setup-skills

Fail

Audited by Snyk on Nov 19, 2026

Risk Level: CRITICAL
Full Analysis

CRITICAL E004: Prompt injection detected in skill instructions.

  • Potential prompt injection detected (medium risk: 0.30). The skill is a legitimate repo-setup workflow with disclosed behavior, but the ADO reviewer prompt tells the reviewer to skip line-by-line review of derived skill files, which weakens a protection over files that drive agent actions beyond what the task needs; this is risky but shows no malicious intent.

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

  • Third-party content exposure detected (medium risk: 0.30). The required reviewer detection in step 3 reads comments on the last ten merged PRs (via gh api) for the Codex connector, so outsider-authored PR comment text is fetched on every run without the agent selecting an item, and the Claude reviewer scaffold it installs ingests PR diffs and linked issue text.

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

  • Potentially malicious external URL detected (medium risk: 0.30). The skill instructs installing sibling skills with npx skills add Gharib89/skills at an unpinned ref, and the installed skill material governs the agent. That repo is individual-owned and community-maintained, with no first-party vendor provenance, so the tier is medium. The other install sources (the mattpocock/skills, upstash/context7 and humanlayer/skills commit pins) do not raise it above medium.

Issues (3)

E004
CRITICAL

Prompt injection detected in skill instructions.

W011
MEDIUM

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

W012
MEDIUM

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

Audit Metadata
Risk Level
CRITICAL
Analyzed
Nov 19, 2026, 11:12 AM
Issues
3