The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes several shell commands to automate the development environment, including git worktree for isolation, docker compose for service orchestration, sed for modifying configuration files, and kill for process management.
[CREDENTIALS_UNSAFE]: The skill performs operations on sensitive environment files that often contain credentials. Evidence: cp ../../.env* . in apps.md copies environment files into worktree subdirectories, duplicating sensitive information like API keys and database passwords across the local filesystem.
[INDIRECT_PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection if used with untrusted codebases.
Ingestion points: Reads and acts upon files within the repository like .env and watch.sh.
Boundary markers: Absent; the agent does not distinguish repository data from system instructions.
Capability inventory: High, including file manipulation and arbitrary shell execution.
Sanitization: None; repository content is executed directly.
[DYNAMIC_EXECUTION]: The workflow requires executing local scripts and binaries such as make tailwindcss, ./tailwindcss, and ./watch.sh, which can result in code execution if the repository contains malicious versions of these files.
[EXTERNAL_DOWNLOADS]: The skill setup involves downloading external assets during execution. Evidence: make tailwindcss is described as a step that downloads the Tailwind CSS binary if it is missing from the environment.

worktrees