snap-forge

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

• Third-party content exposure detected (high risk: 0.80). The skill's "Orient" step explicitly instructs fetching GitHub issue comments via gh issue view <number> --comments (SKILL.md), which ingests public, user-generated GitHub content (e.g., "Durable Decisions" / phase headers) that the agent is expected to interpret and that can materially influence its actions.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly fetches GitHub issue comments at runtime via the command "gh issue view --comments" (i.e., the GitHub issue URL), and those comments—e.g., "## Durable Decisions" / "## Phase" plan sections—are read and used to direct the agent's behavior, so remote content can directly control agent instructions.