Skip to content

Release Loop-Engineering v1.1.0#6

Merged
sheepxux merged 1 commit into
mainfrom
codex/v1.1.0-release
Jul 15, 2026
Merged

Release Loop-Engineering v1.1.0#6
sheepxux merged 1 commit into
mainfrom
codex/v1.1.0-release

Conversation

@sheepxux

@sheepxux sheepxux commented Jul 15, 2026

Copy link
Copy Markdown
Owner

What changed

  • Added the Idea-to-Loop proposal plane for turning vague requests into evidence-bearing Goal Contracts and human-approved executable contracts.
  • Added one domain-neutral finite Work Plan model with stable Parts, dependencies, Part-specific verification, immutable artifact snapshots, independent Part Gates, and a final Goal Gate.
  • Added proposal/decision/Goal-evaluation schemas, CLI flows, runtime replay and recovery, generic examples, bilingual documentation, and digest-bound fresh-session Skill evals.
  • Hardened state-lock acquisition and stale recovery against an ABA race, including async/thenable release behavior and fail-closed recovery.

Why

Loop-Engineering should support suitable long-horizon tasks in any domain without creating a separate Skill per use case. A user can start with a vague idea, review a verifiable Goal and suggested Loop structure, then let one universal Skill advance the approved work Part by Part without skipping failed checks or claiming completion early.

User impact

  • Finite composite goals can now make durable, independently verified progress across Parts.
  • Continuous monitoring remains on bounded discovery instead of being forced into a fake finite plan.
  • Domain-specific tools and evidence stay inside generic worker/evaluator boundaries.
  • Capability and runtime limitations are documented explicitly in Chinese and English.

Validation

  • npm test: 161/161 passing
  • npm run validate:examples
  • npm run validate:proposals
  • npm run doctor
  • canonical and official Skill validation
  • gh skill publish --dry-run: clean
  • npm run package:smoke
  • npm audit --omit=dev --audit-level=high: 0 vulnerabilities
  • npm publish --dry-run --access public: passing
  • multi-process stale-lock stress: passing

Summary by cubic

Release v1.1.0 adds an Idea‑to‑Loop “Propose” mode that turns vague requests into verifiable Goal Contracts and approved loops, plus a domain‑neutral finite Work Plan with Part gates and a final Goal Gate. It also introduces canonical SHA‑256 digests, hardened state locks, new schemas/templates, updated docs, and pinned runtime commands.

  • New Features

    • Proposal plane: non-executable LoopProposal with at most three material questions, explicit assumptions/readiness, and complete traceability; human LoopProposalDecision binds an exact proposal SHA‑256; deterministic compile to loop.yaml.
    • Finite Work Plan: stable Part IDs and acyclic dependencies; Part‑specific acceptance and artifact digest bindings; dependency‑ready scheduling; runner next now signals work, goal-evaluation, or completed; independent final Goal evaluation (goal-evaluation.schema.json).
    • Schemas, templates, and CLI: new proposal/decision/goal‑evaluation schemas and assets; CLI flows to validate, decide, and compile proposals; validation hardened for semantic/traceability checks.
    • Canonical digests: all artifact hashes computed over RFC 8785 canonical JSON; duplicate‑key rejection and tests added.
    • Reliability and safety: state‑lock acquisition/recovery hardened against ABA/thenable races with fail‑closed behavior; clearer troubleshooting; evaluator vs. authority boundaries documented and enforced.
    • Docs, examples, and evals: bilingual docs updated; new examples for proposals and a finite project; evaluator sessions and results for v1.1.0 included.
  • Migration

    • Pin the runtime to v1.1.0 in CI and local helpers: use npm exec --yes --package=github:sheepxux/Loop-Engineering#v1.1.0 -- loopctl and update any pinned fallback in adapters/scripts.
    • Use the proposal flow before execution: create a LoopProposal, obtain a human LoopProposalDecision bound to its SHA‑256, then compile to loop.yaml; do not execute unapproved candidates.
    • For finite composite work, adopt workPlan in loop.yaml, run one Part at a time (or as approved), and perform an independent final Goal evaluation before marking completed.
    • Re-run loopctl doctor and loopctl validate after upgrading; follow new lock‑gate guidance when recovering stale locks.

Written for commit ccc388c. Summary will update on new commits.

Review in cubic

@sheepxux
sheepxux merged commit c0f6b62 into main Jul 15, 2026
4 checks passed
@sheepxux
sheepxux deleted the codex/v1.1.0-release branch July 15, 2026 18:01
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