The Agent Skills Directory

[SAFE]: The skill provides instructional guidelines for analytical tasks in the Monitoring, Evaluation, and Learning (MEL) domain. Analysis of the instructions and portable configuration blocks found no evidence of malicious intent, credential harvesting, or unauthorized network operations.
[INDIRECT_PROMPT_INJECTION]: The skill describes a workflow that ingests data from external sources, creating a potential surface for indirect prompt injection. (1) Ingestion points: External research materials, the 'MEL Wiki', and results from web search tools. (2) Boundary markers: The instructions mandate clear citation formatting and an 'Evidence base line' for all referenced sources, which helps delineate external data from agent reasoning. (3) Capability inventory: The workflow utilizes web search for fact-finding and Grep/search tools for formatting verification. (4) Sanitization: The 'Five Gates' methodology requires the model to 'adversarially reason' and physically verify each citation against its original source, providing a procedural defense against untrusted data.

mel-discipline