OpenSocrates 1.3.0

OpenSocrates 1.3.0

OpenSocrates can retrieve complete reasoning methods when a judgment changes
within the same request, while preserving the task's goals, permissions, evidence
and completion requirements. Mechanical work does not need a method.

The release adds conservative, presentation-only English and Korean Guided
profiles. They help organize optional prose without rewriting authored procedures,
removing required public outputs, hiding uncertainty or imposing universal length
caps. Guided instructions are not verified compliance or a humanizer/detector claim.

• Claude/Codex use the existing native eligibility command first when available;
  empty eligible selection stays valid, and genuine runtime failure retains a
  constrained reference fallback. Other host delivery models remain unchanged.
• Multi-question guidance keeps each conclusion, missing input and reopening
  condition scoped to its question; model compliance remains probabilistic.
• All 48 canonical methods and 96 authored EN/KO procedure bodies are retained.
• All six existing host packages and the complete standalone Claude Chat ZIP ship.
• Codex discovery exposes the controller, rigor and trace. Method IDs remain stable;
  use the controller or complete internal references for prior method invocations.
• The packaged decision <host> --stream route supports volatile same-process
  delivery reuse. Context/epoch changes and process restarts invalidate assertions.
• Default native discovery avoids first-use product-state initialization while
  retaining safe cleanup of existing artifacts. Selected, emitted, agent-reported
  available and actually applied states remain distinct.
• Native runtimes target Apple-silicon macOS. Other native platforms are not shipped.

Evidence and support limits

Source, generated-content, installer and native-package gates protect this release.
Bounded synthetic model evaluation and complete visible-output measurements are
recorded in the repository; they do not establish full-catalog empirical superiority
or lower end-to-end cost. No discarded Compact or procedure-rewrite variant is
promoted.

Synthetic Codex runs using gpt-5.6-luna/max included300-second final-delivery
deadlines, including cases with a correct artifact already written. All failed
attempts remain recorded. A separate one-shot finite observation retains its
300-second marker; it is not evidence of a fixed response-time guarantee.
This release makes no demonstrated quality, naturalness, token-cost or latency
improvement claim.

Claude Chat support is artifact export/layout/reference integrity validated;
v1.3 cloud upload and activation unvalidated. An existing account-level custom
skill is not upgraded by the local installer. Exact public assets are verified
against the release job's validated bytes after publication. That provenance is
separate from cloud activation or method-application evidence.

The existing host-specific evidence limitations, including Desktop GUI, separate
Mac installation and signing/notarization status, remain explicit in the release
record. See release progress and evidence.

Install

After the matching npm publication:

npx --yes opensocrates@1.3.0 install --host codex

Choose the matching host for other packages. Review native hook trust normally;
do not bypass it. Verify the named ZIP and checksum before manual installation.

Post-release correction: the publication workflow refused to overwrite an
existing release and verified the uploaded bytes, but the GitHub API reports
immutable: false for v1.3.0. Therefore GitHub's platform immutable-release
protection does not protect this release. Do not replace its tag or assets.
GitHub documents
that enabling the repository setting applies only to future releases; the
draft follow-up PR #92
adds a workflow gate that requires the setting before future publication.