OPEN THE VALIDATIONFORGE SITEOPEN VALIDATIONFORGE SITE

PLUGIN·beta

Functional validation for agents — evidence over mocks.

Ship functional evidence, not passing tests. Every completion claim produces screenshots, API responses, and build logs that a skeptical reviewer can open.

Install the plugin ↗INSTALL ↓READ THE LAUNCH POST →

withagents.dev/validationforge

Read the launch post →

Docs When Readers Aren't Human

2026-05-19 · 16 min read

642

TEST FILES BLOCKED

VALIDATION PHASES

SKILLS SHIPPED

MOCKS TOLERATED

“A successful build only proves it compiled. A passing test confirms the mock behaved correctly. ValidationForge proves the real system does the real thing.”

ValidationForge is a Claude Code plugin that replaces unit-test theater with functional validation against the real running system. Every completion claim produces evidence — screenshots, API responses, build logs — that a skeptical reviewer can open.

Ships as 52 validation skills, 19 slash commands, 7 enforcement hooks, 7 agents, and 9 rules files. The block-test-files hook refuses to create test files. The validation-not-compilation hook rejects completion claims that cite only a successful build. The evidence-gate-reminder hook injects an evidence checklist before any task is marked done.

01. ARCHITECTURE

How it's built

MODULE 00

VALIDATE

Seven-phase pipeline from preflight to verdict. Every journey captures real evidence — screenshots, API responses, build logs — before a gate opens.

MODULE 01

CONSENSUS

N independent validators each run the full journey list. The synthesizer votes per-criterion. Split verdicts escalate to human rather than downgrade silently.

MODULE 02

FORGE

Autonomous fix-and-revalidate loop with a 3-strike cap per journey. Each attempt targets a different root cause. Rollback on exhaustion.

02. FEATURES

No Mocks, Ever

Enforcement hooks block creation of .test.*, mock libraries, and in-memory test doubles. Fix the real system instead.

Evidence Chain of Custody

Every validation step captures screenshots, API bodies, and log output to disk. Missing files = FAIL.

Three Engines

VALIDATE (beta) runs single-pass validation. CONSENSUS (V1.5) votes contested verdicts. FORGE (V2.0) runs autonomous fix loops with rollback.

Dependency-Aware Waves

DB → API → Web/iOS. Downstream validators never run against a failing upstream.

03. QUICK START

// install

# install via Claude Code marketplace
claude plugin marketplace add krzemienski/validationforge
claude plugin install validationforge@validationforge

# plan a validation sweep
/validate-plan user-signup-flow

# execute with evidence capture
/validate-sweep
# → screenshots to e2e-evidence/web/
# → API responses to e2e-evidence/api/

# render the evidence dashboard + per-journey verdicts
/validate-dashboard

04. INSTALL

01claude plugin marketplace add krzemienski/validationforge
02claude plugin install validationforge@validationforge
03Run /vf-setup to initialize configuration + preflight checks
04Run /validate-plan <journey-name> to define PASS criteria + evidence requirements
05Run /validate-sweep to execute against the real system and capture evidence
06Run /validate-dashboard to render the final evidence-backed verdict

05. FEATURE MATRIX

FEATURE

STATUS / NOTE

VALIDATE engine

OKBeta — 7-phase pipeline

CONSENSUS engine

PARTIALV1.5 — multi-validator vote

FORGE engine

PARTIALV2.0 — autonomous fix loops

Evidence chain of custody

OKScreenshots + API bodies + logs

Block-test-files hook

OKPreToolUse exit 2

Dependency-aware waves

OKDB → API → Web/iOS

06. DOCS

Documentation

Read the VF docs

vf.withagents.dev

06. RESOURCES

GitHub

Source code

krzemienski/validationforge

v1.0.02026-04-21Stable VALIDATE engine, 52 skills, 19 commands, 7 hooks shipped.
v0.9.02026-03-28Dependency-aware wave scheduling for multi-platform projects.
v0.8.02026-03-12Evidence dashboard + per-journey verdict synthesis.