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 agentmods add skills/apify/apify-mcp-server/creating-workflow-evalsnpx skills add apify/apify-mcp-server --skill creating-workflow-evalsgit clone --depth 1 https://github.com/apify/apify-mcp-serverWhat 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 | $0.00069 | $0.01850 |
| Opus 5 | $0.00034 | $0.00925 |
| Sonnet 5 | $0.00014 | $0.00370 |
| Haiku 4.5 | $0.00007 | $0.00185 |
Grade A, and why
creating-workflow-evals 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 yesterday.
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 — 69 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Creating workflow evals for an MCP tool family
Overview
Build a small, calibrated Langfuse eval suite for one tool family (tasks, storage, runs, …), then use its failures to fix the tools. Core principle: evals are designed from user intent, never from tool descriptions — the eval defines what should work; descriptions get fixed afterward to make naive agents pass it.
Commands, item shapes, probe patterns, and sweep queries: reference.md.
The flow
- Inventory the tools — every tool and every argument group needs at least one case (the coverage matrix at the end proves it).
- Probe the platform first. Before writing any case that depends on API behavior (required fields, uniqueness rules, limits, error messages), verify it with a throwaway
tsxscript against the real API. Never write a case on an assumed contract — that's how you get input values the schema rejects. - Two datasets, never one:
<family>-evals(proper suite, zero tool errors tolerated) and<family>-evals-errors(cases that provoke errors on purpose: collisions, not-found, requirement discovery). Mixing them masks real failures. - Write cases in waves: 2–3 easy (single tool, explicit input) → 1–2 medium (cross-tool chains, run options) → 2–3 hard (vague user language, error recovery, collisions). Run and review each wave before writing the next.
- Calibrate on the strongest model first (Opus). A failure there is a case defect or a product gap — never a description problem. Only a calibrated suite (strong model 100%) can attribute weaker-model failures to descriptions.
- Ladder down (Sonnet → Haiku). Passes-on-Opus-fails-on-Haiku = the tool description or output doesn't carry a naive agent. That's the signal you built the suite for.
- Fix tools via outputs before descriptions. A steering sentence in the tool's response summary/nextStep reaches every agent on every call; description text gets skimmed. Both output nudges that fixed Haiku failures in the original build were response-text changes.
What ships with it
1 file beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
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.
- yesterday First seen · 69 lines · 69 tokens per session scan A c5f7dbe446d3
creating-workflow-evals is a skill published in the GitHub repository apify/apify-mcp-server (5,636 stars, last pushed 2d ago), licensed MIT. It adds 69 tokens to every session and 1,850 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 2026-09-02.
Other skills, from other repositories
unity-model-agnostic-inference
Use for on-device or server inference that is not tied to one vendor model: ONNX Runtime, Sentis, Barracuda legacy, custom .onnx/.tflite/.pb assets, and swapping models without rewriting gameplay code.
unity-llm-integration
Use when adding LLM/chat/RAG features to a Unity game or tool: API keys, prompt assets, safety, latency, and server-side vs client-side calls.
unity-ml-agents
Use when working on Unity ML-Agents: training configs, BehaviorParameters, sensors, actuators, inference in builds, or com.unity.ml-agents packages.
unity-sentis-inference
Use when integrating ONNX/Sentis/Unity Inference in Unity: model assets, IWorker, runtime performance, or com.unity.sentis packages.
ai-elements
Build AI chat interfaces using ai-elements components — conversations, messages, tool displays, prompt inputs, and more. Use when the user wants to build a chatbot, AI assistant UI, or any AI-powered chat interface.
grpo-finetune
Fine-tune a model with GRPO on Fireworks-managed GPUs from a plain-English task description and a dataset. Use this skill whenever the user wants to fine-tune, RL-tune, or GRPO-train a model on their own data — or says things like "train a model to extract/classify/score X", "fine-tune on this dataset", "set up a GRPO…