The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests service and endpoint configuration data from external files such as services.json and endpoints.json to drive its monitoring logic. This configuration-driven approach creates a potential surface where malformed or malicious data in these files could attempt to influence the agent's shell command generation. However, the skill mitigates this by enforcing rigorous validation constraints, requiring narrow process patterns, and using evidence-based verification gates.
Ingestion points: The skill reads services.json, endpoints.json, and tech-source-registry.json to build manifests of services, endpoints, and CVE feeds.
Boundary markers: The instructions emphasize an "evidence-based" core principle, requiring the agent to independently verify process status, health file timestamps, and port bindings rather than trusting file content alone.
Capability inventory: The skill uses Bash, Read, Write, and Edit tools to perform checks and generate reports.
Sanitization: The instructions require narrow process patterns validated via ps aux | grep and absolute health file paths to prevent false matches or path traversal risks.
[COMMAND_EXECUTION]: The skill executes diagnostic commands such as pgrep, ss, and curl via the Bash tool to monitor service status. These operations are essential for the skill's primary purpose. Safety is maintained by explicitly instructing the agent to never perform remediation actions, such as systemctl restart, without an explicit user request and human review of the findings.
[DATA_EXFILTRATION]: The skill performs network probes for endpoint validation and CVE source audits. The documentation provides clear security guidance, such as warning against mutating production data, recommending read-only service account tokens, and mandating environment variable interpolation (${API_TOKEN}) to prevent the exposure of credentials in configuration files or logs.

service-health-check