caliper-eval 0.8.0

CLI for evaluating Claude Code skills and AI agents

Caliper — Reliability testing for agent skills

Know whether your skill actually works. Write a short spec of what "good" looks like, run it k times, and get a success rate you can track. Caliper also runs the tasks without the skill, so you can see whether it's the skill or the base agent doing the work. Works with the agent you already use: Claude Code, Codex, Pi, or Hermes.

Teach your agent to evaluate:

npx skills@latest add edonadei/caliper

Or run it yourself:

caliper run commit-commands.eval.yaml --k 3 --baseline

You write a spec — a few lines of YAML describing what "working" means, which you hand-write or have /grill-skill generate for you. With --baseline, Caliper runs each task with and without the skill and diffs the two runs task by task:

─────────────────── CALIPER  —  compare  —  commit-commands ────────────────────
    no skill → with skill   ·   k=3

╭───────────────────────┬─────────────────┬────────┬───────────╮
│ Task                  │         success │      Δ │ attempts  │
├───────────────────────┼─────────────────┼────────┼───────────┤
│ Commits a new feature │  33.3% → 100.0% │ +66.7% │ ✓✗✗ → ✓✓✓ │
│ Commits a bug fix     │  33.3% → 100.0% │ +66.7% │ ✗✓✗ → ✓✓✓ │
╰───────────────────────┴─────────────────┴────────┴───────────╯

 Overall  33.3% → 100.0%   Δ (matched) +66.7% ↑
 Tokens  290K → 180K   Δ -38% (-110K)
 Wall    1m 1s → 42s   Δ -31% (-19s)

Every task got more reliable, and commit-commands got there on 38% fewer tokens and 31% less wall-clock time. The attempts column shows every attempt (✓/✗) before → after, so the bare agent's failures are right there — not hidden in an average.

---

Agent skills are hard to test. A skill that works on your machine, on this prompt, today, might fail tomorrow after a model update or a one-line prompt edit. Caliper makes reliability measurable: define what success looks like, run the skill repeatedly, and get a success rate you can track over time.

Use Caliper to answer questions like:

• Did my prompt edit actually improve the skill?
• Is the skill doing the work, or would the base agent pass without it?
• Does it still pass the workflows it passed last week?
• Which agent — Claude Code, Codex, Pi, or Hermes — runs this skill more reliably?

---

Quick start

Path A — Agentic (let your agent drive)

1. Install the skills

npx skills@latest add edonadei/caliper

2. Generate a spec interactively

In your agent (Claude Code or Codex):

/grill-skill ./my-skill/SKILL.md

grill-skill reads your SKILL.md, interviews you, and writes a 3-task .eval.yaml (happy path, edge case, adversarial).

3. Run and measure

/evaluate-skill run my-skill.eval.yaml --k 3 --baseline

Browse past runs:

/evaluate-skill list
/evaluate-skill report my-skill

Path B — CLI (run it yourself)

1. Install the CLI

pipx install caliper-eval   # requires Python 3.10+

2. Write a spec

# my-skill.eval.yaml
skill:
  path: ./SKILL.md

tasks:
  # Autorater — the LLM judge reads the transcript and decides
  - name: Writes a conventional commit message
    prompt: "Summarize the staged git diff as a commit message."
    expect: >
      The response is a conventional-commit message: a concise subject
      line under 72 characters, followed by a body explaining why the
      change was made, not just what changed.

  # Script execution — a deterministic Python assertion
  - name: Generates a valid config file
    cleanup: rm -f /tmp/app.config.json
    prompt: "Generate a config at /tmp/app.config.json with a 'port' of 8080."
    assert: |
      import json
      from pathlib import Path
      data = json.loads(Path("/tmp/app.config.json").read_text())
      assert data["port"] == 8080

expect: is graded by the judge LLM; assert: runs locally as Python. Use either or both.

The spec never names an engine — the skill and judge default to claude-code, and you pick a different agent/model at run time with --model / --judge-model (see Choosing an engine).

