The Agent Skills Directory

[SAFE]: The core instructions in SKILL.md define a specialized persona for reviewing workshop agendas and session designs. The workflow is restricted to textual analysis and critique without involving sensitive tools or external data exfiltration.
[COMMAND_EXECUTION]: The skill includes Node.js utility scripts (evals/collect-l1-results.mjs and evals/generate-trigger-report.mjs) used by developers to aggregate and report on evaluation data.
These scripts utilize standard Node.js modules (node:fs, node:path, node:url) for local file operations.
evals/collect-l1-results.mjs accesses a specific local directory within the user's home path (.cursor/projects/...) to process agent transcripts, which is consistent with the intended development use case.
[PROMPT_INJECTION]: The pressure-scenarios.md file contains test prompts that use common injection prefixes such as "IMPORTANT: This is a real task". These are explicitly used as part of a testing suite to verify the agent's robustness and are not part of the runtime instructions that would manipulate the agent's behavior maliciously.
[DATA_EXPOSURE]: The skill processes user-provided facilitation plans. While it interacts with external content, its capabilities are limited to generating a prioritized critique. It does not possess network access or the ability to write to sensitive system files based on the content of the plans it reviews.

facilitation-review