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 0xDarkMatter/claude-mods --skill evals-opsgit clone --depth 1 https://github.com/0xDarkMatter/claude-modsWrote 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/0xdarkmatter/claude-mods/evals-ops)<a href="https://agentmods.dev/skills/0xdarkmatter/claude-mods/evals-ops"><img src="https://agentmods.dev/badge/skills/0xdarkmatter/claude-mods/evals-ops/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/0xdarkmatter/claude-mods/evals-ops"><img src="https://agentmods.dev/badge/skills/0xdarkmatter/claude-mods/evals-ops.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 | $0.00154 | $0.04308 |
| Opus 5.5 | $0.00062 | $0.01723 |
| Sonnet 5.5 | $0.00031 | $0.00862 |
| Haiku 4.5 | $0.00015 | $0.00431 |
Grade A, and why
evals-ops 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 1mo 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 — 300 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Evals Ops
Evals are the prerequisite, not the polish. You cannot tune a prompt, a retriever, a compaction strategy or a memory layer without a harness that says whether the change made things better. Teams that skip this ship vibes and learn about regressions from users.
This skill is the operational layer: what to measure, how to build the dataset, how to make a judge trustworthy, and how to gate CI on it without teaching everyone to ignore red.
Route first
| The ask | Go to |
|---|---|
| "What should I even measure?" | Three levels → references/eval-taxonomy.md |
| "Where do the test cases come from?" | Golden set → references/golden-datasets.md |
| "My judge disagrees with me / is it any good?" | Judges → references/llm-judge.md |
| "Verify a finding is real, not plausible" | Refuters → references/adversarial-verification.md |
| "Is my RAG retrieval any good?" | Retrieval → references/retrieval-eval.md |
| "Where do the human labels come from?" | references/annotation-workflow.md |
| "Should this block the merge?" | Gating → references/regression-gating.md |
| "Did this change really make it worse?" | Is the drop real |
| "Optimise against the eval / run it overnight" | Hillclimbing → references/hillclimbing.md |
| "Which platform should we use?" | references/tooling-landscape.md |
| "Just give me a starting file" | Assets — golden set, rubric, runner, CI gate |
The 60-second version
- Write 20 cases before you write a metric. A dataset you can eyeball beats a metric you cannot interpret. Grow to 100-300, then freeze it.
- Prefer a deterministic assertion to any judge. The JSON parsed, the tool was called with the right argument, the query returned 3 rows - free, instant, zero variance. Reach for a judge only where correctness is genuinely a matter of language.
- Score the trajectory, not just the answer. Most teams only check the final artifact and are surprised when a right answer came from a wrong path that breaks tomorrow.
- Calibrate the judge against humans before trusting it. Cohen kappa, not raw
agreement.
scripts/judge-calibration.pydoes the arithmetic and the verdict. - Blocking gates must be deterministic. Judge metrics start advisory. One flaky red permanently devalues the signal.
What ships with it
17 files 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.
- assets/eval-gate.template.yml 6.3 KB
- assets/eval-runner.template.py 6.3 KB runs code
- assets/golden-set.example.jsonl 5.9 KB
- assets/judge-rubric.template.md 3.4 KB
- references/adversarial-verification.md 5.2 KB
- references/annotation-workflow.md 6.0 KB
- references/eval-taxonomy.md 5.4 KB
- references/golden-datasets.md 6.8 KB
- references/hillclimbing.md 9.8 KB
- references/llm-judge.md 7.7 KB
- references/regression-gating.md 8.4 KB
- references/retrieval-eval.md 6.4 KB
- references/tooling-landscape.md 5.0 KB
- scripts/eval-baseline.py 20 KB runs code
- scripts/goldenset-audit.py 17 KB runs code
- scripts/judge-calibration.py 13 KB runs code
- tests/run.sh 35 KB runs code
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.
- 1mo ago First seen · 300 lines · 154 tokens per session scan A 80089054569c
evals-ops is a skill published in the GitHub repository 0xDarkMatter/claude-mods (78 stars, last pushed 13d ago), licensed MIT. It adds 154 tokens to every session and 4,308 once invoked, about $0.0006 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 2026-09-17.
Other skills, from other repositories
llm-eval
LLM evaluation: build evaluation datasets, choose metrics (RAGAS, G-Eval, LLM-as-judge), run automated evals, monitor production quality, and detect regressions.
scikit-learn-train-test-splittings
Guidelines and instructions for Scikit-learn train test splittings.
cli-eval
Create and run evaluation suites, watch live benchmark progress, view scorecards, compare model performance, and integrate eval runs with CI workflows from the CLI.
data-quality-frameworks
Implement data quality validation with Great Expectations, dbt tests, and data contracts. Use when building data quality pipelines, implementing validation rules, or establishing data contracts.
emerging-techniques-model-merging
Merge multiple fine-tuned models using mergekit to combine capabilities without retraining. Use when creating specialized models by blending domain-specific expertise (math + coding + chat), improving performance beyond single models, or experimenting rapidly with model variants. Covers SLERP, TIES-Merging, DARE, Task…
darwinian-evolver
Evolve prompts/regex/SQL/code with Imbue's evolution loop.