Skip to content
Agentmart
Submit a tool

asc-ml-eval

Design and run ML evaluations, slices, and release gates.

Skill by cavaunpeu

IndexedMIT
Write a review

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/cavaunpeu-asc-ml-eval

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Design and run ML evaluations, slices, and release gates. Use when you need to define what good looks like, build an offline eval, compare model variants, or decide whether an ML change is ready to ship.

Use cases: design, ml, agent-building

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/asc-ml-eval

What's inside

Skill name
asc-ml-eval
Description
Design and run ML evaluations, slices, and release gates. Use when you need to define what good looks like, build an offline eval, compare model variants, or decide whether an ML change is ready to ship.
License
MIT
Instructions
73 lines in SKILL.md · view on GitHub

Files (2)

  • SKILL.md
  • openai.yaml · agents/

SKILL.md (first lines)

# Asc Ml Eval

Evaluation is the control surface.
If you cannot say how the change is measured, the work is still too fuzzy.

## Workflow

### 1. Define the decision

State what the eval is trying to decide.

- Is the model good enough.
- Is one variant better than another.
- Did a change regress a slice.
- Is the system ready to ship.

### 2. Pick the right unit of measurement

- Use per-example checks when the task is local.
- Use grouped slices when failures cluster by domain, language, risk, or customer segment.
- Use end-to-end checks when the user experience matters more than turn-level accuracy.

### 3. Write the rubric

Keep the rubric aligned to the real task.

- correctness
- policy adherence
- required conventions
- helpfulness or preference, when relevant
- slice-specific failure checks

Prefer deterministic checks where the rule is lexical or structural.
Use a judge when the judgment is genuinely semantic.

### 4. Build the eval set

- Hold out examples that match the target workload.
- Include hard cases and common cases.
- Keep provenance clear.
- Separate training data from eval data cleanly.

### 5. Set the release gate

Define the threshold before you look at results.

- What must pass.
- What can degrade a little.
- What is an automatic stop.
- What needs human review.

### 6. Track regression over time

- Compare against the last shipped version, not just a single baseline.
- Watch slices, not only the average.
- Keep the eval reproducible so the same change can be checked again later.

## Output Shape

When you use this skill, answer with:

- the decision the eval is supporting
- the rubric
- the eval set shape
- the gate
- the failure slices that matter most

## Design Rule

If the task needs a new metric, ask whether a simpler slice or rubric already answers the question.
Do not invent a new score just because the old one is inconvenient.

From cavaunpeu/agent-skills-closet, MIT.

More from cavaunpeu/agent-skills-closet

Signals

Stars
0
Forks
0
Open issues
0
Last commit (this folder)
Jun 20, 2026 (4 months ago)

From GitHub (fetched Oct 24, 2026).

Details

License
MIT
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from cavaunpeu

Cite this page

asc-ml-eval skill. Agent Mart. https://agentmart.dev/skills/cavaunpeu-asc-ml-eval (last updated 2026-10-24)