Skip to content
Kitbay
Submit

operational-value-designer

MITUpdated on Nov 11, 20265,616GitHub starsIncludes scripts

A gh-aw skill for defining how a GitHub Agentic Workflow's recorded sessions are graded on outcomes rather than activity.

What the skill teaches

A gh-aw skill for defining how a GitHub Agentic Workflow's recorded sessions are graded on outcomes rather than activity. Given a repository and workflow name, the agent chooses among built-in graders, a JavaScript grader that runs in an isolated job, or a separate Bash evaluator, and writes the configuration into the workflow file. The skill sets rules for where grading code may run, and bundled shell scripts check an evaluator against fixtures and verify changes to the grading contract.

Capabilities

  • Pick built-in graders for an agentic workflow
  • Write a JavaScript grader that runs in an isolated job
  • Write and test a Bash evaluator against fixtures
  • Add grading configuration to a workflow file