The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute vercel env pull .env.vercel-prod --environment=production. This involves using a CLI tool to fetch production-level secrets and environment variables, which is a high-privilege administrative action for an AI agent to perform.
[CREDENTIALS_UNSAFE]: The instructions require the agent to manually extract, export, and use the production DATABASE_URL. Handling production database credentials directly within the agent's environment and session history increases the risk of credential exposure.
[INDIRECT_PROMPT_INJECTION]: The skill processes content from local files (seed-skills/<slug>/SKILL.md) and uses it to update a production database.
Ingestion points: Local markdown files in the seed-skills/ directory are read and processed.
Boundary markers: No specific delimiters or safety instructions are provided to the agent or the underlying scripts to prevent the interpretation of instructions that might be embedded in the markdown files being seeded.
Capability inventory: The skill possesses capabilities to build software, execute Node.js scripts, perform database seeding, and push changes to a remote repository via Git.
Sanitization: Although a validator CLI is mentioned, its described role is for structural and format verification rather than security sanitization of the markdown content.
[DATA_EXFILTRATION]: The instructions contain hardcoded absolute paths to the user's home directory (e.g., /Users/promode/qaskills). This reveals information about the host system's user account and directory layout.
[EXTERNAL_DOWNLOADS]: The skill performs network requests to qaskills.sh using curl to verify skill existence and database counts. These operations target the infrastructure associated with the skill's purpose to validate the success of the seeding process.

add-seed-skills