The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the execution of tools against external service providers (e.g., Slack, Gmail) using data processed by the agent host.
Ingestion points: The tool_exec.py script takes external input via the --tool-input argument, which is then used in tool execution.
Capability inventory: The skill can execute arbitrary tools within the Scalekit ecosystem and perform proxied HTTP requests via the actions.request method.
Boundary markers: The skill instructs the user to wait for exact dashboard values and connection names rather than inventing them.
Sanitization: Input is parsed as JSON but passed directly to the underlying SDK; the skill relies on the platform's tool-execution guardrails.
[COMMAND_EXECUTION]: The skill documentation guides the user to install additional host-level capabilities using specialized CLI tools.
Evidence: Commands such as clawhub install scalekit-agent-auth (in SKILL.md) and hermes skills install scalekit-inc/authstack/... (in references/hermes.md) are used to deploy vendor-maintained host skills.
[DATA_EXPOSURE]: The included tool_exec.py script contains a utility function to display active credentials for setup verification.
Evidence: The --get-authorization command calls the get_authorization function, which prints OAuth access_token and refresh_token to standard output. This is a legitimate administrative feature for verifying integration status.

integrate-agentkit-host