Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add MLOps-Courses/agentops-open-course --skill agent-evaluationgit clone --depth 1 https://github.com/MLOps-Courses/agentops-open-courseWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/mlops-courses/agentops-open-course/agent-evaluation)<a href="https://agentmods.dev/skills/mlops-courses/agentops-open-course/agent-evaluation"><img src="https://agentmods.dev/badge/skills/mlops-courses/agentops-open-course/agent-evaluation/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/mlops-courses/agentops-open-course/agent-evaluation"><img src="https://agentmods.dev/badge/skills/mlops-courses/agentops-open-course/agent-evaluation.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00070 | $0.00591 |
| Opus 5.5 | $0.00028 | $0.00236 |
| Sonnet 5.5 | $0.00014 | $0.00118 |
| Haiku 4.5 | $0.00007 | $0.00059 |
Grade A, and why
agent-evaluation scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 29d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 36 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Agent Evaluation
Score an agent's behavior over fixed cases, not one exact string. Let model-free structure checks gate merges. Treat every model-backed scorer as evidence with an explicit floor, and never make an LLM judge the sole release criterion.
When to use
- A prompt/model change can pass a smoke test yet call the wrong tools, cost more, or hallucinate.
- You want deterministic evalset validation in pull requests and behavioral evidence before release.
- You need to choose between two prompt versions with numbers, not opinion.
Steps
- Score the trajectory, not the wording. Assert which tools were called, with which arguments, in order (allow extra reads) over fixed seed cases; hold writes to an exact count. This survives non-determinism that exact-match scoring cannot.
- Grow the set from real failures. When a trace shows a wrong or unsafe trajectory, distil it into one case that pins that single behavior and makes a recurrence visible.
- Add a groundedness check. Require every recognized claim to appear in that turn's retrieved evidence or the user's question. Document the recognizer's vocabulary; use a broader extractor or judge for claims it cannot parse.
- Tripwire the cost. Record per-case tokens and model-call counts against a committed baseline; fail when a change exceeds a tolerance. Trajectory scores tolerate waste, so this is the only thing that catches a correct-but-expensive regression.
- A/B prompt versions. Run the eval set under two pinned prompt versions in isolated processes and print a per-scorer delta; promote or roll back on the numbers.
- Split gates from evidence. Deterministic, model-free checks gate CI; model-backed evals (judge, cost, groundedness) run on a schedule where a failure means "investigate", not "blocked".
Reference implementation
From the AgentOps Open Course (agent modules live under agents/python/src/agent/, eval scripts under agents/python/evals/):
evals/mlflow_eval.py— deterministic trajectory/response scorers + optional judge.evals/groundedness_eval.py,evals/cost_eval.py,evals/prompt_ab.py.- Course chapter
4.4. Evaluations.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 29d ago Changed · -1 lines · -1 tokens per session 308cdc936745
- 1mo ago First seen · 37 lines · 71 tokens per session scan A 1539d25490ca
agent-evaluation is a skill published in the GitHub repository MLOps-Courses/agentops-open-course (2 stars, last pushed yesterday), licensed MIT. It adds 70 tokens to every session and 591 once invoked, about $0.0003 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
skill-release-gate
Evaluate an Agent Skill bundle for structural integrity, trigger quality, artifact improvement, script correctness, safety, installed-tree integrity, and target-host portability before release.
check-tests
Run the local test evidence check.
mirrord-prev-env
Help users create and manage mirrord preview environments — running a modified service as an isolated pod in a shared Kubernetes cluster, scoped by an environment key and HTTP/queue traffic filtering, so teams can validate and review changes against real traffic without affecting live services. Use when a developer…
mirrord-chaos
Help users chaos test their app with mirrord: inject artificial latency or connection errors into a mirrord session's outgoing traffic via per-session chaos rules managed with the mirrord chaos CLI. Use when a user wants to add latency or delay to outgoing connections or a dependency (e.g. a slow database), simulate…
routing-calibration
Use when calibrating or validating a maintained routing recipe against live model backends, including probe manifests and evidence reports.
testing-patterns
Testing patterns for Spring Boot - JUnit 5, Mockito, MockMvc, Testcontainers, positive and negative test cases. Use when user asks about testing, writing tests, test coverage, or test failures.