agentleFS
hiteshbandhuhiteshbandhu/skills-i-use/skills/ai-engineer-talks/run-llm-evals/SKILL.md

run-llm-evals

Skill1 starsChanged 5 months agoMITView on GitHub

Runs checklists and workflows for designing, operating, and improving LLM evals — domain harnesses, agent/voice evals, LLM judges, metrics workshops, product flywheels, dynamic search eval, enterprise conversation intelligence. Use when the user says "evals", "LLM judge", "eval harness", "regression on prompts", "agent eval", "eval dataset", or needs to ship with measurement before prod.

The file

We split this skill into the rules it gives an agent

1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use the decision tree below.

    line 25

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Run LLM evals

Action playbook from twenty-five AI Engineer eval/benchmark talks. Do not summarize talks — pick a workflow and execute it.

Supporting files (read when needed):

  • workflows.md — workflows A–N (steps, deliverables, stop conditions)
  • source-index.md — src-NNN → talk learnings in ingest-into-skills

Optional: {SKILL_OUTPUT_DIR}/run-llm-evals/


Step 0 — Pick workflow

Use the decision tree below. Open the matching section in workflows.md.

What is the user trying to do?
├─ Start or justify an eval program                    → A
├─ Build domain-specific eval from scratch             → B
├─ Design product/LLM app eval (not unit tests)        → C
├─ Define metrics/rubrics that match humans            → D
├─ Braintrust-style playground → CI → online eval      → E
├─ Ship agentic apps with trace-native eval            → F
├─ Voice / phone agent simulation eval                 → G
├─ Close observability ↔ eval gap (Azure agents)       → H
├─ Calibrate LLM judges / adversarial fuzzing          → I
├─ Benchmark models (perf + knowledge + honesty)       → J
├─ Enterprise CX: 100% conversations + labeling        → K
├─ Zapier-style PM+eng eval ops & post-ship feedback   → L
├─ Evaluate web/RAG search APIs (dynamic sets)         → M
└─ Subjective media eval or long-lived architecture    → N

Stop summarizing once a workflow is identified — run its checklist.


Install

Copy this folder into your agent’s skills directory:

cp -r skills/run-llm-evals ~/.claude/skills/
cp -r skills/run-llm-evals ~/.cursor/skills/
cp -r skills/run-llm-evals ~/.codex/skills/

From skills-i-use or ingest-into-skills (playlists/evals-benchmarks-ai-engineer/).


Cross-cutting rules

RuleSource
Evals ≠ unit tests; cover distributions and offense, not one-shot demos[src-005 @ 2:57]
Look at data before fancy judges[src-006 @ 7:15], [src-009 @ 8:25]
Align offline scorers with online logging[src-001 @ 3:23]
Layer evals on routers, tools, and final answers[src-007 @ 4:46]
Calibrate LLM judges; meta-eval the judge[src-013 @ 8:15]
Separate core-traffic vs aspirational datasets[src-020 @ 15:03]

Disputed steps: see source-index.md. Name workflow A–N; save artifacts to ./skill-outputs/run-llm-evals/ when requested; do not auto-commit.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.