The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses Bash to execute diagnostic scripts, including checking file existence and permissions with stat, listing session directories with ls, and using node to parse JSON configuration data.
[DATA_EXFILTRATION]: The skill reads a sensitive configuration file (~/.promptbook/config.json) and extracts an api_key to send it to the vendor's official domain (promptbook.gg) via curl for validation. This behavior is standard for a diagnostic tool verifying service connectivity and credential validity.
[INDIRECT_PROMPT_INJECTION]: The skill processes content from a local configuration file to report status. While this is an ingestion point for potentially untrusted local data, the skill uses specific parsing logic and diagnostic reporting that limits the risk of the data influencing agent behavior.
Ingestion points: Reads metadata and configuration from $HOME/.promptbook/config.json and file listings from ~/.promptbook/sessions/.
Boundary markers: Not explicitly defined in the data ingestion, but results are formatted into a structured diagnostic report.
Capability inventory: The skill has access to Bash (curl, node, stat, ls) and Read tools.
Sanitization: Uses JSON.parse within Node.js to programmatically extract status fields (e.g., has_api_key, telemetry_consent) without exposing the full content to the primary prompt flow.

doctor