The Agent Skills Directory

[SAFE]: The skill implements an evaluation board for comparing agent versions using mock data. Analysis of all code and instructions confirms that the skill performs its stated function without secondary malicious intent.\n- [COMMAND_EXECUTION]: The skill provides management scripts (generate_eval_run.mjs, export_release_report.mjs, setup.mjs, publish_airapp.mjs) intended for project setup and data export. These scripts interact with the Busabase API and the local filesystem in a transparent and appropriate manner for their utility.\n- [INDIRECT_PROMPT_INJECTION]: While the skill processes transcripts that may contain adversarial examples (e.g., jailbreak attempts or phishing examples in the mock test suite), it treats this content strictly as data to be displayed in the UI. The web interface correctly uses HTML escaping (escapeHtml) to sanitize this content before rendering, preventing cross-site scripting (XSS) or confusion. The agent is not instructed to follow instructions contained within the test data.\n- [CREDENTIALS_SAFE]: The management scripts utilize environment variables (BUSABASE_BASE_URL, BUSABASE_API_KEY) for authentication, which is a standard and recommended practice for secure credential management. No secrets are hardcoded in the skill's source files.

kelly-agent-eval