3. Run it

caliper run my-skill.eval.yaml --k 3          # add --baseline to diff vs the bare agent

4. Read the output

─────────── CALIPER  —  my-skill  (claude-code)  —  2026-06-19 14:23 ───────────

╭──────────────────────────────────┬───────┬─────────┬────────┬──────┬───────────╮
│ Task                             │ k (3) │ success │ Tokens │ Wall │           │
├──────────────────────────────────┼───────┼─────────┼────────┼──────┼───────────┤
│ Writes a conventional commit msg │  3/3  │  100.0% │    79K │  27s │  ✓ PASS   │
│ Generates a valid config file    │  2/3  │   66.7% │    82K │  33s │ ~ PARTIAL │
╰──────────────────────────────────┴───────┴─────────┴────────┴──────┴───────────╯

 Score   83.3%  █████████████████░░░

 Tokens   159K in / 2K out
 Wall     1m 0s  10.0s per attempt

Results saved to .caliper/results/my-skill/2026-06-19T14-23-01Z.json

--verbose adds pass@k and pass^k columns (both derived from the raw rate).

Not sure what to put in a spec?

The Eval Starter Pack has four copy-paste templates, each catching a real agent failure (false success, tool misuse, runaway loops, prompt regressions). Every template runs green as-is against a bundled example, then points at your own skill by editing two or three commented lines.

---

How it works

.eval.yaml spec
      │
      ▼
  Harness  ──── runs your skill against the agent (Claude Code / Codex / Pi)
      │
      ▼
   Judge   ──── LLM autorater and/or deterministic Python assertions
      │
      ▼
  success rate + saved transcript

Each attempt runs in an isolated temporary home with no session history. Results are saved as JSON you can inspect and diff later.

---

Agent skills

The repo ships two agent skills. Install both with:

npx skills@latest add edonadei/caliper

evaluate-skill — run and manage evals

Create, validate, run, and summarize evals from inside your normal workflow — no separate terminal needed. The skill installs Caliper automatically if it's missing.

Then use it in Claude Code:

/evaluate-skill run my-skill.eval.yaml --k 3
/evaluate-skill validate my-skill.eval.yaml

Or in Codex:

Use the evaluate-skill skill to run my-skill.eval.yaml with k=3 and summarize the result.

grill-skill — create evals interactively

Don't have evals yet? grill-skill guides you through creating them. It reads your SKILL.md, interviews you about what good behavior looks like, and generates a 3-task spec (happy path, edge case, adversarial). Then it runs the eval and loops — k=1 to validate, k=3 to measure, baseline before you commit.

/grill-skill ./my-skill/SKILL.md

No path needed if you're already in the skill's directory:

/grill-skill

If an .eval.yaml already exists next to your skill, grill-skill reads the existing tasks and interviews you about gaps instead of starting from scratch.

---

Core concepts

Term	What it is
Spec	A .eval.yaml file that describes the skill, judge, and tasks to run
Backend	The CLI agent that executes the skill (claude-code, codex, pi, hermes)
Judge	What decides pass/fail — an LLM reading the transcript (expect:), Python assertions (assert:), or both
success rate	The primary score: run k times, measure how often a single run works (pass@k/pass^k are secondary views, under --verbose)
Baseline	Re-run the same tasks without the skill to prove the skill is doing the work
Attempt	One isolated run of a single task — fresh temporary home, no session history

---

Choosing an engine

The engine (backend + model) is a runtime axis, not a spec field — the spec describes what is tested and how success is judged, and you pick the agent that runs and grades it at invocation. Both default to claude-code; select a different one with --model / --judge-model:

caliper run my-skill.eval.yaml                          # claude-code (default)
caliper run my-skill.eval.yaml --model codex            # codex, its default model
caliper run my-skill.eval.yaml --model codex:gpt-5-codex
caliper run my-skill.eval.yaml --model pi --judge-model claude-code

Backend	Requires	Best for
claude-code	Claude Code CLI installed and authenticated	Testing Claude Code slash-command skills
codex	Codex CLI installed (npm install -g @openai/codex)	Testing Codex skills
pi	pi CLI installed (npm install -g @earendil-works/pi-coding-agent) and authenticated	Testing pi skills (agentskills.io)
hermes	Hermes Agent CLI installed and authenticated (Nous Research)	Testing skills on Hermes; hermes:<provider>/<model> selects the model

Caliper runs skills only through CLI agents — every backend can actually load and run a skill. There is no direct-API backend: to run against API-priced billing, configure one of these CLIs with an API key (e.g. ANTHROPIC_API_KEY / OPENAI_API_KEY) rather than selecting a separate backend.

