The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses standard CLI commands (claude gateway, claude plugins, claude config) to manage the agent's environment. These commands are used as intended for service installation and configuration.
[EXTERNAL_DOWNLOADS]: Installs the manifest plugin via the platform's native plugin manager. This is a standard extension mechanism.
[DATA_EXPOSURE]: Requests a Manifest API key from the user and stores it in the local configuration. It also reads from a local log file (~/.claude/logs/gateway.log) to verify that the observability pipeline is active. Both actions are transparently documented and necessary for the skill's stated purpose.
[INDIRECT_PROMPT_INJECTION]: The skill reads from a local log file. While logs can technically contain untrusted data, the use of grep to find a specific status string is a low-risk verification step. Evidence includes:
Ingestion points: Reading ~/.claude/logs/gateway.log via shell commands.
Boundary markers: None implemented for the log read.
Capability inventory: Subprocess execution for CLI tools (claude), configuration management, and log inspection (grep, tail).
Sanitization: None performed on the log content, though usage is restricted to a specific search pattern.

manifest