agentmods.dev

dym-harness-sdk-evals

dym-harness-sdk-evals is a skill for Claude Code from Rheinmir/setup. It costs 68 tokens per session (661 once invoked), scanned A, original, MIT.

A documented, but not yet verified, workflow for scoring Strands agents with the Strands Evals SDK. It describes tests that judge an answer, the agent's action trace, or a whole session, and can fail CI below a chosen threshold.

In plain words
What is it for?
Use it to design agent evaluation cases, score outputs or traces, investigate failed runs, and set a quality gate in CI, while treating the documented behavior as unverified.
Why use it?
It separates whether an agent succeeded from why it succeeded or failed, and can inspect saved production traces without rerunning the agent.

Skill for Claude Code

Written for Claude Code: disable-model-invocation in frontmatter.

not rated 11repo 20d ago A scan 68 tokens original MIT

Good fit Use it to design agent evaluation cases, score outputs or traces, investigate failed runs, and set a quality gate in CI, while treating the documented behavior as unverified.

Compare 6 skills from other repositories ↓
Install with agentmods
npx agentmods add skills/rheinmir/setup/dym-harness-sdk-evals
View source ↗ Rheinmir/setup
Install

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.

Any agent
npx skills add Rheinmir/setup --skill dym-harness-sdk-evals
Clone the repo
git clone --depth 1 https://github.com/Rheinmir/setup

Made for: Claude Code.

Wrote 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.

agentmods badge for dym-harness-sdk-evals

README.md
[![agentmods](https://agentmods.dev/badge/skills/rheinmir/setup/dym-harness-sdk-evals/github.svg)](https://agentmods.dev/skills/rheinmir/setup/dym-harness-sdk-evals)
Your own site
<a href="https://agentmods.dev/skills/rheinmir/setup/dym-harness-sdk-evals"><img src="https://agentmods.dev/badge/skills/rheinmir/setup/dym-harness-sdk-evals/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.

agentmods 80×15 button for dym-harness-sdk-evals

Your own site · 80×15
<a href="https://agentmods.dev/skills/rheinmir/setup/dym-harness-sdk-evals"><img src="https://agentmods.dev/badge/skills/rheinmir/setup/dym-harness-sdk-evals.svg" alt="Reviewed on agentmods" width="80" height="20"></a>
Per session 68 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 661 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe.
Origin original No closer match found in the catalogue.
Token cost

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.

ModelPer sessionOnce invoked
Fable 5.1 $0.00068 $0.00661
Opus 5.5 $0.00027 $0.00264
Sonnet 5.5 $0.00014 $0.00132
Haiku 4.5 $0.00007 $0.00066

Measured yesterday against content hash 83b392a4594c, method: parsed. Prices are Anthropic first-party input rates as of 2026-10-07, from the pricing page.

Security

Grade A, and why

dym-harness-sdk-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.

.overstack/doyourmagic/harness-sdk/skills/dym-harness-sdk-evals/SKILL.md · 27 lines

What it actually says

Skill: dym-harness-sdk-evals — chấm điểm agent (chưa kiểm chứng)

When to use

  • Cần điểm số chất lượng (không chỉ pass/fail hiệu ứng) cho agent Strands, hoặc chấm trace production đã có.
  • Trạng thái: mọi thứ dưới đây lấy từ docs site/src/content/docs/.../evals-sdk/ trong harness-sdk; code nằm ở repo strands-agents/evals chưa clone, chưa chạy. Coi là giả thuyết.

Steps

  1. Mô hình: Experiment → Case → Task → Evaluator → Detector (how-evaluation-works.mdx:17-48).
    • Evaluator chấm ở 3 mức: OUTPUT (câu trả lời) · TRACE (chuỗi span/tool) · SESSION (cả phiên).
    • Evaluator có loại LLM-judge và loại tất định (evaluators/deterministic_evaluators.mdx).
    • Detector đọc trace của run FAIL để tìm nguyên nhân (tách "có đạt không" khỏi "vì sao").
    • Simulator đóng vai user/tool cho test nhiều lượt.
  2. Chấm trace production không cần chạy lại agent: trace provider lấy span OTel từ CloudWatch/Langfuse (how-to/trace_providers.mdx). Harness phát span khi OTEL_TRACES_EXPORTER được đặt.
  3. Gate CI: CLI run ... --fail-on threshold:0.8 (cli/run.mdx:104,110); kết quả task cache theo case.name (how-to/result_caching.mdx:44) — đổi input mà giữ tên case thì dính cache cũ.
  4. Đa phương thức (design 0010): MLLM-as-judge, Likert-5 cho chất lượng tổng + nhị phân cho correctness/faithfulness.

Rules

  • Chưa kiểm chứng — trước khi dùng: clone strands-agents/evals, đối chiếu --help thật, sửa file này.
  • Harness-sdk KHÔNG dùng evals SDK trong CI của chính nó (grep strands_evals trong harness-py/ts/.github = rỗng) → đổi default harness không có regression eval tự động.
  • Con số "benchmarked defaults" (trung bình 6 benchmark: alfworld, ContextBench, GAIA, WebShop, τ³-bench, Terminal-Bench 2.1) chép tay trong site/public/embeds/strands-harness-benchmarks.html; file số liệu gốc và harness chạy benchmark (strands-labs/benchmark-harnesses, Harbor trên EC2) KHÔNG có trong repo → không tái lập được.
Changes

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.

  1. yesterday First seen · 27 lines · 68 tokens per session scan A 83b392a4594c

Subscribe to this mod's changes

dym-harness-sdk-evals is a skill published in the GitHub repository Rheinmir/setup (11 stars, last pushed 20d ago), licensed MIT. It adds 68 tokens to every session and 661 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 2026-11-07.

Building agentsTesting
Related

dym-harness-sdk-evals compared with skills from other repositories

dym-harness-sdk-evals next to 6 skills from other repositories
ModTypeStarsLast pushLicenceScanTokensWorks with
dym-harness-sdk-evalsRheinmir/setupSkill1120d agoMITA68Claude Code
imagegenopenai/codexSkill134ktodayApache-2.0A113Codex
local-ai-agentsmicrosoft/ai-agents-for-beginnersSkill79kyesterdayMITA200Claude Code, Codex
playwright-climicrosoft/playwrightSkill100kyesterdayApache-2.0A19Claude Code
browser-automationopenclaw/openclawSkill394ktodayMITA36Claude Code
systematic-debuggingobra/superpowersSkill314k4d agoMITA21Claude Code
next-cache-components-adoptionvercel/next.jsSkill143ktodayMITA95Claude Code