The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches GitHub Actions from trusted vendors (Google, HashiCorp, AWS, GitHub) and reputable infrastructure tools (Aqua Security, Bridgecrew, Infracost, Gruntwork). Each resource is identified by a specific commit SHA for integrity.
[REMOTE_CODE_EXECUTION]: Executes automation scripts and security scanners via GitHub Actions. All external actions are pinned to specific 40-character commit SHAs to prevent supply chain attacks.
[COMMAND_EXECUTION]: Runs Terraform and Terragrunt CLI commands for infrastructure management. Commands are configured with security flags like -input=false and -no-color to ensure safe, non-interactive execution.
[PROMPT_INJECTION]: The skill processes untrusted pull request data and plan outputs.
Ingestion points: Pull request branch names (github.event.pull_request.head.ref) and Terraform plan outputs (steps.plan.outputs.stdout) are ingested from the environment.
Boundary markers: Plan outputs are wrapped in markdown triple backticks and HTML details tags within pull request comments.
Capability inventory: Employs actions/github-script to write and update comments on GitHub pull requests using the pull-requests: write permission.
Sanitization: Implements content truncation at 60,000 characters to prevent excessive content injection. Note: One template interpolates a branch name directly into a script string, which could allow for indirect injection if a malicious branch name contains escape characters.

terraform-github-actions-deploy