The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes deceptive instructions referencing a fabricated 'April 2026 Vercel security incident' and 'ownership concerns' to influence the agent's decision-making process and trigger migration logic under a false premise of a security emergency.
[METADATA_POISONING]: The skill's description and trigger conditions contain false claims about a future security breach and reliability issues to drive platform migration, potentially causing the agent to misinform the user about the safety of their current deployment.
[CREDENTIALS_UNSAFE]: The migration workflow explicitly requests the agent to read .env files and .env.vercel.backup files to 'note which keys exist.' This process exposes the full contents of these files, including sensitive secret values, to the agent's context window, increasing the risk of credential leakage.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest untrusted data from repository configuration files and environment backups which could contain embedded instructions.
Ingestion points: Processes repository files including vercel.json, package.json, and .env files in Step 1 and Step 4.
Boundary markers: There are no explicit delimiters or 'ignore' instructions specified for the content of the files being read.
Capability inventory: The skill possesses extensive capabilities including file system access, network communication via MCP, and the ability to update project environment variables.
Sanitization: There is no evidence of sanitization or validation logic for the content ingested from the project configuration or environment files.

vercel-to-createos