infra-ci-cd-github-actions

Pass

Audited by Gen Agent Trust Hub on Sep 20, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests project data such as source code, git history, and package configuration to generate optimized CI/CD workflows and determine affected packages. This creates a surface for indirect prompt injection if project files are controlled by an untrusted party.\n
  • Ingestion points: Reads local repository files including package.json, .github/workflows, and executes git diff to analyze changes as described in examples/testing.md and SKILL.md.\n
  • Boundary markers: Absent. There are no explicit instructions for the agent to treat content within project files as potentially untrusted data that should not be followed as instructions.\n
  • Capability inventory: The skill provides patterns for writing configuration files, executing shell commands via GitHub Action run: steps, and performing network operations for deployment and authentication.\n
  • Sanitization: Absent. Project data and shell outputs are used to filter tasks or configure builds without specific escaping or validation logic.\n- [COMMAND_EXECUTION]: The skill provides examples for executing system commands within CI environments for build tasks and runner maintenance.\n
  • Evidence: examples/core.md includes commands to free up disk space on GitHub Actions runners using sudo rm -rf /usr/share/dotnet and sudo rm -rf /opt/ghc. These are documented standard cleanup procedures for maximizing available storage on ephemeral GitHub-hosted runners.\n- [EXTERNAL_DOWNLOADS]: The skill references several GitHub Actions and CLI tools to be downloaded and used during CI execution from official and well-known sources.\n
  • Evidence: Mentions and provides configuration for standard actions such as actions/checkout, actions/cache, and official CLI tools like vercel and aws via package managers or setup actions.\n
  • Trusted Sources: All external references target well-known services (GitHub Marketplace, Vercel, AWS) or the author's own infrastructure (e.g., the metadata schema registry at agents-inc/cli).
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 20, 2026, 03:57 PM
Security Audit — agent-trust-hub — infra-ci-cd-github-actions