model-grader plugin
Asks the 26 questions needed to build a model grader and nothing else: what the grader itself is, output shape, the eight integrity questions whose absence lets a model fake success, and the verdict rules. Tracks four gates per rule, exits on a buildability dry-run, and blocks freeze without a recorded trail.
How to install the model-grader plugin
Run both commands inside Claude Code. The first adds the marketplace once; the second installs the plugin.
/plugin marketplace add Abhillashjadhav/AI-PM-essential-skills
/plugin install model-grader@ai-pm-skillsWhat's inside model-grader
1 part read from the plugin's own files.
Skills 1
Claude loads it when the task matches- model-grader
Use this skill when someone needs to define what correct output means for an AI feature so that a grader can be built from the answers. Triggers on "build a grader", "define good", "what should the eval check", "how do we know the agent got it right", "write the acceptance criteria", "is this spec complete enough to test", "the agent keeps doing X and nobody agreed whether that is wrong", or any handoff where a model will be judged against a description a human wrote. Runs a twenty-six question interview in three parts — output shape, integrity, verdict and judgment — then a buildability test, and emits a contract an implementer can build from plus an explicit list of undecided items. Do NOT use to write grader code, run an eval suite, score model output, or repair an existing grader.
Using model-grader in Claude Code
Skills need no command. Claude reads them when your request matches what they describe.
Update, turn off or remove
- Update the plugin (shell)
- claude plugin update model-grader@ai-pm-skills
- Refresh the marketplace
- /plugin marketplace update ai-pm-skills
- Turn it off for now
- /plugin disable model-grader@ai-pm-skills
- Remove it
- /plugin uninstall model-grader@ai-pm-skills
Skills in model-grader
- model-graderAbhillashjadhav/AI-PM-essential-skills
Use this skill when someone needs to define what correct output means for an AI feature so that a grader can be built from the answers. Triggers on "build a grader", "define good", "what should the eval check", "how do we know the agent got it right", "write the acceptance criteria", "is this spec complete enough to test", "the agent keeps doing X and nobody agreed whether that is wrong", or any handoff where a model will be judged against a description a human wrote. Runs a twenty-six question interview in three parts — output shape, integrity, verdict and judgment — then a buildability test, and emits a contract an implementer can build from plus an explicit list of undecided items. Do NOT use to write grader code, run an eval suite, score model output, or repair an existing grader.
8 files9stars
model-grader plugin questions
How do I install the model-grader plugin?
In Claude Code run /plugin marketplace add Abhillashjadhav/AI-PM-essential-skills, then /plugin install model-grader@ai-pm-skills.
What does the model-grader plugin include?
model-grader bundles 1 skill.
How do I update or remove the model-grader plugin?
Update it with claude plugin update model-grader@ai-pm-skills in your shell, or from the Installed tab of /plugin. Remove it with /plugin uninstall model-grader@ai-pm-skills.