FORGE-v0.6.0 — Phase I: Governed AI Engineering Foundation

FORGE v0.6.0 is the first public Phase I release of Framework for Orchestrated Reasoning, Governance & Engineering.

FORGE is a credit-aware engineering control layer for AI-assisted software development. It is designed to reduce repeated repository rediscovery and move deterministic preparation out of the model loop, so AI coding clients can spend more of their reasoning budget on judgment rather than reconstructing facts that local tooling can establish first.

Intent → Context → Reasoning → Evidence → Ship

What ships in Phase I

Deterministic AI preparation

FORGE can inspect a repository, select bounded context, classify task risk, discover relevant verification commands, and prepare a structured work packet before an AI coding client begins deeper reasoning. The aim is not to replace the model; it is to give the model a cleaner, more deliberate starting point.

FORGE also caches reusable preparation so repeated work can avoid unnecessary rediscovery when the underlying context has not changed.

Privacy-preserving efficiency metrics

forge metrics and forge metrics --json expose local telemetry for understanding how FORGE is preparing work. Current metrics include preparation events, cache-hit rate, preparation latency, deterministic steps, context considered versus selected, estimated context tokens avoided, and model-tier routing information.

The metric log is intentionally privacy-preserving. It does not store task descriptions, source code, filenames, repository identity, Git branch or commit information, or secret-like values. Token figures are estimates for engineering feedback; they are not provider billing data and are not a promise of a fixed percentage reduction in AI usage.

Risk-aware planning and resumable runs

FORGE can produce a deterministic plan for a task:

forge plan --task "Add passwordless sign-in"

Governed work can also be started and resumed through persisted run state:

forge run start --task "Add passwordless sign-in"
forge run status

This creates a clearer boundary between assumptions, planned work, evidence, approvals, and completion state instead of treating a long AI conversation as the only record of execution.

Portable project layer

FORGE can install a portable .forge/ operating layer into a project:

forge init --dry-run
forge init
forge doctor
forge validate

The project layer contains workflows, policies, context, intelligence modules, validation contracts, and engineering guardrails that can travel with the repository.

AI coding-client integration

FORGE includes a provider-neutral MCP server:

forge mcp serve

The current server exposes read-oriented preparation and status capabilities including forge_prepare, forge_plan, forge_context, forge_run_status, and forge_metrics. A Codex plugin packages the same governed preparation approach for supported AI coding workflows.

Governance and safety boundaries

Phase I establishes explicit engineering boundaries around AI-assisted changes. FORGE excludes secret-like paths from prepared context, keeps selected context bounded, separates project rules from task context, supports approval requirements for sensitive work, persists progress and evidence, and keeps verification distinct from assumptions and instructions.

FORGE does not silently execute arbitrary application commands simply because an AI model proposed them.

Verified release distribution

The v0.6.0 release pipeline is designed to fail closed. Before GitHub publishes the release, it verifies package, lockfile, and CLI version consistency; requires an annotated tag whose signature GitHub verifies; confirms the tagged commit is already part of main; runs type checking and the full test suite; builds and smoke-tests the packed CLI; generates and verifies SHA-256 checksums; and creates GitHub/Sigstore-backed provenance attestations for the release assets.

Linux and macOS:

curl -fsSLO https://github.com/gODtECH-Ctl-Create/gODtECH-FORGE/releases/download/v0.6.0/install.sh
sh install.sh 0.6.0
forge --version

Windows PowerShell:

Invoke-WebRequest https://github.com/gODtECH-Ctl-Create/gODtECH-FORGE/releases/download/v0.6.0/install.ps1 -OutFile install.ps1
.\install.ps1 -Version 0.6.0
forge --version

Both installers verify the release checksum before invoking the global npm installation.

Release provenance can be checked independently with GitHub CLI:

gh attestation verify forge-0.6.0.tgz \
  --repo gODtECH-Ctl-Create/gODtECH-FORGE \
  --source-ref refs/tags/v0.6.0

Phase I boundaries

This release does not claim a universal percentage reduction in AI tokens or credits. It does not ship native standalone binaries, automatic trusted execution of arbitrary project commands, dedicated integrations for every AI provider, or a production-complete autonomous software-development system. npm registry availability is also separate from this GitHub release and should not be assumed until it is explicitly published there.

Those boundaries are deliberate. Phase I is the foundation: make AI-assisted engineering more prepared, governed, measurable, repeatable, and auditable without pretending deterministic tooling can replace model judgment.

Why this release matters

AI coding systems are increasingly capable, but capability alone does not make an engineering process efficient, reproducible, or safe. FORGE establishes a different operating boundary:

Deterministic systems prepare facts; models spend reasoning on judgment.

v0.6.0 turns that principle into a working local CLI, planning and context engine, resumable workflow layer, MCP interface, governance framework, privacy-preserving efficiency telemetry, and verified release path.

License: Apache-2.0
Runtime: Node.js 20+
Release: v0.6.0
Status: Phase I foundation

Built with FORGE itself as the governing engineering workflow.