Make your AI 10^79× stronger

Why this works

ProjectGate is an AI workflow gate system that turns unreliable AI output into evidence-backed execution.

It forces AI-assisted work through facts, SOPs, known bug rules, runtime gates, and reviewable TaskRun evidence.

Current alpha: v0.4.6

Start here

Background and problem statement: BACKGROUND.md / BACKGROUND_CN.md.

Read QUICKSTART.md first.

中文文档：QUICKSTART_CN.md。

For copy-paste commands, see COMMANDS.md.

中文命令速查：COMMANDS_CN.md。

What it gives you

• Core governance rules
• Project Pack compiler
• Codex adapter
• Generic documentation template
• Example docs inbox
• One installer script
• Low / medium / high profile syntax: -p L, -p M, -p H

What it does not promise yet

• It is not a complete hard-gate system for every AI tool.
• It does not train a model.
• It does not automatically promote rules without owner review.
• It does not modify your project unless you explicitly authorize adapter installation steps.

Quick install

Initialize a workspace only:

python install_projectgate_alpha.py --dry-run --workspace-root "D:\ProjectGate"
python install_projectgate_alpha.py --install --workspace-root "D:\ProjectGate"

Compile a project pack from docs:

python install_projectgate_alpha.py --install `
  --workspace-root "D:\ProjectGate" `
  --project-name "MyProject" `
  --project-root "D:\MyProject" `
  --docs-dir "D:\MyProjectDocs" `
  --build-pack

Build and install a Codex pack:

python install_projectgate_alpha.py --install `
  --workspace-root "D:\ProjectGate" `
  --project-name "MyProject" `
  --project-root "D:\MyProject" `
  --docs-dir "D:\MyProjectDocs" `
  --build-pack `
  --build-codex-pack `
  --install-codex-skill

Use in Codex:

/goal $projectgate -p L: Review this issue and produce an action queue. Do not modify project files.

ProjectGate Runtime v0.2

Alpha v0.2.0 adds Runtime Gate support. SOPs and KnownBugRules are no longer passive folders only. Real workflows must create a TaskRun.json, load active SOPs / KnownBugRules into the task, pass stage gates, use REPAIR_AND_RECHECK on gate failure, and pass a delivery check before final delivery.

ProjectGate Knowledge Router v0.3

Alpha v0.3.0 adds knowledge routing and learning loops. L / M / H now controls SOP / KnownBugRule selection scope, not just cost. Successful reusable TaskRuns can create SOP candidates. Failures can create KnownBugRule candidates. Candidates require owner approval before becoming active.

ProjectGate Auto Capture v0.4.5

Alpha v0.4.5 adds automatic learning-loop capture. Stage gate / delivery check failures automatically create incidents and KnownBugRule candidates. Successful delivery checks automatically create SOP candidates. The owner only approves or rejects candidates instead of manually writing them. If a similar rule already exists, ProjectGate records whether the active rule was not selected, selected but not enforced, or too coarse.

Operator Workflow v0.4.5

• Added candidate lifecycle tooling for listing, showing, approving, rejecting, and merging SOP / KnownBugRule candidates.
• Added Project Pack manager for pack info, validation, export, and controlled install.
• Added projectgate_cli.py as a unified command wrapper for start, stage, delivery, observe, continuity, candidates, pack, status, and controlled command execution.
• Added root pg.py and pg.bat entrypoints for shorter local commands.
• Added pg exec so ProjectGate-controlled commands can automatically write execution logs and run observation gate.

v0.4.5 Safety Hardening

ProjectGate v0.4.5 clarifies the current enforcement boundary and hardens installer behavior.

ProjectGate is an AI workflow evidence layer for ProjectGate-controlled flows. It is not yet an OS-level sandbox or enterprise compliance platform.

Safety changes:

• ProjectGate-managed directory replacements now use .projectgate-managed.json markers.
• Installers refuse to overwrite unmarked non-ProjectGate directories.
• Existing managed directories are backed up before replacement.
• Workspace-root and target-depth checks reduce accidental deletion risk.
• SECURITY.md documents the current enforcement boundary and Alpha limitations.

v0.4.5 Packaging and Test Foundation

ProjectGate v0.4.5 adds a Python package entry point and automated test foundation.

New files:

• pyproject.toml
• projectgate/cli.py
• projectgate/__init__.py
• tests/test_projectgate_package_and_runtime.py

pg.py is now a compatibility shim. The stable entry point is routed through projectgate.cli:main.

The test foundation covers:

• package / pg entry points
• installer safety refusal
• managed replacement backup
• candidate lifecycle
• runtime start / stage / delivery
• observation gate
• pg exec

Run:

python -m unittest discover -s tests

v0.4.5 Philosophy Docs

ProjectGate v0.4.5 adds philosophy and implementation-map documents.

New docs:

• docs/WHY_GATES_MATTER.md
• docs/CUBE_MODEL.md
• docs/IMPLEMENTATION_MAP.md

These documents explain why gates matter as search-space reduction, present the Cube Model without tying it to a specific private project, and map the philosophy to current ProjectGate implementation files.

v0.4.5 Headline

ProjectGate v0.4.5 surfaces the Cube Model headline and direct evidence link at the top of the README.

v0.4.5 Public Surface Sync

ProjectGate v0.4.5 aligns the public homepage with the current Cube Model positioning.

• The README now opens directly with Make your AI 10^79× stronger.
• The Cube Model explanation is linked immediately under the headline.
• The first screen now describes ProjectGate as an AI workflow gate system for evidence-backed execution.

v0.4.6 Documentation Hygiene

ProjectGate v0.4.6 removes private local paths from public documentation.

• Quick Start now points first-time users to the public GitHub repository.
• The Codex adapter README now uses repository-relative paths instead of local machine paths.
• No runtime behavior changes.