The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill instructs the agent to retrieve sensitive authentication information. This includes reading tokens from ~/.config/configstore/firebase-tools.json and executing gcloud auth print-access-token. It also involves verifying API keys in terraform.tfvars for Namecheap DNS management.
[DATA_EXFILTRATION]: Extracted credentials and authentication tokens are intended to be sent to external API endpoints, specifically firebasehosting.googleapis.com, to manage hosting sites and custom domains.
[REMOTE_CODE_EXECUTION]: The skill provides templates for JavaScript scripts intended to be created at /tmp/firebase-create-domain.js and /tmp/firebase-list-domains.js, which are then executed by the agent using node to perform API calls.
[COMMAND_EXECUTION]: The instructions rely heavily on the bash tool to manage environment setup and deployment using Nix shell wrappers, such as nix shell nixpkgs#nodejs -c npm install and nix shell nixpkgs#firebase-tools -c firebase deploy.
[EXTERNAL_DOWNLOADS]: The build process downloads numerous third-party dependencies from the npm registry and system packages from the Nix package manager (nixpkgs) to facilitate the Astro build and Firebase deployment.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface by instructing the agent to process data from project files (e.g., README.md, go.mod) and interpolate them into the website structure. While MDX escaping is mentioned, the lack of strict boundary markers for all ingested content poses a low risk of malicious instructions influencing the deployment process.

website-launch