eval-engineering

Review, improve, and author MoltNet evals — one discipline across every use case (agent-runtime prompt evals in evals-v2/, practice/incident evals in evals/moltnet-practices/, and context-pack evals). Use when asked to "review an eval", "is this eval trustworthy", "improve/rewrite an eval scenario", "write an eval", "create an eval from an incident", "check the baseline", or "gap-test". Merges the LangChain eval-engineering process with MoltNet's battle-tested gap-test principles and the two-stage gate+judge harness.

Quality

94%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Passed

No findings from the security scan

Quality

Content

92%Weight 40%Scale 1-5

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

A well-structured, lean process skill that sequences a concrete eval-engineering workflow with explicit validation and feedback loops, and cleanly delegates detail to three real one-level-deep reference files. Its only weakness is that some executable artifact specifics live in references rather than inline.

Suggestions

Inline a minimal annotated eval.json/rubric.json/gates.json skeleton (even 3-4 fields each) so an author can start without first opening scenario-format.md.

Add one short worked example inline (e.g., the dbos-after-commit case summarized in 2-3 lines) to ground the abstract method steps in a concrete artifact.

Dimension	Reasoning	Score
Conciseness	The body is lean and assumes Claude's competence; it defines only MoltNet-specific framing (gates vs judge, gap-test) rather than concepts Claude already knows, and every section and list item earns its place without padding.	5 / 5
Actionability	Concrete, executable guidance is present (named tools run_eval/judge_eval_attempt, `moltnet eval run --pack`, thresholds like ≥40%/≥2/≥4, a 20% gold-standard example), but the exact field schemas for eval.json/rubric.json/gates.json are deferred to references, leaving minor gaps versus fully copy-paste-ready inline detail.	4 / 5
Workflow Clarity	The Method is a clearly sequenced 7-step process with explicit validation (the two-stage gate failure → composite 0 / judge-skipped rule, baseline measurement of ≥2/≥4 runs) and a feedback loop in step 7 (audit, then re-run on non-capability failures), satisfying the destructive/batch validation requirement.	5 / 5
Progressive Disclosure	The body is an overview that points to three well-signaled, one-level-deep references (scenario-format.md, review-checklist.md, incident-to-eval.md), each introduced in context; all three referenced files exist in references/ and none are nested deeper.	5 / 5
	Total	19 / 20 Passed

Description

96%Weight 40%Scale 1-5

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

A dense, information-rich description that explicitly covers what the skill does, when to invoke it via numerous natural trigger phrases, and a distinct MoltNet eval-engineering niche. It is comprehensive and third-person, with only minor overlap risk on the generic "review" trigger.

Dimension	Reasoning	Score
Specificity	"Review, improve, and author MoltNet evals" lists multiple concrete verbs, supplemented by create-from-incident, check-baseline, and gap-test actions across three named corpora, giving comprehensive coverage rather than the minor gaps of a 4.	5 / 5
Completeness	It explicitly answers both what ("Review, improve, and author MoltNet evals") and when ("Use when asked to..." with concrete trigger phrases), matching the anchor for clearly and explicitly answering both.	5 / 5
Trigger Term Quality	Seven natural trigger phrases are quoted verbatim ("review an eval", "is this eval trustworthy", "improve/rewrite an eval scenario", "write an eval", "create an eval from an incident", "check the baseline", "gap-test"), covering synonyms and common variations a user would actually say.	5 / 5
Distinctiveness Conflict Risk	The MoltNet-anchored niche and gap-test/baseline triggers are mostly distinct with minimal conflict risk, but the phrase "review an eval" has minor overlap with a generic review skill, so it sits just below the clear-niche anchor of 5.	4 / 5
	Total	19 / 20 Passed

Validation

100%

Warnings & errors only

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation — 16 / 16 Passed

Validation for skill structure

No warnings or errors.

Repository: getlarge/themoltnet
Path: .agents/skills/eval-engineering/SKILL.md
Commit: 5daa9ca

Reviewed: ~14 hours ago

Table of Contents

Discovery Implementation Validation

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.