The skill engine and judge engine are independent — you can test a Codex skill with a Claude judge, or any other combination, by pairing --model with --judge-model.

Claude Code setup

Install and authenticate the claude CLI. --model claude-code uses your existing Claude Code auth — no extra configuration needed.

Codex setup

npm install -g @openai/codex
codex login

--model codex calls codex exec. If the Codex desktop app is installed, Caliper prefers the app-bundled binary over codex on PATH. Set CODEX_CLI_PATH to force a specific binary.

pi setup

npm install -g @earendil-works/pi-coding-agent
pi   # then authenticate (e.g. /login for a subscription provider, or set the provider API key)

--model pi runs pi --print --mode json and loads the skill natively via pi's --skill flag (the agentskills.io standard). It reuses your ~/.pi/agent auth and settings — the :model half of --model pi:<model> overrides pi's configured default when set. Set PI_CLI_PATH to force a specific binary. Note: pi's built-in default provider is google, so running --model pi with no model relies on your pi config to resolve a provider you are authenticated for.

Hermes setup

curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes login   # authenticate; pick a default model/provider you have credits for

Hermes is a stateful, always-on agent (persistent memory, a persona, auto-generated skills), so Caliper normalizes it to a neutral agent to keep its score apples-to-apples with the other backends: every attempt runs in an isolated HERMES_HOME seeded with your ~/.hermes auth/config only (never SOUL.md/MEMORY.md) and with --ignore-rules, and the skill-under-test is staged as the sole local skill. --model hermes runs hermes -z (oneshot) then hermes sessions export to recover the full tool-call trajectory; --model hermes:<provider>/<model> (e.g. hermes:anthropic/claude-sonnet-4.6) selects the model, otherwise your ~/.hermes/config.yaml default is used — point it at a provider you have credits for. Set HERMES_CLI_PATH to force a specific binary. Hermes updates itself (hermes update), so it is not part of caliper update-cli.

Check installed CLI versions:

caliper update-cli --check

---

Recommended workflow

1. Create a spec for one behavior you care about.
2. Run with --k 1 while iterating on the spec.
3. Add assert: for facts an LLM judge might guess wrong (files, JSON, command output).
4. Move to --k 3 or higher once the task is stable.
5. Add --baseline to prove the skill is making a difference.
6. Commit the spec alongside the skill so contributors can run the same eval.

/evaluate-skill run my-skill.eval.yaml --k 3 --baseline --verbose

---

Spec format

To scaffold a spec, use the evaluate-skill or grill-skill skill, or hand-write the YAML below.

skill:
  path: ./SKILL.md              # path to the skill file (optional for baseline-only runs)

# Note: there is no `backend`/`model` or `judge:` block. The engine is a runtime
# axis — pass `--model` / `--judge-model` at run time (default: claude-code).

sandbox:
  extra_path:
    - ./bin                     # prepended to PATH inside each attempt
  forbidden_files:
    - ".*\\.eval\\.yaml$"       # prevents agent from reading the spec
    - "./.caliper/.*"           # prevents agent from reading saved results

tasks:
  - name: Short task name
    setup: <shell command>      # optional, runs before each attempt
    cleanup: <shell command>    # optional, always runs after each attempt
    prompt: <prompt sent to the agent>
    expect: <natural-language success condition>
    assert: |
      # optional inline Python assertion
      assert True

  - name: Task with external assertion script
    prompt: "Generate a report"
    assert: ./assertions/check_report.py

Each task needs at least one of expect or assert. Task IDs are assigned automatically as task-001, task-002, and so on.

