The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes standard Python scripts for fetching (gmail_label_fetch.py), splitting (gmail_label_split.py), merging (gmail_label_merge.py), and applying (gmail_label_apply.py) labels to Gmail data. These are local scripts within the skill's own package.
[DATA_EXPOSURE]: The skill accesses Gmail messages but limits data collection to metadata (Subject, From, Date) and a truncated snippet (first 120 characters) to optimize for LLM classification cost and privacy. OAuth tokens are managed via a local gmail_accounts.json registry file, which is standard practice for multi-account management.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (emails). However, it fetches only specific metadata and snippets, reducing the surface for injection. The instructions also define clear classification categories (Action Required, Waiting On, Reference), which helps steer the agent towards structured output and reduces the likelihood of obeying instructions hidden in email content.
[DYNAMIC_EXECUTION]: The skill uses parallel subagents (Tasks) to classify email chunks. While this involves spawning new agent contexts, the scope is limited to reading a specific JSON chunk and writing a classified JSON output file.

gmail-label