Skill Evaluation Execution

Use this reference to run an evaluation of an existing Skill. Load Skill Quality Review for the Gates, D1-D8 scorecard, target profiles, and E0-E3 evidence ceilings. Use Skill Discovery instead when the question is whether to create or extend a Skill.

This protocol joins two complementary sources:

• Better Harness agent-lint supplies repository-aware, explicit-target E0 evidence for frontmatter, descriptions, routed references, size, and progressive disclosure.
• Plugin Eval supplies a deterministic local evaluation-result document for structure, estimated context budgets, helper-code checks, and benchmark preparation.

Neither source proves selection, task lift, repeatability, safety in use, or a later improved outcome by itself.

Resolve the Evaluation Envelope

Record these fields before running commands:

• canonical absolute Skill directory and SKILL.md entrypoint;
• target host/client, users, distribution profile, and supported platforms;
• requested mode: static analysis, benchmark preparation, or measured run;
• model, tools, permissions, workspace fixture, and allowed side effects;
• Better Harness commit and Plugin Eval version or resolved script entrypoint.

Use plugin-eval when it is on PATH. When the plugin is installed but the command is not linked, invoke its shipped Node entrypoint as