The Agent Skills Directory

[CREDENTIALS_UNSAFE] (HIGH): The skill's configuration helper (scripts/config-helper.js) writes sensitive credentials, including the secret_key and connection_id, to a plain text file at ~/.nango-mcp/credentials.json. This predictable path and lack of encryption allow any local process to harvest these credentials if the environment is compromised.
[PROMPT_INJECTION] (LOW): The SKILL.md file contains instructions tagged 'CRITICAL
CONTINUOUS EXECUTION' that explicitly command the agent to skip user confirmation and status reporting during multi-step tool execution. This instruction is designed to bypass human-in-the-loop oversight for automated actions.
[INDIRECT_PROMPT_INJECTION] (LOW): The skill processes external data from HubSpot, creating an attack surface where malicious task content could influence agent behavior.
Ingestion points: Data returned by nango-mcp-server_whoami and various HubSpot task fetching tools referenced in SKILL.md.
Boundary markers: Absent; no delimiters are used to encapsulate external content in the system prompt instructions.
Capability inventory: File system access via config-helper.js, network access to Nango's API, and task creation/deletion capabilities.
Sanitization: No input validation or output escaping is performed on data retrieved from HubSpot before it is processed by the agent.

hubspot-task-manager