---

Judging

LLM autorater (expect:)

The judge engine reads the full attempt transcript and decides whether the expect condition was met. When the backend captures tool-call traces (Claude Code, Codex, pi, Hermes), those traces are included — the judge can verify things like "the agent used tool X" without relying on the final text alone.

The judge engine is chosen at run time and defaults to claude-code; point it at a different agent with --judge-model (e.g. --judge-model codex), independently of the skill's --model.

Deterministic assertions (assert:)

Python assertions run locally. Use these for facts the LLM judge might guess:

• file exists / exact file contents
• JSON / schema validity
• command output
• images or screenshots
• repository state

tasks:
  - name: Writes an output file
    cleanup: rm -f /tmp/out.txt
    prompt: "Write hello world to /tmp/out.txt"
    assert: |
      from pathlib import Path
      path = Path("/tmp/out.txt")
      assert path.exists(), "Output file was not created"
      assert path.read_text().strip() == "hello world"

When both expect and assert are present, both must pass.

---

CLI reference

Command	Description
caliper run <spec>	Run an evaluation spec
caliper validate <spec>	Validate a spec file
caliper list [spec]	List specs and saved runs
caliper report <spec-or-result>	Re-render saved results
caliper compare <A> <B>	Diff two saved runs of the same eval, task by task
caliper update-cli [backend]	Check or update installed agent CLI versions

caliper run flags

Flag	Default	Description
--k INT	3	Attempts per task
--baseline	off	Also run each task without the skill
--workers INT	4	Parallel task workers
--timeout INT	120	Seconds per attempt
--fail-fast INT	0	Stop a task after N consecutive infra_error/timeout attempts (0 disables)
--model TARGET	claude-code	Skill engine — backend and/or model (see below)
--judge-model TARGET	claude-code	Judge engine — backend and/or model (see below)
--verbose	off	Show per-attempt judge reasoning
--output PATH	—	Also save results JSON to a specific path

--model and --judge-model syntax

The engine is not stored in the spec — these flags select it, defaulting to claude-code when omitted. Both accept a backend:model compound value, a bare backend name, or a bare model name:

# Backend and model together
caliper run my-skill.eval.yaml --model codex:gpt-5-codex

# Backend only (that backend's default model)
caliper run my-skill.eval.yaml --model codex

# Model only (backend stays claude-code)
caliper run my-skill.eval.yaml --model claude-sonnet-4-6

# Select the judge engine independently
caliper run my-skill.eval.yaml --model codex --judge-model claude-code:claude-haiku-4-5-20251001

Accepted backends: claude-code, codex, pi, hermes (alias: claude → claude-code). The actual engine used is recorded in each saved run's RunMeta — the skill backend/model, and the judge_backend/judge_model that graded it — so results stay traceable even though the spec doesn't pin it. When you don't name a model and the CLI uses its own default, RunMeta records the concrete model the agent resolved rather than a bare "default", wherever the backend reports it — the skill model from hermes' session export, and the judge_model from the claude-code judge's JSON output. judge_model stays empty for an assert:-only run, where no LLM judge fired.

---

Comparing two runs (caliper compare)

An ablation compares two runs of the same eval — a full skill vs. a shortened variant, or the same skill over time. caliper compare <A> <B> diffs two already-saved runs task by task, so you don't hand-write a JSON script to answer "did this change regress?".

# Latest run of each spec (a bare spec name resolves to its latest run)
caliper compare commit-simple-full commit-simple-short

# Pin specific runs by pointing at their results JSON
caliper compare .caliper/results/demo/2026-07-01T10-00-00Z.json \
                .caliper/results/demo/2026-07-02T09-00-00Z.json

# Machine-readable diff for a ship / no-ship decision
caliper compare A B --format json

Each positional (A, B) is addressed exactly like report's argument: a spec name (→ its latest run) or a path to a results JSON. There are no --run-a/-b flags — pin a historical run by naming its JSON path.

