v4.3.7

Verify this release before installing — see
documentation/how-to/release-verification.md.

Changes: CHANGELOG.md.

What's Changed

• fix(reporting): refresh active language after Claude tools by @Arcanada in #493

Full Changelog: v4.3.6...v4.3.7

v4.3.6

Verify this release before installing — see
documentation/how-to/release-verification.md.

Changes: CHANGELOG.md.

What's Changed

• fix(test): enforce shared source-history deadline regression by @Arcanada in #492

Full Changelog: v4.3.5...v4.3.6

v4.3.3

Verify this release before installing — see
documentation/how-to/release-verification.md.

Changes: CHANGELOG.md.

What's Changed

• ci(deps): bump the actions-minor-and-patch group with 3 updates by @dependabot[bot] in #486
• Adopt the approved Datarim brand identity by @Arcanada in #487
• fix: validate selected document revisions; prepare 4.3.3 by @Arcanada in #488

Full Changelog: v4.3.2...v4.3.3

Datarim 4.3.2

Datarim 4.3.2 includes Human Outcome Reporting 0.3.1. Supported Codex and Claude Code clients now receive current reply and artifact language preferences through synchronous startup callbacks, before the first progress message. Settings remain independent and default to English; changing them needs no reinstall.

Reusable notes and document excerpts inside a reply follow artifact language, including their examples. The surrounding explanation follows reply language. Explicit bilingual or translation requests, quotations, code and machine identifiers retain their exceptions.

Installers preserve foreign and safety hooks and intervening edits. A new Codex hook definition requires native review through /hooks; installation checks establish configuration, while activation and observed model behavior require a fresh session.

The English expectations compatibility fix from 4.3.1 remains included. Orchestration, /dr-auto, /dr-quick and JEV classification remain the main workflow features.

See the changelog, language setup, standalone installation, and release verification.

v4.3.1

Datarim 4.3.1 lets you choose the language of AI replies separately from the language of generated documents. It includes all changes since the last published release, 4.2.3. The language changes recorded under 4.3.0 in the changelog are included here; no GitHub Release was published for 4.3.0.

Choose your reply and document languages

Both settings default to English:

language:
  replies: en
  artifacts: en

For example, you can request French replies while keeping plans and reports in English. Settings live in:

• Personal defaults: $XDG_CONFIG_HOME/datarim/config.yaml, or ~/.config/datarim/config.yaml when XDG_CONFIG_HOME is unset.
• Shared project defaults: <project>/datarim/config.yaml.
• Private project overrides: <project>/datarim/config.local.yaml.

Shared defaults require an explicit team decision to track that file; keep local overrides private.

Explicit task or launch overrides take priority, followed by delegated language settings. After those, replies use private project, personal, shared project, then English defaults. Documents use private project, shared project or the legacy project document-language directive, personal, then English defaults. Conflicting project document settings are reported as an error. Native project and security rules retain their authority.

Preferences need no reinstall and are shared with standalone reporting. Existing conversations may retain older instructions. Follow the language setup guide and precedence reference.

Clearer reports and compatible checklists

Human Outcome Reporting 0.3.0 uses the same language preferences within Datarim or as a standalone skill. Fixed report labels have English, Russian, French, Arabic and Japanese catalogs; missing coverage produces a disclosed English fallback. Original quotations, supplied prose, code and machine identifiers are preserved.

The expectations parser now recognizes existing English labels such as Operator expectations and Status history consistently, so validation and requirement links use the same interpretation. Added checklist items and every acceptance-criterion link receive the same checks as original items. Missing statuses, unresolved links and other unmet conditions still block the applicable checks.

Four ways Datarim helps

• Orchestration coordinates specialist agents, checkpoints and feedback; an optional plugin also manages terminal sessions.
• /dr-auto takes a multi-stage task through its required work, checks and reflection. Archiving remains separate, and permission boundaries still apply.
• /dr-quick handles tiny edits and lookups. Edits still need acceptance and delivery evidence; larger requests return to the full workflow.
• JEV advises a model tier and reasoning effort. Its separate command safety checks work without an API key; classification advice requires configured API access and does not switch the model already running in your conversation.

Start with Datarim for everyone or the README. Before installing or updating, follow release verification. See the full changelog for details.

v4.2.3

Verify this release before installing — see
documentation/how-to/release-verification.md.

Changes: CHANGELOG.md.