Repository navigation
Roadmap
This page describes direction, not a release promise. Tagged releases, versioned source, issues, and current PRs determine execution status.
- Preserve the 48 authored methods, 96 complete EN/KO procedure bodies, and v1.4 Codex foundation on Apple-silicon Mac and Windows x64.
- Add supplemental reader guidance for source reconciliation, stakeholder context, understandable conclusions, and practical messages.
- Add native Claude entry on Apple-silicon Mac, a portable content-only Claude account skill, and Antigravity workspace/global content on Mac/Windows x64.
- Use
Powered by OpenSocratesonly after an eligible canonical method was read in full and actually applied; keep internal identities and evidence boundaries. - Preserve owned lifecycle operations, privacy, fail-open behavior, and rollback.
Windows native Claude Code, dedicated coding-specialist support, project memory,
and broad orchestration are outside this version. --host all remains Codex.
See v1.5.0
and the README for release scope.
- Reader usefulness: preserve what sources say, settled follow-ups, unknowns, and the authority needed before a draft makes a commitment.
- Host truthfulness: qualify each live surface and account origin separately; keep installation, loading, full reads, application, and outcomes distinct.
- Lifecycle safety: preserve unrelated settings, owned removal, disabled state, transactional recovery, and platform-specific filesystem boundaries.
- Canonical integrity: keep bilingual procedures, IDs, revisions, generators, and package identities reproducible without rewriting authored methods.
- Evaluation rigor: preserve failures and disagreements, then assess practical usefulness with a stated protocol rather than infer improvement from guide checks.
Candidate work needs an issue, acceptance criteria, scope, and evidence plan before it becomes scheduled. Relevant follow-ups include broader supported-host matrices, fresh-conversation workspace/global qualification, separate Chat and Cowork observations, and confirmation-grade held-out quality evaluation. No general quality, speed, or cost advantage is established by the release.
Earlier v1.2 question-led intervention and adjudication work is retained in PR #68 and PR #69, with its versioned evidence and limitations. Those records are not current draft status or human gold. The six-host v1.3 integrations were retired in v1.4; v1.5's new Claude and Antigravity profiles do not restore the old adapters or their receipt gates. The earlier coding/memory/orchestration v1.5 candidate, PR #95, was withdrawn in full and is not an implementation baseline.