──────────────────── CALIPER  —  compare  —  commit-simple ─────────────────────
    2026-07-01T10-00-00Z (claude-code) → 2026-07-02T09-00-00Z (claude-code)   ·   k=5

╭──────────────────┬─────────────────┬────────┬───────────────╮
│ Task             │         success │      Δ │ attempts      │
├──────────────────┼─────────────────┼────────┼───────────────┤
│ commits cleanly  │ 100.0% → 100.0% │      — │ ✓✓✓✓✓ → ✓✓✓✓✓ │
│ handles conflict │  100.0% → 20.0% │ -80.0% │ ✓✓✓✓✓ → ✓✗✗✗✗ │
│ pushes upstream  │       80.0% → — │      — │ ✓✓✓✓✗ → ⊘⊘⊘⊘⊘ │
╰──────────────────┴─────────────────┴────────┴───────────────╯

 Overall  100.0% → 60.0%   Δ (matched) -40.0% ↓
 Tokens  1.2M → 700K   Δ -42% (-500K)
 Wall    6m 18s → 3m 40s   Δ -42% (-2m 38s)
 ⚠ 1 regression: handles conflict
 ⊘ 1 unmeasured (excluded from Δ): pushes upstream
 unmatched — only in A: flaky task   only in B: new task

How the diff reads:

• Each row reads before → after. The two runs are named once in the header (here, the two timestamps; a --baseline run says no skill → with skill), so there's no A/B legend to decode.
• Tasks are matched by name. task_id is only positional, so name is the stable identity — reordered tasks still line up. A task present in only one run is listed as unmatched and left out of the delta.
• Δ is after − before. A negative Δ renders red and flags the task as a regression (any-below rule: the after side below the before side).
• The score excludes unusable attempts. The attempts column reuses the run report's glyphs; ⊘ marks an unusable attempt (rate-limit / timeout / judge error). A side with no usable attempts shows — (unmeasured) and is never counted as a regression — infra noise can't fake a loss.
• The headline Δ (matched) averages each side over only the tasks measured on both sides, so it is strictly like-for-like.
• Token and wall-clock deltas sit under the headline as secondary signals: a drop renders green (cheaper), a rise red (costlier). They are never a regression — a token drop at a flat score is the win an ablation is looking for, and a rise is a trade-off to weigh, not a failure. Only the score feeds has_regression. The token row is shown only when both runs reported tokens; wall time is always shown. Dollar cost is deliberately not tracked (tokens are the volume signal).
• Guards for a k mismatch or different spec names print as warnings in the header and in --format json (k_mismatch, spec_mismatch, warnings), so an agent driving compare sees them too.

The --format json output serializes the full comparison (per-task scores, deltas, regression/has_regression flags, unmatched task lists, the warnings, and per-side usage totals a_usage/b_usage with token_delta/wall_delta) for scripting.

---

Scoring

Every attempt carries a typed outcome, so infrastructure and judge noise are not scored as task failure:

Outcome	Meaning	Counts toward the score?
pass	satisfied the task's judge(s)	✅ success
task_fail	the skill genuinely failed the task	✅ attempt
cheat	a forbidden-file read was detected	✅ attempt
infra_error	harness failure — nonzero exit, or a detected rate-limit / spending-cap	❌ unusable
timeout	exceeded the time budget with no result	❌ unusable
judge_error	the judge produced no verdict (unparseable / errored autorater)	❌ unusable

passed is retained in the JSON as a convenience, equal to outcome == pass.

The primary metric is the raw success rate — how often a single run works, computed over the usable attempts (those that got a fair shot); unusable attempts leave the denominator and are reported as a separate "N unusable" count:

usable  = pass + task_fail + cheat
score   = successes / usable                # raw rate; None if usable == 0

Two secondary views are kept for anyone who wants them (shown under --verbose, and on every task in the JSON as pass_at_k / pass_hat_k):

pass@k  = 1 - (1 - score) ^ usable   # P(≥1 of k passes)
pass^k  = score ^ usable             # P(all k pass)

Which one to look at depends on how the skill is actually used:

The question you're asking	Metric	For a 1/3 skill (k=3)
How reliable is a single run? (default)	success rate	33%
If I retry up to k times and keep any win, do I get one?	pass@k	70%
Will it work on every run, no exceptions?	pass^k	4%

Reach for pass@k when a failure is cheap to retry and you can keep the good run (a human re-runs it, or you regenerate until it works) — it's the retry-optimistic, "eventual success" view, always ≥ the raw rate. Reach for pass^k when the skill runs unattended or as one link in a chain, where a single failure breaks the whole thing — it's the strict "must never fail" view, always ≤ the raw rate. When in doubt, the success rate is the honest middle: pass@k is the code-generation metric (generate k, keep the best) and flatters flaky skills (1/3 → 70.4%), which is why Caliper leads with the rate.

