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 Arenukvern/skill_steward --skill skill-eval-improvegit clone --depth 1 https://github.com/Arenukvern/skill_stewardWrote 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/arenukvern/skill_steward/skill-eval-improve)<a href="https://agentmods.dev/skills/arenukvern/skill_steward/skill-eval-improve"><img src="https://agentmods.dev/badge/skills/arenukvern/skill_steward/skill-eval-improve/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/arenukvern/skill_steward/skill-eval-improve"><img src="https://agentmods.dev/badge/skills/arenukvern/skill_steward/skill-eval-improve.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 | $0.00061 | $0.02411 |
| Opus 5 | $0.00030 | $0.01205 |
| Sonnet 5 | $0.00012 | $0.00482 |
| Haiku 4.5 | $0.00006 | $0.00241 |
Grade A, and why
skill-eval-improve 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 9d 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.
This is a copy
100% identical to skill-eval-improve — 0 lines differ, which has more behind it and is treated as the original. This page carries a canonical link to it rather than competing with it.
How it starts
The opening of the file, as written. The whole thing — 203 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Skill eval & improve
Improve skills measurably: baseline → measure → bounded edit → re-validate. Combine local tooling, Codex plugin-eval (when installed), and research-backed loops (SkillOpt).
When to use
- Skill triggers wrong or never loads (description routing)
- Bloated
SKILL.md, high token cost, weak outcomes - After adding a new procedure—need regression checks
- Porting patterns from product MCP / plugin-eval research into Skill Steward skills
When not to use
- Bulk repo validation — e.g. “validate every skill in this repo” →
pnpm run validateonly (skill-authoring-lifecycle for audit); do not start benchmark or SkillOpt loops. - Automated SkillOpt / cluster training — Skill Steward documents a manual bounded-edit loop; no overnight optimizer pipeline.
- Creating a new skill — use skill-authoring-lifecycle first; eval-improve applies after a skill exists.
Cursor scope (optional): activate when editing under skills/** or scripts/validate-skills.mjs.
Mixture of experts (evaluation stack)
| Layer | Expert | Tool / method | Cost |
|---|---|---|---|
| 0 — Gate | Lint | pnpm run validate, skill-authoring-lifecycle |
seconds |
| 0b — Rules | Routing/docs SSOT | pnpm run eval (T1 behavior-critical YAML cases) |
seconds |
| 1 — Static | Structure | Codex plugin-eval analyze (if available) |
seconds |
| 2 — Human | Behavior | 3–5 prompts with/without skill | minutes |
| 3 — Measured | Usage | plugin-eval benchmark + measurement-plan |
minutes–hours |
| 4 — Evolve | Text optimization | SkillOpt-style bounded edits + held-out gate | hours |
| 5 — Navigate | Telemetry / dogfood | Current steward benchmark scenarios on compact traces |
seconds |
Use the cheapest layer that answers the question. Do not skip layer 0.
Layer 0 — Skill Steward validator (always)
What ships with it
11 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.
- evals/cases/duplicated-guidance-compression-trigger.yaml 702 B
- evals/cases/improve-routing-trigger.yaml 444 B
- evals/cases/tool-loop-drift-trigger.yaml 521 B
- evals/cases/validate-only-dormant.yaml 425 B
- references/chrome-eval-design.md 2.1 KB
- references/eval-case-schema.md 1.8 KB
- references/evals-template.md 645 B
- references/evals.md 2.5 KB
- references/plugin-eval.md 1.5 KB
- references/skillopt-and-research.md 1.5 KB
- references/sources.md 1.7 KB
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.
- 9d ago First seen · 203 lines · 61 tokens per session scan A 95275b22371b
skill-eval-improve is a skill published in the GitHub repository Arenukvern/skill_steward (11 stars, last pushed 24d ago), licensed MIT. It adds 61 tokens to every session and 2,411 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it A with 0 findings. It is 100% identical to skill-eval-improve, differing in 0 lines, and is treated as a copy.
Other skills, from other repositories
testing-patterns
Plan, write, and maintain automated tests in any language, with worked examples in Python and JavaScript/TypeScript. Use this skill whenever the user wants to write or plan tests, apply TDD (red-green-refactor), structure unit tests with arrange-act-assert and given-when-then naming, choose between fakes, stubs, and…
bruno-api-testing
Create, run, and maintain API test collections using Bruno (OpenCollection YAML format and legacy Bru format). Use when the user wants to: (1) create a Bruno API test collection from scratch or from OpenAPI/Swagger specs, (2) write API request files with tests and assertions, (3) run API tests using bru CLI, (4)…
openspec-verify-change
Verify implementation matches change artifacts. Use when the user wants to validate that implementation is complete, correct, and coherent before archiving.
zsh-script-writer
Write and test Zsh scripts following project coding standards with ShellSpec BDD testing. Use when the user wants to: (1) create a new Zsh script, (2) write or fix ShellSpec tests for Zsh scripts, (3) review Zsh code for best practices, (4) add error handling or dependency checking to shell scripts, (5) implement API…
Test Engineer
Creates or completes a medium-coverage test suite: unit, component, and critical e2e flows.
convex-verify
Prove a Convex feature works — seed, drive as multiple mocked users via convex-test, assert behavior including the negative authz cases (wrong user refused, data-scope enforced).