The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The file openapi/twenty-metadata.json contains two hardcoded JWT tokens within the description field. These tokens are formatted as active API keys with expiration dates set far into the future (year 2126).
[EXTERNAL_DOWNLOADS]: The script scripts/fetch_openapi_specs.sh is designed to download OpenAPI JSON files from remote Twenty CRM instances provided via the --base-url argument.
[COMMAND_EXECUTION]: The skill documentation and scripts authorize the execution of shell commands (bash, curl) and Python scripts to facilitate CRM operations and data processing.
[PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection by ingesting and relying on external OpenAPI specifications. Malicious content within a compromised CRM's metadata could potentially influence agent behavior.
Ingestion points: scripts/fetch_openapi_specs.sh downloads external JSON data into the local openapi/ directory.
Boundary markers: Absent. Instructions suggest treating the live instance as the "source of truth" without explicit validation of text-based metadata.
Capability inventory: The skill utilizes subprocess execution (curl, bash), file-system writes, and network operations.
Sanitization: Validation is limited to ensuring the downloaded file is a valid JSON OpenAPI document; it does not sanitize or filter the content of the fields within the document.

twenty-crm