The aggregate is the average task success rate, skipping tasks with no usable attempts. With --baseline, Caliper runs the same tasks without the skill and reports the delta. A throttled or judge-flaked run no longer masquerades as a regression.

For persistent infrastructure failures, --fail-fast N can stop scheduling new attempts for a task after N consecutive infra_error or timeout outcomes. The default 0 keeps the historical behavior and runs all k attempts. An early-stopped task is shown as ABORTED in the report; if every completed attempt was unusable, its score remains null and it is skipped in the aggregate score.

---

Token & wall-clock usage

Pass@k tells you whether a skill works; usage tells you what it costs to get there. Two runs can have identical scores while one burns twice the tokens. Caliper records token volume and wall-clock time per attempt and rolls them up per run:

 With skill    100.0%  ████████████████████

 Tokens   1.2M in / 340K out
 Wall     6m 18s  12.6s per attempt
 ⊘ unusable spend: 180K tokens, 42s  (2 attempts, not counted in the average)

• The results table carries per-task Tokens (total) and Wall columns, so you can see which task is the expensive one at a glance; the run summary line below it aggregates across the whole run.
• Each AttemptRecord carries an optional usage object with input_tokens (non-cached prompt), output_tokens, cache_read_tokens, cache_creation_tokens, and a computed total_tokens. The four token fields are disjoint — input_tokens excludes cache — so total_tokens never double-counts. duration_seconds (already recorded) is the wall-clock half.
• in = input + cache_read + cache_creation; out = output. Every attempt counts toward the run total; the unusable slice (timeout / infra / judge error) is broken out separately so wasted spend is visible without distorting the per-usable-attempt average.
• All usage fields are optional — a backend that can't report them leaves them null and the line renders —. Support: claude-code, codex, pi, and hermes all report token usage; codex uses OpenAI semantics (its input_tokens includes cache) so it is normalized to the non-cached contract.
• Dollar cost is deliberately not tracked — it is inconsistent across backends and would need a maintained price table. Tokens are the volume signal; a dollar figure can be derived downstream if needed.
• With --baseline, the whole no-skill run is retained (RunResults.baseline_task_results) and the report renders as a compare view — the same table, attempt strips, and token/wall deltas — so the skill-vs-bare- agent difference is shown side by side (see the quickstart at the top).
• report --format json includes a derived usage_totals block; the saved results JSON keeps the raw per-attempt usage (totals are derived, never persisted), plus the full baseline_task_results when --baseline ran. compare surfaces token/wall deltas across two runs — see above.

---

Project layout

caliper/
  commands/       CLI command implementations
  harness/        Agent execution backends (Claude Code, Codex, pi, Hermes)
  judge/          LLM and script judging implementations
  schema/         Eval spec and result models
  runner.py       Evaluation orchestration
skills/
  evaluate-skill/ Agent skill for running Caliper from Claude Code or Codex
  grill-skill/    Agent skill for creating and iterating on evals interactively
tests/            Pytest coverage for harnesses, judges, and runner behavior

---

Contributing

Contributions are welcome. See CONTRIBUTING.md for good first areas, the pre-PR checklist, the ruff formatting convention and pinned version, and the one-time pre-commit install step.

---

Troubleshooting

codex judge failed: model ... is not supported The model name is not available to your Codex account. Use a model that codex exec --model <name> accepts.

codex CLI not found

npm install -g @openai/codex

claude command not found Install and authenticate Claude Code, or switch the backend to codex or pi.

A task passes only because of assert: When a task has only assert:, no LLM judge runs. Add expect: if you also want an LLM to evaluate the transcript.