pm-improvement-skill-research

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

• Third-party content exposure detected (high risk: 0.90). SKILL.md's mandatory workflow (see "MANDATORY: Skill Improvement After Every Retrospective" step 2 and "Level 2: Research with MCP" -> "Reference URLs to Fetch") requires using MCP tools to WebSearch and fetch content from public sites like https://agent-skills.md and GitHub and then update/commit agent skill files based on that content, so untrusted, user-generated third‑party pages will be read and can materially influence agent behavior.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly requires using MCP tools at runtime to fetch external guidance (e.g., https://github.com/bmad-code-org/BMAD-METHOD and multiple https://agent-skills.md/... links) and then use that fetched content to update AGENT.md/skills which directly control agent prompts/instructions, so this is a high-risk runtime dependency.