claude-software-factory

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

• Third-party content exposure detected (medium risk: 0.30). The required loop has the PM read GitHub PR review threads and review-bot findings (which include comments from any commenter on the PR) via gh api graphql and gh pr checks, and an outsider can post comments on a PR without the agent selecting a specific item, so this is medium under the GitHub PR reading rule.

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

• Potentially malicious external URL detected (medium risk: 0.30). The only inbound source is the instructed npx skills@1.5.23 add stealth-factory/skills install of related skill material, which is a setup-time repository locator, but publisher provenance rests only on an unverified community owner name with no first-party or independently authenticated evidence, so it is common/reputable unofficial and rated medium.