Skip to main content

STAGE 0.6.4

Released: 2026-07-23

The plugin package version is 0.6.4. SPEC.md is method version 0.6.

Release Decision​

This patch removes research ceremony from the ordinary external-audit route. An adoption-ready audit and a formal STAGE case study are different products: the former serves a project owner deciding whether to adopt a map, while the latter serves STAGE's optional Maintainer Laboratory.

No Safety Kernel rule, artifact schema, or project-facing runtime architecture changes in this release.

What Changed​

  • Renamed mapping's study depth to inspection depth.
  • Removed study.yaml from the default external-audit bundle.
  • Required an explicit formal-research request and an available Maintainer Laboratory before the mapping skill creates a case-study record or makes transferability claims.
  • Standardized the packaged workflow on external audit terminology while preserving historical trial and study names in archival records.
  • Recorded a bounded fresh-consumer probe of the released 0.6.3 visual correction behavior.

Compatibility​

  • Existing STAGE maps, audits, and historical studies remain valid.
  • Project-map and case-study schemas remain unchanged.
  • The plugin still packages only stage-deliver-change and stage-map-project.
  • Formal studies remain available through the STAGE repository; they are no longer implied by an ordinary installed-skill audit.

Evidence​

The release candidate passed the complete STAGE release gate, direct plugin validation, direct validation of both packaged skills, and git diff --check.

The recorded 0.6.3 consumer probe used gpt-5.5 with Codex CLI 0.142.5. It produced four trace events, zero action calls, and no detached visual substitute after the owner rejected that route. This is one bounded semantic observation, not a deterministic release gate or proof of causal prompt behavior.

Limitations​

  • The behavior probe does not establish visual quality, implicit routing, cross-model consistency, or independent adoption.
  • External audits still require accountable human review before adoption.
  • Historical study machinery remains repository-owned maintainer tooling, not a bundled practitioner CLI.
  • STAGE remains optimized first for the owner's personal workflow and does not claim empirical proof or universal fit.
Last updated on