The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill explicitly asks the user for authentication secrets including email/usernames and passwords in Step 2. These credentials are then written into a temporary local file (screenshot-script.mjs) as plain-text strings to be used by the Playwright script, creating a significant risk of credential exposure if the file is intercepted before deletion.- [COMMAND_EXECUTION]: The skill dynamically constructs a Node.js script based on user input and project discovery. This script is then executed on the host system using the node command. This represents arbitrary command execution where the behavior depends on the content of the generated script.- [PROMPT_INJECTION]: The skill reads and processes external data from the project's README.md, CHANGELOG.md, and source code files to build a feature list. This creates a surface for indirect prompt injection where an attacker could place malicious instructions in project documentation to manipulate the agent's planning phase.
Ingestion points: Reads project documentation and routing files in Step 3.
Boundary markers: None; the skill reads file contents directly without delimiters.
Capability inventory: Filesystem access (mkdir, rm), shell execution (node), and user interaction.
Sanitization: None; external content from the codebase is interpolated into the reasoning process without escaping.

screenshots