The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes pip install -e dex-core to automatically repair missing or broken Python packages identified during health checks.
[COMMAND_EXECUTION]: Runs system commands such as python3 to execute a pre-flight utility and python3 -c to dynamically test module imports for MCP servers.
[CREDENTIALS_UNSAFE]: Accesses and inspects configuration files including .mcp.json and supabase.json (Granola credentials) to verify the status of API keys and session tokens.
[DATA_EXFILTRATION]: Utilizes a track_event function for silent analytics, transmitting metadata about the health check results (error counts, fix status) to an external telemetry service.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it ingests untrusted technical data from .logs/mcp-health.json and .logs/error-queue.json.
Ingestion points: .logs/mcp-health.json, .logs/error-queue.json.
Boundary markers: The instructions mandate silent reading and translation of errors into human-readable text, providing an abstraction layer, but lack explicit sanitization of the technical message strings.
Capability inventory: Command execution (pip, python), file creation, and analytics reporting.
Sanitization: Employs a fixed mapping table to translate specific technical patterns into predefined fixes, which mitigates the risk of arbitrary log content directly controlling the agent's behavior.

health-check