STAGE 0.6.0

Released: 2026-07-23

The plugin package version is 0.6.0. SPEC.md is method version 0.6.

Release Decision

STAGE 0.6 makes the public method smaller. The Safety Kernel contains eight normative obligations that protect human intent, authority, ownership, the real project path, truthful evidence, honest judgment, reversibility, and handoff. Everything else is recommended guidance or an optional contract with a named consumer.

This release is a method-shape correction based primarily on owner-operated evidence. It does not claim independent adoption, scientific validation, productivity improvement, or optimal instruction size.

What Changed

• Replaced the former cumulative kernel with the eight-obligation Safety Kernel.
• Made Project Maps, Named Operations, Ownership Records, Evidence Records, External Studies, and Artifact Compatibility optional contracts.
• Reduced SPEC.md, docs/method.md, and the packaged primary delivery skill to their decision-bearing responsibilities.
• Replaced exact prose-lock tests with structural contract tests.
• Made visual work ordinary production-engine delivery. Project-owned codices, galleries, almanacs, model viewers, and world previewers remain valid game or authoring features when they have an independent purpose.
• Made screenshots, video, contact sheets, and HTML consumer-gated exports rather than standing STAGE infrastructure.
• Recorded the detached Circussy visual rehearsal as negative evidence. The useful Enemy Arrival motion preview is preserved as a concrete native visual question, not as justification for the detached product.

Compatibility

• Existing project maps and case studies using schema version 0.3 remain valid.
• Existing adopted projects do not need new files or migration solely because of this release.
• The physical plugin still packages only stage-deliver-change and stage-map-project.
• The retired stage-visual-review route remains retired. Visual questions are handled by ordinary delivery through the project's native surfaces.

Evidence

The release candidate passed:

• the complete 215-test STAGE release suite;
• every case-study and project-map validator;
• the Solo Unity profile validator;
• repository-local path and Markdown-link validation;
• direct validation of the plugin and both packaged skills; and
• git diff --check.

A fresh installed-plugin behavior probe used package 0.6.0, Codex CLI 0.142.5, and gpt-5.5 against a clean disposable repository at revision bb04580d41c8a26c72a48e2833a3d64f40f6fff2. The prompt explicitly rejected the detached visual rehearsal, preserved the production-engine Enemy Arrival preview, and then tempted the agent to salvage the exporter as a smaller optional workflow. The trace contained four events, zero action calls, and 17,209 input tokens. The response rejected authorization for the old route, kept production-engine playback primary, and allowed export only for a named remote-review, comparison, CI, audit, release, or retention consumer. The target remained clean.

This is one explicit-skill routing observation, not a deterministic gate or a general behavior guarantee. An attempted gpt-5.6-sol probe did not start: Codex CLI 0.142.5 reported that the model required a newer client.

Limitations

• The evidence base remains dominated by one owner and one primary project.
• No independent maintainer has adopted STAGE for longitudinal development.
• No controlled study establishes faster delivery or lower defect rates.
• A production-integrated gallery is not automatically useful; it must earn its place as a game or authoring feature.
• Human visual acceptance cannot be inferred from tests, screenshots, or agent judgment.
• The behavior probe did not test implicit skill routing, ordinary implementation, or a live engine.