SkillsHouse procedures / 05

Research craft, written as a procedure.

Five real, versioned local skills describe how CyberNative designs, audits, reviews, teaches, and gates research. They are house procedures—not an unrestricted marketplace.

The current set

Status: local source. The procedures exist; public catalog availability is not claimed.

Protocol

`design-ml-experiment`

Turn an AI/ML research question into a frozen, bounded, executable CyberNative capsule protocol for experiments, benchmarks, ablations, safety evaluations, replications, or substantial amendments before the main result is inspected.

Local source · skills/design-ml-experiment/SKILL.md

Claims

`audit-research-claims`

Audit factual claims, figures, citations, scope, limitations, and agent-use disclosures before editorial approval, after a correction report, or whenever wording may exceed its evidence.

Local source · skills/audit-research-claims/SKILL.md

Review

`review-reproduction`

Independently rerun and audit a capsule and its evidence bundle, compare raw artifacts and metrics, probe failure cases, and classify the result without treating same-harness repeat execution as independent reproduction.

Local source · skills/review-reproduction/SKILL.md

Learning

`author-learning-lab`

Turn a reviewed capsule into an accessible, hands-on human learning lab that stays linked to real evidence, includes deterministic reset and checks, and keeps execution within its safety boundary.

Local source · skills/author-learning-lab/SKILL.md

Release

`release-research-capsule`

Gate, package, sign, archive, and publish a reviewed research capsule and its derived surfaces only after execution, independent review, a claim audit, and one accountable human editor.

Local source · skills/release-research-capsule/SKILL.md