ai-forge-cli

Forge

Forge is a Python CLI and framework for V4 system design and delivery. It keeps C1/C2 architecture central, keeps C3 implementation architecture beside code, and gives skills narrow context from the merged model.

Forge is skills-first. The skills drive the work. The CLI exists to provide scoped context, validation, and audit artifacts to the active skill.

What This Repository Contains

• src/cli: the Forge CLI implementation
• skills/: the Forge skill source used by initialized repositories
• FRAMEWORK_V4.md: the V4 framework process used to seed forge/FRAMEWORK_V4.md
• SCHEMA_REFERENCE_V4.md: the V4 schema contract used to seed forge/SCHEMA_REFERENCE_V4.md
• examples/: example Forge repositories and audit artifacts

CLI Commands

• forge init: scaffold a new Forge workspace and explain how to use it
• forge crawl: extract the merged V4 model from central files and code annotations
• forge context: render scoped context for a system, flow, container, entity, component, operation, or data shape
• forge knowledge: list supporting Markdown knowledge docs attached to Forge refs such as containers, flows, entities, components, and operations
• forge update skills: refresh only forge/skills/ using conflict-safe manifests
• forge audit: generate a self-contained architecture audit dashboard

What Forge Is

• a framework for capturing architectural truth before implementation detail
• a skills-first workflow for moving from business discovery to a flow-complete delivery outcome
• a scoped context generator for build, review, and security tasks
• an audit artifact generator for human review
• a lightweight index for repo-native delivery knowledge such as runbooks, testing notes, security notes, operations notes, and domain glossaries

What Forge Is Not

• a generic project scaffolder for arbitrary app stacks
• a replacement for implementation skills such as build, review, and security
• a reason to model every payload, component, or environment detail up front
• a reason to widen scope beyond an approved, coherent and verifiable outcome

Recommended Workflow

1. Run forge init in an empty repository.
2. Read:
   • forge/USING_FORGE.md
   • forge/FRAMEWORK_V4.md
   • forge/SCHEMA_REFERENCE_V4.md
3. Use forge-business to create business-plan.md for new ideas.
4. Use forge-schema to define:
   • forge/system.yaml
   • forge/containers.yaml
   • forge/entities.yaml
   • forge/decisions.yaml
   • forge/crawler.yaml
5. Use forge-hydrate when an existing codebase needs to be reverse-engineered into Forge V4.
6. Use business actions to speculate cross-container runtime flows before settling containers.
7. Define a complete actor/system outcome, its acceptance conditions, and all material flow outcomes.
8. Add C3 annotations beside implementation code as the outcome is built.
9. Use forge crawl, forge context, and forge audit to validate the merged model.
10. Use forge-review and forge-security before build work.
11. Use forge-build to implement the approved outcome and forge-test to independently prove it.

The intended operating mode is:

1. choose the active skill first
2. ask that skill what scope it needs
3. use forge crawl to validate extraction
4. use forge context only for narrow scope
5. use forge audit when you need a review artifact

An initialized repository keeps the Forge-owned schema workspace under ./forge/. The repo root stays available for product code, app docs, and non-Forge tooling.

Golden Path Examples

• examples/forge_minimal_web_app: the V4 example used for docs, smoke tests, crawler extraction, context, and audit rendering

Local Development

Use the project virtual environment:

.venv/bin/python -m pip install -e .[dev]

Validation Commands

make lint
make typecheck
make test
make compile
make build
make verify-package
make smoke-init
make check

make verify-package builds the wheel, installs it into a fresh Python 3.11+ virtual environment, and confirms that the installed forge entrypoint works.

Packaging

Build artifacts are generated with:

.venv/bin/python -m build

This produces:

• dist/*.tar.gz
• dist/*.whl

Validate package metadata before release with:

make check-dist

Local Testing

Run the full local verification path:

make check

That covers linting, type checking, tests, compile validation, build output, distribution metadata checks, clean wheel install, and forge init smoke validation.

If you want to test the richer example schema directly:

EXAMPLE="examples/forge_minimal_web_app"
.venv/bin/forge audit --project-dir "$EXAMPLE" --output /tmp/forge-audit-example.html

That generates the audit artifact and opens it unless --no-open is supplied.

Framework Notes

• Forge is flow-complete: model broadly, then deliver one approved, meaningful outcome at a time.
• The framework should reduce context, not increase it.
• Components and exact schemas should only appear when runtime boundaries and business action intent are already clear.
• Important decisions should be captured in forge/decisions.yaml.

Maintainer Notes

• Release instructions live in docs/RELEASING.md.