The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill analyzes repository content to generate setup scripts, creating a surface for indirect prompt injection from untrusted files in the repository.
Ingestion points: The agent is instructed to inspect the repository (e.g., lockfiles, .env.example, docker-compose.yaml) to determine setup requirements as described in SKILL.md.
Boundary markers: The skill includes a mandatory instruction to "Show the user the proposed files and get explicit approval before writing," providing a manual verification gate.
Capability inventory: The agent has access to the Bash tool (specifically superset:* tools) to write files and execute workspace creation commands which run the generated scripts.
Sanitization: No specific automated sanitization is mentioned; the process relies on the user review step to filter malicious content interpolated into the scripts.
[COMMAND_EXECUTION]: The skill's primary purpose is to generate and run shell scripts (.superset/setup.sh, .superset/teardown.sh) and commands (e.g., bun dev) through the superset CLI. These commands are executed locally within the user's workspace environment to automate development setup.
[DATA_EXFILTRATION]: The instructions suggest copying environment variables and secrets (such as .env files) from the main repository to the local workspace worktree. This process is localized to the user's development environment and is intended for configuration rather than external transmission.

setup