The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires the mcp-skill and fastmcp Python packages to handle MCP communication and authentication.
[DATA_EXFILTRATION]: The skill establishes network connections to the official Airtable service at https://mcp.airtable.com/mcp for its primary operations.
[DATA_EXFILTRATION]: The application accesses the local file system path ~/.mcp-skill/auth/ to store OAuth tokens, which is necessary for maintaining authenticated sessions.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it ingests data from external Airtable records which could contain malicious instructions intended to manipulate the agent.
Ingestion points: Untrusted data enters the agent context through the list_bases, list_tables_for_base, get_table_schema, and list_records_for_table methods in app.py.
Boundary markers: The skill does not implement boundary markers or instructions to ignore embedded content when presenting retrieved data to the model.
Capability inventory: The skill possesses both read and write capabilities (create_records_for_table, update_records_for_table), creating a risk if an injected instruction triggers an automated write operation.
Sanitization: No sanitization or validation is performed on the data retrieved from the Airtable API before it is processed by the agent.

airtable