The Agent Skills Directory

[SAFE]: The skill is designed with a strong human-in-the-loop requirement, ensuring that no data is written to the platform without the user's explicit review and approval of the final draft.
[EXTERNAL_DOWNLOADS]: The skill directs the host to configure the Well MCP server using the vendor's official endpoint at https://api.wellapp.ai/v1/mcp. This is a legitimate vendor resource used for establishing the necessary communication channel.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external sources and user input which could serve as injection surfaces.
Ingestion points: The skill retrieves company information via well_query_records and accepts conversational invoice descriptions from the user as described in SKILL.md.
Boundary markers: While explicit delimiters are not mentioned for the data objects, the instructions mandate that the agent presents the full draft to the user for confirmation.
Capability inventory: The skill possesses write capabilities through the well_create_invoice_from_data tool.
Sanitization: The primary mitigation is a mandatory human-in-the-loop confirmation process where the user must explicitly verify all fields (issuer, receiver, amounts, and dates) before the write operation is executed.

draft-invoice