harness

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

• Third-party content exposure detected (high risk: 0.80). The harness skill (see SKILL.md examples like "Example 1: Research Harness" and the README "Deep Research" prompt, and the required reference references/agent-design-patterns.md which endorses a "general-purpose" agent type with WebSearch/WebFetch) explicitly instructs creating researcher agents (media/community/researcher) that fetch and ingest public web, news, and forum/social content, so the agent will read untrusted third‑party user-generated content that can influence orchestration and decisions.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill includes scripts that fetch remote repository content at runtime—scripts/install.sh can git clone https://github.com/revfactory/harness (via _direct_install) and scripts/sync-upstream-harness.sh pulls files from https://raw.githubusercontent.com/revfactory/harness/main using curl—which would import external SKILL/agent definitions that directly influence agent prompts and behavior if those scripts are executed.