The Agent Skills Directory

[COMMAND_EXECUTION]: The skill employs dynamic context syntax (!command) to execute shell probes on the host system at the moment the skill is loaded. These probes perform file system searches and content inspection (find, ls, grep).
[DATA_EXFILTRATION]: One of the dynamic context commands specifically targets the .env file to check for the presence of the DATABASE_URL secret (grep -q '^DATABASE_URL=' .env). This represents sensitive file access within an automated execution context, which is a high-risk pattern for credential harvesting.
[COMMAND_EXECUTION]: Instructions in references/connect.md guide the agent to load all environment variables from .env files into the shell session (set -a; . ./.env; set +a) and execute various CLI tools like psql, npx wrangler, and pnpm exec drizzle-kit. This provides a broad surface for command execution and potential secret exposure if the environment is not strictly controlled.
[PROMPT_INJECTION]: The skill is designed to ingest data from live databases, which are considered untrusted external sources. While instructions include using LIMIT and targeted queries as mitigations, there is an inherent risk of indirect prompt injection if the database contains malicious payloads intended to influence the agent's logic during the schema comparison or data reasoning steps.

inspect-db