Skip to content

Release Bearing Lite 2.0.0: opt-in lifecycle and simpler planning - #217

Merged
1wgrumph merged 2 commits into
mainfrom
docs/orchestrator-intake
Oct 18, 2026
Merged

1wgrumph merged 2 commits into
mainfrom
docs/orchestrator-intake

Conversation

@1wgrumph

@1wgrumph 1wgrumph commented Oct 18, 2026 •

Copy link
Copy Markdown
Contributor

What changed

Prepares Bearing Lite 2.0.0 and simplifies its planning lifecycle around the existing Orchestrator: architectural alignment and scope share one session, profile-enabled specialists contribute in dependency order, and the Orchestrator owns final design, assembly and handoff. Removes the Plan Integrator and separate Planning and Design routes, with explicit migration failures for retired configuration.

Makes Bearing hooks opt-in for an explicitly requested current native session/task. Ordinary sessions stay quiet and never fetch or fast-forward. Bound sessions retain ownership and assurance budgets. Review runs only when requested; author self-checks remain inside the author’s assignment. Removes fixed three-attempt limits and uses event-driven coordination.

Updates role loops, onboarding, role explanations, optional OpenCodeReview/Reverify and extra verification tool configuration. Preserves those capability selections in frozen dispatches. Fixes final register binding and unchanged-input evidence adoption.

Why

The goal is lower token usage and faster lifecycles while preserving a production-grade SDLC. Removes redundant planning sessions, uses optional specialist roles and event-driven coordination, and prevents installed hooks or old plans from activating a workflow without owner intent.

Validation

  • Parent personally reviewed the stable hook source with OpenCodeReview delegation; 95 targeted checks passed.
  • Isolated Claude Code native session completed two shell calls and a write with successful hooks; Muse loaded all 16 skills.
  • Final Node suite: 966/966; pinned isolated schema checks: 319/319; public/conformance subset: 56/56.
  • Pinned Codex 0.162.1 native discovery/assurance: 4/4; five consecutive runs and two concurrent copies passed, followed by the final 2.0.0 gate.
  • All seven manifests agree on 2.0.0; frozen install, dependency audit, npm package dry run and diff hygiene passed.
  • VnV profile: REFUSED_NO_VNV; this repository has no VnV policy directory.

Residual risk

Breaking profile migration: remove retired plan_integrator and planning_and_design roles and use the Orchestrator. Existing plans require explicit current-task opt-in before general hooks apply.

Cursor native loading remains unqualified. After disclosure of this gap, the owner directed merge and release on 2026-10-18; the release proceeds with that disclosed limitation. All required GitHub checks passed. Live installations and real-project lifecycle trials are a later phase.

@1wgrumph 1wgrumph changed the title Simplify opt-in lifecycle planning and hook guards Release Bearing Lite 2.0.0: opt-in lifecycle and simpler planning Oct 18, 2026
@1wgrumph
1wgrumph marked this pull request as ready for review October 18, 2026 14:31
@1wgrumph
1wgrumph merged commit f936510 into main Oct 18, 2026
6 checks passed
@1wgrumph
1wgrumph deleted the docs/orchestrator-intake branch October 18, 2026 14:31
1wgrumph pushed a commit that referenced this pull request Oct 21, 2026
Resolve conflicts in favor of main: PR #217 (2.0.0) already removed the
skill-text caps and MAX_CORRECTIONS. Keep the recovery test-model change,
the last remaining hard-coded correction cap, and reword the changelog.
1wgrumph pushed a commit that referenced this pull request Oct 21, 2026
The recovery contract-test model still forced OWNER_DECISION_REQUIRED on a
third correction and rejected a fourth. Skill-text caps were removed in
2.0.0 (#217); task-state.md now caps corrections only by an owner-supplied
allowance. A correction with new evidence now returns READY.

Refs #211. Supersedes #212.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant