setup-harness
Warn
Audited by Snyk on Nov 19, 2026
Risk Level: MEDIUM
Full Analysis
MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).
- Third-party content exposure detected (low risk: 0.10). The install check reads public registry metadata (npm, PyPI, Go proxy, crates.io, NuGet, Maven Central, Docker Hub) for each catalog-named package, including publisher-controlled fields such as maintainers and repository, and reads the glue of hook repos. Outsiders can publish to these registries, but the agent must select a specific package rather than receive a feed, so this is low.
MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).
- Potentially malicious external URL detected (medium risk: 0.30). The only inbound locator the skill instructs is the printed
npx skills@1.8.1 add Gharib89/skillscommand, which points to an individually owned repository, so it is unofficial provenance and yields a medium tier, while the Anthropic plugin, Maven, Eclipse, Docker and Playwright sources are strong first-party publishers.
MEDIUM W013: Attempt to modify system services in skill instructions.
- Attempt to modify system services in skill instructions detected (low risk: 0.30). The skill writes only project-scoped files (committed
.claude/hooks and settings,scripts/check.sh,CLAUDE.md) and never asks for sudo, system file edits, user account creation, or ssh/systemd changes, so it sits mostly in the allowed project-specific category. I raise it slightly because it installs third-party tools and a cloud SessionStart script that installs them, which changes machine state, though it is gated by an install check and human confirmation.
Issues (3)
W011
MEDIUMThird-party content exposure detected (indirect prompt injection risk).
W012
MEDIUMUnverifiable external dependency detected (runtime URL that controls agent).
W013
MEDIUMAttempt to modify system services in skill instructions.
Audit Metadata