Practical notes: Evals: A Crash Course for Agents and Skills

The following notes reconstruct a practical path around “Evals: A Crash Course for Agents and Skills”. Emphasis stays on contracts, checks, and drop-in code placeholders rather than motivational framing. When working through the Overview stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Keep configuration outside application code. Environment files, secret stores, and feature flags belong in one place operators can audit without reading the whole graph.

The mental model

The The mental model stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish. Budget tokens per turn and per session. Agentic tools expand context aggressively; hard caps keep demos from becoming surprise invoices.

The evaluation layers

The The evaluation layers stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Prefer small, testable units over sprawling scripts. When a step fails, the failure should point at a single responsibility rather than a tangled pipeline. Keep graph state flat and typed. Nested blobs hide which node wrote which field and break resume after interrupts.

Skills need two eval suites

The Skills need two eval stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Treat this stage as a contract between inputs and validated outputs. Name the artifacts, define success checks, and refuse silent partial completion. Keep graph state flat and typed. Nested blobs hide which node wrote which field and break resume after interrupts. The Skills need two eval stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Keep configuration outside application code. Environment files, secret stores, and feature flags belong in one place operators can audit without reading the whole graph.

- prompt: "Patch the vulnerable npm dependencies"
  expected_skill: cve-remediation
- prompt: "Review authentication input validation"
  forbidden_skill: cve-remediation

What a good eval case looks like

For the What a good eval stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish. Put human approval on edges that spend money or change production data. Compile-time wiring does not equal business completeness.

{
  "id": "fix-null-condition",
  "prompt": "Fix saving rules with a null condition",
  "fixture": "repos/null-condition",
  "setup": ["npm install"],
  "checks": [
    "npm test -- null-condition.test.js",
    "git diff --check"
  ],
  "rubric": [
    "Fixes the root cause",
    "Preserves existing behavior",
    "Adds a regression test",
    "Avoids unrelated changes"
  ],
  "forbidden": [
    "deleting existing tests",
    "hard-coded fixture-specific output"
  ]
}

Graders: use the strongest one available

For the Graders use the strongest stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Prefer small, testable units over sprawling scripts. When a step fails, the failure should point at a single responsibility rather than a tangled pipeline. Put human approval on edges that spend money or change production data. Compile-time wiring does not equal business completeness.

Metrics that matter

For the Metrics that matter stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Treat this stage as a contract between inputs and validated outputs. Name the artifacts, define success checks, and refuse silent partial completion. Put human approval on edges that spend money or change production data. Compile-time wiring does not equal business completeness. For the Metrics that matter stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Keep configuration outside application code. Environment files, secret stores, and feature flags belong in one place operators can audit without reading the whole graph.

Building a good dataset

When working through the Building a good dataset stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish. Checkpoint after expensive steps. Resume should not re-bill the same LLM call when an operator retries a later node.

A practical loop

When working through the A practical loop stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Prefer small, testable units over sprawling scripts. When a step fails, the failure should point at a single responsibility rather than a tangled pipeline. Checkpoint after expensive steps. Resume should not re-bill the same LLM call when an operator retries a later node.

Common mistakes

When working through the Common mistakes stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Treat this stage as a contract between inputs and validated outputs. Name the artifacts, define success checks, and refuse silent partial completion. Checkpoint after expensive steps. Resume should not re-bill the same LLM call when an operator retries a later node. When working through the Common mistakes stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Keep configuration outside application code. Environment files, secret stores, and feature flags belong in one place operators can audit without reading the whole graph.

Routing metrics, made concrete

The Routing metrics made concrete stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish. Keep graph state flat and typed. Nested blobs hide which node wrote which field and break resume after interrupts.

The best place to start

The The best place to stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Prefer small, testable units over sprawling scripts. When a step fails, the failure should point at a single responsibility rather than a tangled pipeline. Keep graph state flat and typed. Nested blobs hide which node wrote which field and break resume after interrupts.

A real, minimal eval framework you can copy

The A real minimal eval stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Treat this stage as a contract between inputs and validated outputs. Name the artifacts, define success checks, and refuse silent partial completion. Keep graph state flat and typed. Nested blobs hide which node wrote which field and break resume after interrupts. The A real minimal eval stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Keep configuration outside application code. Environment files, secret stores, and feature flags belong in one place operators can audit without reading the whole graph.

evals/
  cases/
    skills/security-review.trigger.json
    agents/fix-null-condition.json
  graders.py
  metrics.py
  runner.py
  run.py        # CLI entry1. Case files. A skill trigger case is just positives and negatives. An agent case is a prompt plus graders and a run count.

