The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill uses environment variables (DIALPAD_API_KEY, ATTIO_API_KEY, GOG_KEYRING_PASSWORD) for API authentication, which is the recommended practice for managing sensitive secrets in agent skills.
[DATA_EXFILTRATION]: Network activity is confined to legitimate integration points with vendor-specific APIs (Dialpad, Attio) and a local webhook server. No unauthorized data transmission or suspicious network patterns were identified.
[COMMAND_EXECUTION]: The skill utilizes local Python scripts for campaign logic and provides instructions for interacting with established CLI tools like gog-shapescale and attio. These operations are well-scoped to the skill's purpose.
[PROMPT_INJECTION]: The skill processes untrusted SMS content received via webhooks, which is stored in a local state file and displayed during status checks. This presents an indirect prompt injection surface typical of communication tools.
Ingestion points: webhook_handler.py (POST /) receives incoming message text from Dialpad webhooks.
Boundary markers: None present in the campaign state storage or CLI output displays.
Capability inventory: The skill can execute local management scripts and call external communication tools (Dialpad, Gmail) and CRM APIs.
Sanitization: No explicit sanitization or filtering of incoming message content was observed before it is stored in campaigns.json or presented to the agent.

campaign-orchestrator