Validation Gating — Development skill for Claude Code
Turn proposed harness changes into candidate-first changes with explicit replay, regression, and rollback expectations.
How to install Validation Gating
Installs to ~/.claude/skills/ilearn-lab-evoharness-validation-gating/SKILL.md
mkdir -p ~/.claude/skills/ilearn-lab-evoharness-validation-gating && curl -fsSL https://raw.githubusercontent.com/iLearn-Lab/EvoHarness/HEAD/.claude/skills/validation-gating.md -o ~/.claude/skills/ilearn-lab-evoharness-validation-gating/SKILL.md Restart Claude Code, or start a new session, for it to be picked up.
What Validation Gating does
name: validation-gating description: Turn proposed harness changes into candidate-first changes with explicit replay, regression, and rollback expectations.
Validation Gating
Use this skill when a change affects commands, skills, or self-evolution flow.
- candidate first, promotion second
- record the smallest replay step that proves the original failure or success pattern
- record the narrowest regression command that guards the touched behavior
- keep rollback possible until the n
Alternatives in Development
- Harness Improvement — Update Kandev harness files from session learnings or explicit cross-platform harness requests 697 ★
- Stop Review Gate — You are reviewing uncommitted changes before the author finishes their turn 664 ★
- Debug User Tx — Replay a user's failing transaction on forked state and map the error to source 105 ★
Full documentation available on GitHub
View Source RepositoryRelated Skills
Run Evals
Replay the regression corpus inside THIS interactive session — no headless, no API key (ADR 0003). Required be
Update Harness
Apply an approved Harness Improvement Plan (HIP) to update instructions, hooks, skills, and configuration file
Debate Replay
Deterministically replay a past debate from a session snapshot — regression tests the deliberation engine
Think Research Framework
Research a thinking framework and produce its evidence-graded dossier plus a schema-valid proposed registry en
Arnold
General-purpose planning and execution harness for LLMs — structured phases, critique, gating, and review
Architecture Read
Architecture reviewer step 1 — read the issue and understand proposed structural changes
Related Agents
Validation Guardian
Review whether a harness change is candidate-first, replayable, and rollback-safe
Petswipe Reviewer
Use proactively for code review, regression hunting, validation gaps, deployment impact analysis, and document
MCP Safety Reviewer
Reviews MCP tool designs for blast radius, three-mode design (read_only/draft_only/approved_execute), refusal