The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of bash -c to wrap curl commands and jq for processing JSON responses. This is primarily used to manage Google Docs resources and handle environment variables during piped operations.
[CREDENTIALS_UNSAFE]: Uses the GOOGLE_DOCS_TOKEN environment variable for authentication. The skill metadata explicitly defines this as a platform-managed secret (vm0_secrets), which is a standard practice for secure credential handling in this context.
[DATA_EXFILTRATION]: Communicates with official Google API endpoints (docs.googleapis.com and googleapis.com). These are well-known services, and the network activity is strictly aligned with the skill's documented purpose of managing Google Docs.
[INDIRECT_PROMPT_INJECTION]: The skill has the surface area for indirect injection as it reads content from external Google Docs which could contain instructions. However, it uses jq to parse data and presents it as structured output, following standard data-retrieval patterns.
Ingestion points: Document content is fetched in steps 2, 3, 16, and 17 using curl from the Google Docs API.
Boundary markers: None explicitly present in the suggested commands to separate document data from agent instructions.
Capability inventory: The agent has shell access via curl, jq, and tr as demonstrated in the skill's examples.
Sanitization: Uses jq for parsing and tr -d '\0' for basic cleanup in plain text extraction, but does not perform semantic validation of document content.

google-docs