Skip to content

Qforge-chen/Q-forge

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

11 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Q-Forge

Q-Forge is a quality-focused agent system for manufacturing workflows, not a generic chatbot.

This public repository is no longer only a proof page. It is now the public proof layer plus the migration blueprint for a local quality-specific OpenClaw runtime.

In roughly 24 hours, Q-Forge moved from an already-proven Goose + MCP setup into a local OpenClaw quality edition with Android mobile proof.

24-Hour Migration Result

Within roughly 24 hours, the four Q-Forge capabilities that had already been proven on Goose + MCP were migrated into a local OpenClaw quality runtime and a hardened Android chat client:

  • 8D
  • RCA
  • Supplier
  • Reporter

The result is not a generic OpenClaw setup with extra prompts. It is a local quality runtime profile with:

  • deterministic tool paths
  • evidence-locked workflows
  • local artifact generation
  • mobile proof on Android

What changed in one fast migration pass

  • the original Goose + MCP proof stack was not discarded; it was migrated
  • the four working capability paths were re-established in a local OpenClaw quality runtime
  • Android mobile proof was completed for Supplier, 8D, and RCA
  • the runtime was narrowed into a quality-specific profile instead of a generic assistant
  • a public framework snapshot was added so other builders can follow the logic, not just the screenshots

What this means

This repository now shows one concrete thing:

Q-Forge was already real before this pass, and after one focused 24-hour migration sprint it became a local OpenClaw quality edition with working mobile proof.

Direct Running Proof

This repository contains direct running proof, not only design notes.

  • the workflows shown here were executed end to end
  • the proof set includes real local HTML artifacts
  • the proof set includes real Android chat screenshots
  • the public examples, rendered outputs, and mobile traces all come from actual runs

8D DOCX review demo

Open the preview image to watch the 40-second 8D demo from the earlier Goose + MCP phase.

Proof At A Glance

Supplier mobile proof 8D mobile proof RCA mobile proof
Supplier mobile proof 8D mobile proof RCA mobile proof
Validation stamp 8D rendered report preview RCA rendered report preview
Validation lock 8D report artifact RCA report artifact

What Is Public Here

  • real public skill packages under skills/
  • sanitized inputs, outputs, and rendered artifacts under examples/ and docs/mobile-proof/
  • public QMS Lite design and migration documents under docs/
  • a minimal framework snapshot under framework/ so other builders can follow the architecture without needing the full private runtime

Already Working Now

Capability Current working path Public proof
8D strict 8D audit path, deterministic D3-D7 checks, local HTML artifact, Android-triggered proof good 8D HTML, bad 8D HTML, good chat screenshot, bad chat screenshot
RCA detective-style branch-pruning chat, strongest cause-path conclusion, local HTML artifact, Android-triggered proof RCA HTML, start screenshot, pruning screenshot, conclusion screenshot
Supplier deterministic spreadsheet analysis, validation lock, local HTML artifact, Android-triggered proof supplier HTML, supplier screenshot
Reporter deterministic markdown-to-HTML renderer used by all three flows above 8D preview, RCA preview

Why Not Generic OpenClaw + Skills

Generic OpenClaw plus skills can add knowledge, but it does not automatically solve the runtime problem of quality work.

Quality-specific deployment needs:

  • local data handling instead of casual cloud routing
  • a reduced-permission mobile channel
  • deterministic tool boundaries
  • separate stable knowledge and daily work inputs
  • logic-locked review paths instead of free-form model output
  • rendered artifacts that can be reviewed and audited later

That is why QMS Lite is a runtime profile, not only a prompt pack.

What Was Customized

OpenClaw local quality edition

  • OpenClaw was reduced into a QMS Lite runtime profile
  • the local gateway was kept, but the work was narrowed to quality workflows
  • qm-review and qm-rca were turned into skill-driven agents instead of generic assistants
  • 8D, RCA, Supplier, and Reporter were wired into deterministic local tool paths
  • local data was split into:
    • knowledge-base
    • work-inbox
    • artifacts
    • audit

Android secure chat edition

  • the Android client was reduced from a general assistant into a secure chat terminal
  • Firebase and update checks were removed
  • cleartext traffic was disabled
  • FLAG_SECURE and gateway-only behavior were enforced during the hardening pass
  • the phone proof shown here was produced against the local OpenClaw QMS Lite gateway, not against a cloud runtime

Follow The Build

Public design docs

Public framework snapshot

Earlier Public Examples

These remain useful as the original Goose + MCP proof layer:

Builder Update

  • Q-Forge had already been proven on Goose + MCP before this migration pass started.
  • The practical thesis was already validated: large models plus deterministic rules can produce useful quality outcomes.
  • The current direction is now clearer: a local OpenClaw quality runtime plus Q-Forge-style skills is a better long-term base than staying on the old runtime.
  • This repository now shows both:
    • what had already worked
    • how that working logic was migrated into a local OpenClaw quality edition

Repository Layout

  • skills/: public skill packages
  • examples/: sanitized source examples from the earlier public proof layer
  • assets/: demo media and report previews
  • docs/: public migration, architecture, and proof documents
  • framework/: minimal public code and config skeleton for QMS Lite

Scope Note

  • Included here:
    • proof artifacts
    • migration explanation
    • framework skeleton
    • public skill packages
  • Intentionally excluded:
    • private project memory
    • private runtime configs and secrets
    • full local fork implementation
    • real customer data
    • full Android private implementation details

Contact

License

Apache License 2.0. See LICENSE.

About

The Local AI Auditor for Manufacturing. Built on Goose & MCP. Zero Hallucination Deterministic Logic Gates.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors