Skip to main content

STAGE 0.5.11

Released: 2026-07-22

STAGE 0.5.11 narrows always-visible routing metadata while keeping the full delivery, authority, stopping, preservation, and evidence rules in the skill bodies selected for a task. It keeps the 0.5 Practitioner Kernel and every supported artifact schema compatible.

Why This Patch Exists​

The installed plugin exposes skill descriptions and one plugin-level default prompt before a task selects and loads a skill body. Those routing surfaces had accumulated workflow policy that was already canonical inside the skills. The duplication increased ambient context and made ordinary delivery sound like a list of prohibitions rather than the primary route.

The visual-workflow dogfood also established a sharper boundary: native visual diagnosis remains useful when one explicit visual question needs specialist inspection, but detached rehearsal products and permanent galleries are not a default STAGE concern. A player or developer Almanac is ordinary product, authoring, or development-tool work when it has independent value.

Changes​

  • Keep all three skill descriptions concise and trigger-focused.
  • Keep full route policy in the selected skill body rather than duplicating it in always-visible metadata.
  • Replace the plugin default prompt's negative checklist with a positive project-native delivery instruction and one explicit-only boundary.
  • Retain stage-visual-review only for bounded diagnosis of one visual question through the real engine or game.
  • Add regression checks for routing-metadata size while preserving canonical policy in each packaged skill.

Practitioner Impact​

Ordinary requests still enter through stage-deliver-change. Project mapping and dedicated visual diagnosis remain opt-in specializations. The shorter metadata changes routing presentation, not authority: once selected, each skill still protects existing work, respects diagnosis-only and human-decision boundaries, verifies through production consumers, and stops rejected routes.

No detached contact sheet, visual-rehearsal package, gallery, or Almanac is created by default. If a game genuinely needs a persistent codex or inspection surface, build it as ordinary project functionality through the primary delivery route.

Compatibility​

The plugin package version is 0.5.11; SPEC.md is method version 0.5. Project-map and case-study artifact versions 0.1 and 0.3 remain supported. No game-project migration is required.

Verification​

The release is checked through:

  • the complete STAGE release gate;
  • direct Codex plugin and all skill-package validators;
  • repository-local project-map, case-study, schema, and Markdown-link validation;
  • focused routing-metadata and canonical-policy regression tests; and
  • a fresh installed-plugin behavior probe recorded after candidate installation.

The probe used the installed 0.5.11 candidate with gpt-5.5 and Codex CLI 0.142.5 against clean Lanternworks revision 541b2747096bb358daeff31b5d21994421582498. The agent refused to shrink, adapt, or continue the rejected detached workflow, proposed no analogue, and did not inspect the target repository. It produced seven trace events, two agent messages, and one command action that read the installed stage-deliver-change skill body. The target revision and working tree were unchanged. This is a semantic route-control pass and a strict zero-action ceiling failure; it is not reported as a structural zero-action pass.

Limitations​

Shorter routing metadata does not by itself prove lower whole-task token use, faster delivery, or more reliable model behavior. The selected skill still loads substantial project-independent policy, and Codex behavior remains bound to model, toolchain, prompt, repository state, and available context. The specialist visual route is retained from one owner's workflow evidence; it is not evidence that every game needs dedicated visual-review infrastructure. The 0.5.11 probe used 40,674 input tokens, including 23,296 cached input tokens, so it supplies no token-reduction evidence. Its one skill-body read also shows that concise routing metadata does not eliminate skill-loading cost.

Last updated on