The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data which presents a potential injection surface. Ingestion points: The skill reads email content and metadata via the Gmail connector to identify and extract receipt information (SKILL.md). Boundary markers: Explicit safety rules instruct the agent to treat all email bodies and attachments as hostile and to disregard any instructions found within them (SKILL.md, Safety Rule 3). Capability inventory: The agent has the capability to write records to an Airtable base and generate or save files to Google Drive (SKILL.md). Sanitization: The skill mandates human-in-the-loop confirmation before any write operation and includes specific logic to identify and flag suspicious phishing domains and lookalike senders (SKILL.md, Safety Rules 2 and 5; references/receipt-extraction.md).
[NO_CODE]: No executable scripts or binary files are shipped with this skill; it consists entirely of markdown instructions and reference documentation.

expense-wrangler