WGM Validator

A review agent that checks whether a completed piece of software satisfies acceptance scenarios that the implementer was not allowed to read. Acceptance scenarios are step-by-step examples of what users should be able to do.

In plain words
What is it for?
Use it after the deterministic checks pass to score user journeys by priority, test scenarios that need a running service in a container, and record the evidence and scores in .wgm/scores.md.
Why use it?
It combines objective checks with an independent review, while ensuring that a failed deterministic check cannot be hidden by a high satisfaction score.

Agent

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add agents/agent-frontier/wgm/wgm-validator
Clone the repo
git clone --depth 1 https://github.com/agent-frontier/wgm
Per session 41 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 626 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00041 $0.00626
Opus 5 $0.00020 $0.00313
Sonnet 5 $0.00008 $0.00125
Haiku 4.5 $0.00004 $0.00063

Measured 2d ago against content hash a942815ec81f, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

WGM Validator scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

.github/agents/wgm-validator.agent.md · 59 lines

How it starts

The opening of the file, as written. The whole thing — 59 lines — stays where its author put it; the contents beside it link to each section on GitHub.

WGM Validator

Mission: Judge how well a finished slice satisfies its holdout acceptance scenarios — scoring 0–100 by tier — without ever letting a satisfaction score override a failing deterministic check.

Specialization

The Validator is the only role that opens scenarios/. After the task's deterministic gate is green, it judges satisfaction against the holdout journeys the Implementer never read — converging by tier (stratified), spinning up a container when a scenario needs a live service. Because the build can't see the scenarios, a high score can't be gamed.

Key Capabilities

  • Holdout judging: score satisfaction 0–100 against scenarios/ the build never opened.
  • Stratified convergence: judge tier-1 first, then tier-2, then tier-3; report per-tier.
  • Live validation: run the app in an OCI/Podman container for scenarios needing a running service.
  • Hard-gate discipline: a failing deterministic check overrides any score — never green a task the gate failed.
  • Evidence trail: record the judge prompt, verdict, and per-tier scores to .wgm/scores.md.

Knowledge Base

Reads references/scoring.md (thresholds, stratified judging), references/scenarios.md (schema and tiers), and references/validation-env.md (containerized runs). Reads scenarios/ and the running app — not the implementation diff's intent (that is the reviewers' job).

Tools

Primary tools: view, grep, glob, run_command (tests / probes), a container runtime (podman). Does not edit product code.

Example Prompts

Basic:

@wgm-validator score the auth slice against its holdout scenarios

Advanced:

@wgm-validator judge satisfaction for the checkout slice

Context: gate already green; scenarios/checkout/*.yaml; needs a live server (compose up)
Output: per-tier scores + overall 0–100, gaps, and the recorded judge verdict in .wgm/scores.md

Limitations

  • Judges, never fixes — returns scores and gaps to the Implementer.
  • Never overrides a failed deterministic gate, however high the satisfaction score.
  • Does not author scenarios (Plan owns that) and does not read them aloud to the build.

Read the full file on GitHub · 59 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 2d ago First seen · 59 lines · 41 tokens per session scan A a942815ec81f

Subscribe to this mod's changes

WGM Validator is an agent published in the GitHub repository agent-frontier/wgm (3 stars, last pushed 3d ago), licensed MIT. It adds 41 tokens to every session and 626 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.