skills.rest
ZaxbyHub
ZaxbyHubOFFICIAL
Explore all skills by this author →
GitHub

qa-sweep

Enforces parallel implementation, adversarial review, and completeness verification before task completion.

Updated Mar 30, 2026
Search Tags:#quality assurance#code review#debugging#verification#regression testing#sub-agents#adversarial review
One-click install
npx skills add https://github.com/ZaxbyHub/ragappv3 --skill qa-sweep-zaxbyhub
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: qa-sweep
Source: https://github.com/ZaxbyHub/ragappv3/tree/main/.opencode/skills/qa-sweep
Command: npx skills add https://github.com/ZaxbyHub/ragappv3 --skill qa-sweep-zaxbyhub

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Implementation and debugging tasks often ship with missed edge cases, unverified assumptions, and incomplete coverage because a single agent reviews its own work. This Skill enforces a structured quality protocol that catches errors before a task is declared done. ## Core Features & Use Cases - Parallel Implementation: Delegates independent units of work to parallel sub-agents, with local fallback passes when delegation is unavailable. - Independent Adversarial Review: Spawns a fresh reviewer agent instructed to distrust all prior work and trace every change end-to-end through the actual source code. - Completeness Verification: Maps every originally planned item to delivered code changes via a second independent agent. - Regression Baseline Testing: Requires each new regression test to fail against the documented pre-fix commit and pass against the fix branch. - Use Case: After fixing a bug in a RAG retrieval pipeline, run this protocol to have an independent reviewer trace the fix, verify the regression test fails on the pre-fix baseline, and confirm all planned items were delivered before stopping. ## Quick Start Apply the qa-sweep protocol to verify this bug fix with an independent adversarial review and regression baseline check before marking it complete.

Frequently Asked Questions about qa-sweep

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

FAQPage Schema
How do I verify a bug fix with independent code review?▼

Use an adversarial review pass where a fresh agent, uninvolved in the implementation, traces every change end-to-end through the actual source code. The reviewer assumes all prior work is incorrect until proven otherwise with direct code evidence.

How to confirm a regression test actually guards a bug fix?▼

Run the new test against the documented pre-fix commit or baseline hash to confirm it fails, then run it against the fix branch to confirm it passes. A test that never fails on the pre-fix baseline is not a real regression guard.

What happens when sub-agents are not authorized for delegation?▼

The protocol falls back to local passes: the same explorer, adversarial reviewer, and completeness checks run locally and are explicitly labeled as fallback passes. The review rigor and stop conditions remain unchanged.

When should I stop working on an implementation task?▼

Stop only when at least 95% confident that all issues and edge cases are covered, all reviewer findings are addressed, and delivered work matches the original plan. Below that threshold, state what remains and continue.

Does this protocol handle external PR review findings?▼

External PR review findings are treated as unverified claims until independently validated against the code. Accepted follow-ups that affect a PR update or publish flow are routed through the commit-pr process.

Related Skills

Explore More Software Engineering Skills→
Bbeierle12Bbeierle12
8

systematic-debugging

Apply a scientific-method approach to debug bugs and test failures.

Community
Basic
kamp-uskamp-us
25

ship-it

Automate squash merging of verified Pull Requests with green CI status.

Official
Advanced
zajcazajca
2

go-linting

Configure golangci-lint with recommended Go linters for static analysis.

Community
Basic
try-workstry-works
4

rlm-subagent

Coordinate parallel RLM sub-phase execution with sequential fallback and two-stage reviews.

Community
Advanced