The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The core logic file contains hardcoded sensitive authentication credentials.\n
Evidence: scripts/twist_api.js defines DEFAULT_CLIENT_SECRET with a plaintext value. Exposure of these credentials could allow unauthorized parties to impersonate the integration or intercept user data.\n- [COMMAND_EXECUTION]: The skill executes shell commands to facilitate the OAuth login flow on the user's machine.\n
Evidence: In scripts/twist_api.js, the exec function is used to call open, start, or xdg-open with a dynamically constructed URL to initiate the browser-based authentication process.\n- [PROMPT_INJECTION]: The skill processes external data from Twist, creating a vulnerability surface for indirect prompt injection.\n
Ingestion points: Data is retrieved from the Twist API via commands like threads, comments, messages, and search in scripts/twist_api.js.\n
Boundary markers: No explicit boundary markers or structural delimiters (such as XML tags or guardrail instructions) are used to wrap external content before presentation to the agent.\n
Capability inventory: The skill has extensive write capabilities, including posting messages (reply), creating channels (add_channel), modifying users (add_workspace_user), and uploading files (upload_attachment).\n
Sanitization: The skill includes a sanitizeForAI function in scripts/twist_api.js that redacts specific injection-related keywords (e.g., 'ignore all previous instructions'), which provides some protection but relies on a limited blacklist.

twist-toolkit