AGENT LAB: SKILLS

autonomous-skill

Warn

Audited by Snyk on Feb 15, 2026

Risk Level: MEDIUM
Full Analysis

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

  • Third-party content exposure detected (high risk: 0.80). The skill explicitly supports a network mode (the --network option in scripts/run-session.sh) which runs codex with --dangerously-bypass-approvals-and-sandbox and even gives the example "Fetch data from GitHub API and analyze", meaning the agent can access and ingest public third‑party/web API content that could carry untrusted, user-generated input.

MEDIUM W013: Attempt to modify system services in skill instructions.

  • Attempt to modify system services in skill instructions detected (high risk: 0.90). The skill explicitly instructs use of Codex flags like --dangerously-bypass-approvals-and-sandbox and --full-auto (full access, file edits + network), which encourages bypassing security/sandbox protections and enabling autonomous file-system and network actions that can compromise the host.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Feb 15, 2026, 09:12 PM