The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell scripts and Python one-liners to construct API requests. It uses Python's json.dumps to process user-supplied parameters like names and search terms before sending them to the API.
[COMMAND_EXECUTION]: scripts/twenty-config.sh contains a hardcoded absolute file path (/Users/jhumanj/clawd/config/twenty.env) for configuration. This hardcoded path creates a dependency on a specific local user directory, which may lead to execution failures or unauthorized access if not properly configured on other systems.
[COMMAND_EXECUTION]: scripts/twenty-create-company.sh writes company payload data to a static temporary file path (/tmp/twenty_create_company.json). The use of static paths in /tmp without unique identifiers or restricted permissions is a security risk on multi-user systems as it can lead to data leakage or symlink-based race conditions.
[DATA_EXFILTRATION]: The skill transmits CRM data and authentication tokens to a user-defined remote endpoint (TWENTY_BASE_URL). While this is the intended purpose of the CRM integration, it constitutes a data egress point that should be monitored.
[COMMAND_EXECUTION]: Ingestion points for untrusted data include search terms and company names in twenty-find-companies.sh and twenty-create-company.sh. While Python's json.dumps provides some sanitization, the skill lacks explicit boundary markers or instructions to the agent to ignore embedded commands in the processed CRM data.

twenty-crm