github-actions-cache-poisoning

Pass

Audited by Gen Agent Trust Hub on Nov 9, 2026

Risk Level: SAFEDATA_EXFILTRATIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill documentation describes techniques for exfiltrating sensitive environment variables, such as /proc/self/environ and CI/CD runtime tokens, to an external server.
  • [COMMAND_EXECUTION]: The skill provides functional shell and Python script templates intended to manipulate the GitHub Actions Cache API and inject malicious lifecycle scripts into package manifests.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests and analyzes untrusted workflow and manifest files from target repositories, creating an attack surface where instructions embedded in data could influence agent behavior. 1. Ingestion points: .github/workflows/*.yaml files and package manifests like package.json. 2. Boundary markers: None. 3. Capability inventory: File reading, grep search, and configuration analysis. 4. Sanitization: None identified in the skill logic.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 9, 2026, 12:00 PM