/agent-manager-skill) and SKILL.md explicitly says skills resolved from ~/.claude/skills or repo paths (SKILL.md files) are injected as system prompts on agent start, so third‑party public skill files can be fetched and directly influence agent instructions and actions.

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

• Potentially malicious external URL detected (high risk: 0.90). The install script explicitly fetches https://github.com/fractalmind-ai/agent-manager-skill (via npx openskills or git clone) during installation, and the repository's SKILL.md files can be installed and later injected as system prompts when agents are started, meaning remote content can directly control agent instructions.

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

• Attempt to modify system services in skill instructions detected (high risk: 0.70). This skill explicitly includes actions that modify host state (e.g., "Write to system crontab"), runs install/setup scripts, and starts/stops tmux sessions — all of which can change the machine and may require elevated privileges, so it poses a moderate-to-high risk.