How do I install the Krt Skill Arbiter skill?

Run npx skillmds@latest add elzawarudo/krt-skill-arbiter in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Krt Skill Arbiter safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Krt Skill Arbiter?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Krt Skill Arbiter free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Krt Skill Arbiter?

elzawarudo (@elzawarudo) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Krt Skill Arbiter by elzawarudo — Agent Skill

Name: Krt Skill Arbiter
Author: elzawarudo

Krt Skill Arbiter

Evaluate a KRT skill portfolio with deterministic corpus checks, supervisor-captured pass/fail/inconclusive scoring, and structural portfolio validation. Use when adding or changing skills, auditing routing and negative triggers, testing fallback/permission/restart/outcome behavior, or checking portfolio metadata and safety wiring. It does not execute models or grade its own behavior.

elzawarudo 2705975 11 files · 54.4 KB Updated Sep 17, 2026

File contents

KRT Skill Arbiter

Measure skill behavior without turning the evaluator into another opaque agent.

Load References

Load references/safety.md before handling prompts, outputs, traces, or tool logs.
Load references/evaluation-contract.md before preparing or scoring a run.
Treat references/portfolio.json as the explicit inventory and safety classification for every KRT skill.

Workflow

Run scripts/check_corpus.py against references/cases.json and references/expectations.json; preserve its corpus version and digest in the run result.
Give each routing case to the evaluated runtime without its expected skill. Give capability cases with their declared target skill.
Keep expected behavior hidden until the runtime response is complete.
Have a supervisor record exactly one pass, fail, or inconclusive result per observed case. Preserve evidence outside the status field.
Run scripts/score_run.py <results.json> to verify corpus identity and aggregate the captured judgments. Do not convert inconclusive results into passes or failures.
Run scripts/check_portfolio.py --repo-root <repo> after adding or modifying skills.

Use the bundled scripts only as deterministic validators and aggregators. They do not invoke models, execute case content, or decide whether a response is correct.

Output

Report:

corpus version and coverage;
pass, fail, and inconclusive counts;
conclusive pass rate;
failures grouped by evaluation category;
portfolio contract errors;
evaluator limitations and evidence location.

Do not claim improvement from one run alone. Compare like-for-like corpus versions and record runtime/model context outside the scored artifact.

---name: krt-skill-arbiterdescription: Evaluate a KRT skill portfolio with deterministic corpus checks, supervisor-captured pass/fail/inconclusive scoring, and structural portfolio validation. Use when adding or changing skills, auditing routing and negative triggers, testing fallback/permission/restart/outcome behavior, or checking portfolio metadata and safety wiring. It does not execute models or grade its own behavior.---# KRT Skill ArbiterMeasure skill behavior without turning the evaluator into another opaque agent.## Load References- Load `references/safety.md` before handling prompts, outputs, traces, or tool logs.- Load `references/evaluation-contract.md` before preparing or scoring a run.- Treat `references/portfolio.json` as the explicit inventory and safety classification for every KRT skill.## Workflow1. Run `scripts/check_corpus.py` against `references/cases.json` and `references/expectations.json`; preserve its corpus version and digest in the run result.2. Give each routing case to the evaluated runtime without its expected skill. Give capability cases with their declared target skill.3. Keep expected behavior hidden until the runtime response is complete.4. Have a supervisor record exactly one `pass`, `fail`, or `inconclusive` result per observed case. Preserve evidence outside the status field.5. Run `scripts/score_run.py <results.json>` to verify corpus identity and aggregate the captured judgments. Do not convert inconclusive results into passes or failures.6. Run `scripts/check_portfolio.py --repo-root <repo>` after adding or modifying skills.Use the bundled scripts only as deterministic validators and aggregators. They do not invoke models, execute case content, or decide whether a response is correct.## OutputReport:- corpus version and coverage;- pass, fail, and inconclusive counts;- conclusive pass rate;- failures grouped by evaluation category;- portfolio contract errors;- evaluator limitations and evidence location.Do not claim improvement from one run alone. Compare like-for-like corpus versions and record runtime/model context outside the scored artifact.

Frequently asked questions