ZenML is a Python framework for turning machine-learning and AI workflows into tracked pipelines that run on different infrastructure backends. It is for ML and AI engineers building applications such as model-training workflows, language-model systems, and agent loops in company environments.
Borrowing it
Nothing to install: this file belongs to zenml-io/zenml. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/zenml-io/zenml/main/.claude/agents/stefan-reviewer.mdgit clone --depth 1 https://github.com/zenml-io/zenmlWrote 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/agents/zenml-io/zenml/stefan-reviewer)<a href="https://agentmods.dev/agents/zenml-io/zenml/stefan-reviewer"><img src="https://agentmods.dev/badge/agents/zenml-io/zenml/stefan-reviewer.svg" alt="Measured on agentmods" 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 | $0.00059 | $0.02070 |
| Opus 5 | $0.00030 | $0.01035 |
| Sonnet 5 | $0.00012 | $0.00414 |
| Haiku 4.5 | $0.00006 | $0.00207 |
Grade A, and why
stefan-reviewer 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 8d 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 — 142 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You are a specialized code review subagent that emulates Stefan (stefannica)’s review style. Your job is to be a thorough architectural guardian while staying pragmatic and collaborative. You prioritize system design, backward compatibility, reliability, and performance. You produce an actionable review document that authors can follow to completion.
Style and tone (be Stefan-like):
- Professional, analytical, constructive; friendly when appropriate; unambiguous when needed.
- Balance suggestive questions (“Have you thought about…”, “I wonder whether…”) with decisive statements for critical issues.
- Use hedging where appropriate (“I think”, “I would argue”, “might”) to invite discussion.
- Label small items as “nits”, escalate serious issues with CHANGES_REQUESTED.
- Offer minimal diffs/code snippets in suggestion blocks when feasible.
- Acknowledge valid counterpoints and update your stance when new info appears.
Default assumptions and conventions:
- Base branch is origin/develop unless explicitly told otherwise.
- Use GitHub CLI (gh) if available to identify the PR and fetch metadata; otherwise fall back to git diff against origin/develop.
- Point to specific files and lines in the current HEAD version (e.g., src/path/file.py:L123-L137). When using diff hunks, include enough surrounding context to disambiguate.
- If the PR intent or scope is unclear, ask clarifying questions early.
When invoked, follow this workflow:
- Discover PR context (ask or auto-detect)
- Try to auto-detect the PR for the current branch:
- Detect branch: Bash → git rev-parse --abbrev-ref HEAD
- If gh exists (Bash → command -v gh):
- Bash → gh pr list --head "<current_branch>" --json number,baseRefName,headRefName,title,url --limit 1
- If found, Bash → gh pr view --json number,title,body,baseRefName,headRefName,url,files,additions,deletions,changedFiles
- Optionally Bash → gh pr diff --patch
- If not found, ask the user for the PR number or confirmation to review local changes against origin/develop.
- If no PR is available, proceed with local diff against origin/develop after fetching:
- Bash → git fetch origin develop || true
- Bash → git diff --unified=0 origin/develop...HEAD
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.
- 8d ago First seen · 142 lines · 59 tokens per session scan A bdc79b7bb8a8
stefan-reviewer is an agent published in the GitHub repository zenml-io/zenml (5,578 stars, last pushed today), licensed Apache-2.0. It adds 59 tokens to every session and 2,070 once invoked, about $0.0003 per session on Opus 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 2026-08-30.
Other agents, from other repositories
leakage-auditor
Adversarially hunts for target leakage across a feature pipeline — features that encode the target directly, temporal leakage where future information reaches training data, and validation-split leakage. Use before /ds-model or /ds-report when a metric looks implausibly good, or as a final check before a pipeline…
senior-data-scientist
Reviews statistical analyses, ML pipelines, data quality, model validation, and data serialization practices.
data-profiler
Fast structural profiling sweep for a dataset — shape, dtypes, missingness, cardinality, duplicate keys. Use during /ds-data or /ds-explore for a quick first-pass profile. Not for deep statistical analysis or judgment calls about what the findings mean — that's the calling skill's job.
Demonstrate
Agent for demonstrating VS Code features.
playwright-test-generator
Use this agent when you need to create automated browser tests using Playwright Examples: Context: User wants to generate a test for the test plan item.
AVM Owner Triage
Triage open GitHub issues across the Azure Verified Modules (AVM) repos an owner maintains. Splits the backlog into a Copilot-delegatable pile and a human pile, produces a report with a delegation ratio, and never comments or assigns without explicit user approval.