The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes output from various external tools and APIs to determine the success or failure of deployment steps, creating a surface for indirect instructions.
Ingestion points: The skill reads deployment status pasted by users, container health check responses via curl/wget, and build logs from Vercel/Coolify (e.g., in recipes/nextjs-supabase-on-vercel.md and recipes/nextjs-supabase-on-coolify.md).
Boundary markers: The instructions do not specify explicit delimiters or "ignore instructions" wrappers for the data returned by these external processes.
Capability inventory: The skill possesses high-privilege capabilities including merging pull requests (gh pr merge), executing database migrations (supabase db push), and managing live deployments (vercel promote, docker run).
Sanitization: No explicit sanitization or filtering of tool outputs is mentioned; the agent is instructed to parse tool responses directly to verify "pass" conditions.
[COMMAND_EXECUTION]: The skill relies on a wide range of local CLI tools to automate the deployment lifecycle.
Evidence: The skill instructions explicitly invoke git, gh, docker, psql, supabase, vercel, curl, and wget across various files.
Operations performed include building Docker containers (docker build), modifying .gitignore and .env files, and executing shell commands to verify service health.
[CREDENTIALS_UNSAFE]: The skill manages access to sensitive environment variables and credentials required for deployment processes.
Evidence: The skill handles SUPABASE_ACCESS_TOKEN, SUPABASE_DB_PASSWORD, and VERCEL_OIDC_TOKEN through environment variables and local files like .env.local.
Safety Context: The instructions include robust negative constraints, explicitly forbidding the agent from printing or logging these secrets during the shipping process (e.g., "Never read, print or show the value" in SKILL.md).

ship