The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The reference/dev-servers-status.sh script contains a _read_api_key function that actively searches for and extracts the value of APP_API_KEY from .env and .env.example files. This value is then printed to the standard output, exposing sensitive credentials in the agent's session context.
[DATA_EXFILTRATION]: The skill performs broad read operations on environment files (.env, .env.example, .wt-ports.env) to extract port configurations and secrets. The programmatic extraction of these values, particularly the API key search, poses an exfiltration risk if the output is processed by other tools or agents.
[COMMAND_EXECUTION]: The skill dynamically generates and writes several Bash scripts to the filesystem. These scripts are constructed by interpolating data from package.json and .env files. If these files contain malicious payloads, they could lead to arbitrary command execution when the generated scripts are run.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted data from project configuration files (package.json, .env) and uses them to determine its behavior and generate executable code without boundary markers or sanitization.
[EXTERNAL_DOWNLOADS]: The skill fetches documentation from well-known external sources, including vitejs.dev and GitHub (coderabbitai/git-worktree-runner), to assist with port configuration detection. These are documented neutrally as they target well-known services.
[REMOTE_CODE_EXECUTION]: The dev-post-checkout.sh script automatically executes package manager install commands (e.g., pnpm install, npm install, cargo build) based on detected project files, which effectively executes code retrieved from external package registries.

dev-setup