reviewing-and-validating-project-changes

Validates project changes against contracts, diffs, CI evidence, and regression checks.

Updated Jul 9, 2026
One-click install
npx skills add https://github.com/alsdmlals4-eng/Base --skill reviewing-and-validating-project-changes-alsdmlals4-eng
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: reviewing-and-validating-project-changes
Source: https://github.com/alsdmlals4-eng/Base/tree/main/skills/reviewing-and-validating-project-changes
Command: npx skills add https://github.com/alsdmlals4-eng/Base --skill reviewing-and-validating-project-changes-alsdmlals4-eng

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) components.

What problem does it solve? It prevents unverified completion claims by requiring that every project change—whether made by a human or an AI agent—is checked against the approved work contract, the actual diff, executable test results, and post-merge repository state before being accepted. ## Core Features & Use Cases - Claim and Intent Verification: Links material claims and approved acceptance criteria to actual diffs, exact HEAD execution results, and post-merge readbacks using a structured evidence record and JSON schemas. - Multi-Lens Review Modes: Covers contract checks, static and runtime validation, accessibility review, performance profiling, CI cost optimization, reference freshness, and regression testing. - Fail-Closed Decisions: Produces ACCEPT, REVISE, REJECT, or UNVERIFIED verdicts and never promotes file existence, test definitions, or producer self-reports into pass evidence. - Use Case: Before merging a pull request, run the claim-and-intent-verification mode with a review evidence record to confirm each acceptance criterion maps to real implementation paths and executed checks at the exact HEAD commit. ## Quick Start Ask the AI to review the current pull request against its approved work contract and produce a validation verdict with evidence for every acceptance criterion.

Frequently Asked Questions about reviewing-and-validating-project-changes

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I verify that an AI agent's completion claim is true?

Use the claim-and-intent-verification mode to atomize material claims into a ledger, then bind each claim to the actual diff, exact HEAD execution results, and post-merge main readback. Claims without direct evidence stay CLAIM_UNVERIFIED or IMPLEMENTATION_UNVERIFIED.

How do I run the review evidence checker?

Copy the REVIEW_EVIDENCE_RECORD.json template, fill in claims, acceptance criteria, and checks, then run tools/check_review_evidence.py with --record, --base-ref, --execute-checks, and --output. The scripts/verify_evidence.py entrypoint provides compatibility for existing callers.

When is local CI fallback allowed instead of GitHub Actions?

Local fallback is allowed only when no canonical remote workflow run, ci-gate check run, or existing commit status exists for the exact head, the worktree is clean, the base is current, and the change is locally reproducible. Code, engine, or CI toolchain changes are blocked without an equivalent local contract.

Does this validation cover accessibility and performance?

Yes, dedicated modes audit accessibility barriers across text, input, UI, time limits, and motion, and compare performance against frame-time budgets on target hardware. Neither is forced for trivial changes; they apply when gameplay, UI, rendering, or platform builds are affected.

Why does a change stay UNVERIFIED even when tests pass?

A passing test at a different SHA, a queued CI run, or a producer's self-report is not direct evidence. The gate requires execution at the exact HEAD, mapped acceptance criteria, and post-merge readback before any completion verdict is issued.