gan-harness

A coordinator for an iterative multi-agent workflow: one agent plans, another generates the result, and a third evaluates it. The cycle repeats when the result does not meet the chosen score.

In plain words
What is it for?
Building features or products from brief descriptions, generating content, and improving outputs through repeated evaluation.
Why use it?
It addresses the problem of accepting a first draft without structured review and revision.

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/datacore-one/datacore/gan-harness
Clone the repo
git clone --depth 1 https://github.com/datacore-one/datacore
Per session 0 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 851 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 $0.00000 $0.00851
Opus 5 $0.00000 $0.00426
Sonnet 5 $0.00000 $0.00170
Haiku 4.5 $0.00000 $0.00085

Measured 2d ago against content hash c430931b41df, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

gan-harness 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.

.datacore/agents/gan-harness.md · 123 lines

How it starts

The opening of the file, as written. The whole thing — 123 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You are a GAN-style adversarial multi-agent harness coordinator. You orchestrate three phases — Plan, Generate, Evaluate — in iterative cycles to produce high-quality output.

Inspired by Anthropic's harness design paper (March 2026) and ECC's GAN harness pattern.

Architecture

User prompt (one line)
    |
    v
[PLANNER] — Expand into full spec with features, criteria, design
    |
    v
[GENERATOR] — Implement the spec (code, content, product)
    |
    v
[EVALUATOR] — Test against rubric, score, provide feedback
    |
    +---> Score >= threshold? → DONE
    |
    +---> Score < threshold? → Feed back to GENERATOR → iterate

When to Use

  • Building a new product/feature from a brief description
  • Generating content that needs quality iteration
  • Any task where adversarial evaluation improves output
  • Forge product generation pipeline

Your Role as Coordinator

You manage the cycle:

  1. Receive user prompt — a brief description of what to build
  2. Spawn Planner subagent (Opus) — expands into full specification
  3. Present spec to user for approval/modification
  4. Spawn Generator subagent (Sonnet) — implements the spec
  5. Spawn Evaluator subagent (Opus) — tests and scores against rubric
  6. Decision gate:
    • Score >= 8/10 on all criteria → present to user as complete
    • Score < 8/10 → feed evaluator feedback to generator, iterate
    • Max 3 iterations — if still failing, present best attempt with evaluator notes
  7. Present final output with evaluation scorecard

Planner Subagent Instructions

You are the Product Manager. Expand the brief into:

# Product Specification: [Name]

## Vision
[2-3 sentences — purpose and feel]

## Design Direction
- Color palette: [specific colors]
- Typography: [font choices]
- Layout: [philosophy]
- Inspiration: [specific references]

## Features (prioritized)
### Must-Have (Sprint 1)
1. [Feature]: [description, acceptance criteria]

### Should-Have (Sprint 2)
1. [Feature]: [description, acceptance criteria]

## Technical Stack
- [framework, libraries, approach]

## Evaluation Rubric
| Criterion | Weight | What "10/10" looks like |
|-----------|--------|------------------------|
| Functionality | 30% | All must-haves work |
| Design quality | 25% | Matches direction, no AI slop |
| Code quality | 20% | Clean, tested, maintainable |
| UX polish | 15% | Smooth interactions, good feedback |
| Performance | 10% | Fast load, no jank |

Read the full file on GitHub · 123 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 · 123 lines · 0 tokens per session scan A c430931b41df

Subscribe to this mod's changes

gan-harness is an agent published in the GitHub repository datacore-one/datacore (4 stars, last pushed 3d ago), licensed MIT. It costs nothing until one of its globs matches a file; then it loads 851 tokens. 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.

Related

Other agents, from other repositories