// cases/skills/security-review.trigger.json
{
  "skill": "security-review",
  "positives": [
    "Audit this endpoint for SQL injection",
    "Check the login flow for auth bypasses",
    "Is this file-upload handler safe?"
  ],
  "negatives": [
    "Upgrade React dependencies",
    "Rename this variable across the repo",
    "Add a loading spinner to the form"
  ]
}

// cases/agents/fix-null-condition.json
{
  "id": "fix-null-condition",
  "prompt": "Fix saving rules with a null condition",
  "fixture": "fixtures/null-condition",
  "setup": ["npm ci"],
  "runs": 5,
  "graders": [
    { "type": "shell", "cmd": "npm test -- null-condition", "critical": true },
    { "type": "shell", "cmd": "git diff --check" },
    { "type": "forbidden", "pattern": "it\\.skip|xit\\(", "message": "must not disable tests" }
  ]
}

# graders.py
import re, subprocess
def shell(step, ctx):
    r = subprocess.run(step["cmd"], cwd=ctx.workdir, shell=True,
                       capture_output=True, text=True)
    ok = r.returncode == 0
    return {"pass": ok, "score": 1 if ok else 0, "detail": r.stdout[-400:]}
def forbidden(step, ctx):
    bad = re.search(step["pattern"], ctx.diff) is not None
    return {"pass": not bad, "score": 0 if bad else 1,
            "detail": step["message"] if bad else ""}
def judge(step, ctx):
    v = ctx.llm.rate(step["rubric"], ctx.artifact)  # 0..1, blinded
    return {"pass": v >= step.get("threshold", 0.7), "score": v}
GRADERS = {"shell": shell, "forbidden": forbidden, "judge": judge}3. Metrics. Success rate, pass@k, pass^k, and the routing confusion matrix — exactly the numbers from earlier.

# metrics.py
def success_rate(rs):
    return sum(1 for r in rs if r["pass"]) / len(rs)
def pass_at_k(rs):
    return 1 if any(r["pass"] for r in rs) else 0
def pass_hat_k(rs):
    return 1 if all(r["pass"] for r in rs) else 0
def routing(positives, negatives, fired):
    tp = sum(1 for p in positives if fired(p))
    fp = sum(1 for n in negatives if fired(n))
    fn = len(positives) - tp
    tn = len(negatives) - fp
    return {"tp": tp, "fp": fp, "fn": fn, "tn": tn,
            "precision": tp / (tp + fp or 1),
            "recall": tp / (tp + fn or 1)}4. The runner. One function per suite. The agent runner repeats each case so pass^k is meaningful; the skill runner just asks the router which skills fire.

# runner.py
import json
from graders import GRADERS
from metrics import success_rate, pass_at_k, pass_hat_k, routing
def run_agent_case(agent, path):
    c = json.load(open(path))
    runs = []
    for _ in range(c.get("runs", 3)):
        ctx = agent.run(c["prompt"], fixture=c["fixture"], setup=c["setup"])
        ok = True
        for step in c["graders"]:
            g = GRADERS[step["type"]](step, ctx)
            if step.get("critical") and not g["pass"]:
                ok = False
        runs.append({"pass": ok})
    return {"id": c["id"], "success": success_rate(runs),
            "pass_at_k": pass_at_k(runs), "pass_hat_k": pass_hat_k(runs)}
def run_skill_trigger(router, path):
    c = json.load(open(path))
    fired = lambda prompt: c["skill"] in router.route(prompt)
    return {"skill": c["skill"], **routing(c["positives"], c["negatives"], fired)}5. Run it. The CLI just dispatches on the case type and prints the metrics.

$ python run.py cases/agents/fix-null-condition.json
fix-null-condition   success=0.80   pass@5=1.00   pass^5=0.20
$ python run.py cases/skills/security-review.trigger.json
security-review      precision=0.86  recall=1.00   (tp=6 fp=1 fn=0 tn=5)

Don’t build from scratch if you don’t have to

For the Don t build from stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish. Put human approval on edges that spend money or change production data. Compile-time wiring does not equal business completeness.

General LLM & prompt evals

For the General LLM prompt evals stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Prefer small, testable units over sprawling scripts. When a step fails, the failure should point at a single responsibility rather than a tangled pipeline. Prefer structured outputs with schema validation over free-form prose when the next step is code or a tool call.

Tracing, datasets & LLM-as-judge platforms

For the Tracing datasets LLM-as-judge platforms stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Treat this stage as a contract between inputs and validated outputs. Name the artifacts, define success checks, and refuse silent partial completion. Prefer structured outputs with schema validation over free-form prose when the next step is code or a tool call. For the Tracing datasets LLM-as-judge platforms stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Keep configuration outside application code. Environment files, secret stores, and feature flags belong in one place operators can audit without reading the whole graph.

Agent-specific harnesses & benchmarks

When working through the Agent-specific harnesses benchmarks stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish. Log tool name, args hash, latency, and outcome for every call. Debugging agent loops without that trail wastes hours.

How to choose

When working through the How to choose stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Prefer small, testable units over sprawling scripts. When a step fails, the failure should point at a single responsibility rather than a tangled pipeline. Checkpoint after expensive steps. Resume should not re-bill the same LLM call when an operator retries a later node.

Operational checklist

For the Operational checklist stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state.

Record timings and token or query cost next to functional results. Cost visibility early prevents surprise bills when the path moves from demo to shared environments.

Put human approval on edges that spend money or change production data. Compile-time wiring does not equal business completeness.

Track cost and latency beside quality. A slightly worse answer that costs 10x less may be the right production trade.

Pin dependency versions and record the image digest that ran the demo. Reproducibility beats tribal knowledge.

Prefer small, testable units over sprawling scripts. When a step fails, the failure should point at a single responsibility rather than a tangled pipeline.

Before promoting the stack, freeze versions, capture a golden transcript for the critical path, and confirm rollback steps. Shared environments need rate limits, tenancy checks, and a clear owner for secret rotation. Prefer boring reliability over clever one-off demos.

Batch note for ce69f1512f25: keep provider keys out of the repo, set a per-session token ceiling, and store transcripts next to the eval fixtures so later model swaps stay comparable.