The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/refresh_token.js utilizes child_process.exec to launch the system's default web browser during the OAuth 2.0 authorization flow. This is used specifically to open platform-appropriate binaries like open, start, or xdg-open with an escaped authorization URL.
[EXTERNAL_DOWNLOADS]: The skill communicates with official Google service endpoints (tasks.googleapis.com and oauth2.googleapis.com) for task management and token refreshment. These are well-known, trusted technology services.
[PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection because it processes task titles and notes that could originate from external sources.
Ingestion points: Data enters the agent context through scripts/get_tasks.sh (retrieving tasks from the API) and scripts/create_task.sh (taking user-provided strings).
Boundary markers: The skill does not use explicit delimiters to isolate retrieved task content from the agent's instructions.
Capability inventory: The skill can execute subprocesses (curl, jq) and perform file-write operations to maintain the token.json file.
Sanitization: The bash scripts utilize jq with the --arg flag for JSON construction, ensuring that task content is properly escaped and cannot break the structure of API requests.

google-tasks