Agent Regression Forge
SpeedRanger/agent-regression-forge
Converta falhas de agentes IA em testes de regressão e regras determinísticas
Ver no GitHubTransforma falhas de agentes IA em testes de regressão determinísticos, guardrails e prompts otimizados. Cria artefatos concretos (regression packs) a partir de erros reais de agentes de código, incluindo modos de falha prováveis, verificações determinísticas e sugestões de scaffolding.
Instalação
npx convez-ai SpeedRanger/agent-regression-forge1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Validar comportamento de agentes de código após falhas
- Criar testes de regressão baseados em erros reais
- Implementar guardrails para evitar comportamentos indesejados em próximas execuções
- Documentar e reproduzir falhas de agentes de forma estruturada
Veredito

Ferramenta especializada e bem-pensada para um problema real: falhas silent de agentes IA que aparentam sucesso. A abordagem de converter failures em regression packs é prática e dirigida, embora dirija-se a um nicho técnico maduro (equipes com agentes IA em produção).
README
A tradução para o português deste README ainda está na fila.
Agent Regression Forge
Agent Regression Forge turns a failed AI coding-agent run into deterministic regression checks, guardrails, and a next prompt.
Live target:
- Site: https://speedranger.github.io/agent-regression-forge/
- Repo: https://github.com/SpeedRanger/agent-regression-forge
- Release: https://github.com/SpeedRanger/agent-regression-forge/releases/tag/v0.2.0
- Public request form: https://github.com/SpeedRanger/agent-regression-forge/issues/new?template=regression_pack_request.yml
Why this exists
AI coding agents are good enough to create dangerous success-shaped work: changed files, confident summaries, skipped tests, broad fallback logic, and adjacent behavior drift. The missing layer is not another coding agent. It is a small proof forge that turns the failure into tests and rules the next run must obey.
Public product
index.html- static first-use app.github/ISSUE_TEMPLATE/regression_pack_request.yml- public-safe paid slot requestsample-regression-pack.md- example outputlaunch-day-buyer-packet.md- buyer-facing offer and proof boundaryreserve-slot.md- public-safe slot reservation pageproduct-hunt-launch-assets.md- Product Hunt launch copyproduct-hunt-submit-handoff-2026-06-08.md- submit fields and manual blockerlaunch-status.json- proof gateslaunch-ledger.md- public launch state and revenue slotsllms.txt- machine-readable product summarySECURITY.md- public safety boundary
Paid wedge
The first paid unit is a $19 public/redacted "failure to regression pack"
service.
The buyer sends one public or redacted agent failure and receives:
- likely failure modes
- deterministic regression checks
- test scaffold suggestions
- guardrail rules
- proof checklist
- next prompt
One paid slot clears the $10 target before fees only when receipt, checkout
export, paid invoice, or equivalent buyer-confirmed payment evidence exists.
The public request path currently uses a GitHub issue template. After a real checkout or private paid-intake URL exists, the deploy script can publish it with:
pwsh -NoProfile -File ui/agent-regression-forge/deploy-github-pages.ps1 `
-RequestUrl "PASTE_CHECKOUT_OR_PRIVATE_INTAKE_URL" `
-ConfirmPublish
Run locally
Open index.html directly or serve the folder:
python -m http.server 4193 -d ui/agent-regression-forge
Then open http://localhost:4193.
Hard limits
Agent Regression Forge does not guarantee Product Hunt placement, bug fixes, agent correctness, or revenue. It creates a concrete regression-pack artifact from public/redacted inputs.