The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's operational logic is built around the execution of local Node.js scripts including check_services.mjs, sync_domains.mjs, ingest_spend.mjs, and execute_decisions.mjs. These scripts are intended to be run by the agent to update the fleet status and plan maintenance tasks. They interact with the environment via standard variables and local JSON roster files.
[EXTERNAL_DOWNLOADS]: The monitoring scripts perform intentional network requests to probe external services. check_services.mjs uses fetch and tls.connect to verify the status and certificate validity of user-configured endpoints. sync_domains.mjs queries the public rdap.org service for domain registration data. These connections are legitimate, well-documented components of the skill's monitoring capabilities.
[INDIRECT_PROMPT_INJECTION]: The skill handles data from external web servers and RDAP APIs, which could theoretically contain malicious instructions.
Ingestion points: Data fetched from service endpoints and RDAP responses.
Boundary markers: The system clearly separates check results from human-actionable cards in the Action Queue.
Capability inventory: Node script execution, filesystem access for rosters/payloads, and outbound network probes.
Sanitization: All external data rendered in the AirApp UI is sanitized using escapeHtml functions in app.js and operations-views.js to prevent injection or cross-site scripting attacks.

kelly-devops