apply-agent-research

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

• Third-party content exposure detected (high risk: 0.75). The skill’s runtime LLM context is not shown; however, the required workflow ingests outsider-authored free text into the agent only via the knowledge base mirror (agent-research-knowledge clone) and host governance docs (e.g., CONTEXT.md, CLAUDE.md, docs/adr/, docs/agents/installed-skills.md), which are then used to generate proposals—these are outsider-authored repo content not chosen by the operating user.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill explicitly requires cloning and reading external git repositories at runtime—notably the agent-research-knowledge repo and the dividedby/skills repo—whose fetched content is consumed as the knowledge/input that directly drives the agent's proposal generation (i.e., controls its prompts), so these runtime fetches are external dependencies that influence agent behavior.