CHANGELOG.md

Changelog

Unreleased

1.5.3 - 2026-06-25

• Fixed the manual release workflow so GitHub Actions configures a bot git identity before creating annotated release tags.

1.5.2 - 2026-06-25

• Added --no-contents to mlab export / npm run export and forwarded it through mlab done / npm run done, allowing reader exports to omit generated Contents pages while keeping export manifests explicit about the choice.
• Aligned the manual release workflow with the public release checklist by creating annotated tags and using the matching CHANGELOG.md section for GitHub release notes.

1.5.1 - 2026-06-19

• Hardened practice benchmark and strategy runs for heterogeneous model testing: per-row provider/model failures now persist as error rows with evaluated/error denominators in summaries, reports, console output, and practice-strategy eval snapshots. Runs with some errors now report partial; runs where every comparison errors fail instead of masquerading as a pass.
• Cleaned generated-artifact discovery so incomplete practice-strategy runs are marked in_progress without polluting recommendations, while completed benchmark/eval artifacts expose evaluated/error row counts.
• Expanded the practice prose guard to catch visible reasoning tags, analysis headings, and prompt/task chatter from reasoning-heavy models before judging.

1.5.0 - 2026-06-19

• Added generated-artifact discovery across driver runs, practice runs, practice benchmarks, strategy comparisons, eval snapshots, and golden-path guides, surfaced through mlab artifacts, status, and report.
• Added mlab eval practice-strategies to snapshot strategy-comparison evidence under state/evals/, compare against baselines, and optionally fail CI/release checks with --fail-on-regression.
• Added mlab golden-path as a first-use product guide that can persist onboarding evidence under state/golden-path/.
• Extended mlab drive with resume-safe continuation, read-only artifact inspection tools, an eval snapshot primitive, and generated-artifact context in model observations.
• Added contributor primitive-contract docs, installed-package smoke coverage for artifacts/evals/golden-path, and a manual GitHub Actions release workflow scaffold.

1.4.0 - 2026-06-19

• Scoped the future mlab drive model-driver primitive: an interactive, policy-bound command loop that exposes Manuscript Lab tools to a model, records decisions under state/driver/, and preserves gates and approvals as the readiness boundary.
• Added the first bounded mlab drive implementation slice with a validated tool catalog, curated policy packs, strict project-relative path fences, dry-run, mock-decision, and live model decision support, persisted state/driver/ step ledgers, wrapper/npm routing, and installed-package smoke coverage.
• Hardened mlab drive after adversarial review: child primitives now inherit pinned workspace/config roots, ephemeral live-model driver runs do not force model-call audit writes, review-only has an enforced allowlist, --no-write blocks mutating primitives, traversal segments are rejected, and --resume is explicitly rejected until real history reconstruction ships.
• Tuned mlab drive for actual model use: model-backed runs now default to a four-step observe/decide/act loop and carry compact parsed-result summaries into the next decision, while heuristic runs still default to one conservative step.
• Made the driver-exposed practice.bench primitive less macro-shaped by adding bounded model-controlled knobs for exercise set/id list, seeds, candidate count, and repair rounds.
• Hardened practice generation from live GLM traces: direct baselines and candidates now prose-guard outputs before judging and retry once with a strict final_prose contract when a model leaks planning/meta text.
• Added mlab practice propose / npm run practice as a safe creative-writing exercise primitive: it generates multiple candidates, judges them against a hidden exercise test, revises the winner, writes state/practice/ artifacts, and is exposed to the driver as the approval-gated practice.propose tool.
• Added mlab practice compare for controlled direct-vs-mlab evidence: it runs a direct same-model baseline, a practice proposal loop, and a blind pairwise judge, then writes comparison artifacts under state/practice-evals/.
• Hardened practice comparisons with structured prose-only revision outputs, prose/meta leakage guards, bounded repair rounds when the direct baseline wins, and copy checks so duplicated direct baselines cannot count as distinct mlab wins.
• Expanded the practice exercise battery and added mlab practice bench, which runs direct-vs-mlab comparison matrices across exercise sets, models, and seeds, writes self-contained state/practice-bench/ ledgers, separates first-pass wins from repair recoveries, supports --judge-model, and exposes practice.bench to the model driver as an approval-gated oracle-guided workflow benchmark tool.
• Added mlab practice strategies, a first-class strategy-comparison command that runs preset practice-loop shapes, writes state/practice-strategies/ ledgers with nested benchmark artifacts, recommends per-exercise strategies from aggregate win/delta/cost/recovery evidence, and exposes practice.strategies to the model driver and installed-package smoke tests.
• Aligned GLM examples and taste-arbiter defaults on OpenRouter GLM 5.2 so the driver, practice lab, provider docs, and release examples point at the same current model route.

1.3.0 - 2026-06-17

• Added room diagnose as the first writers' room step. It writes a deterministic story-foundation diagnosis, readiness grade, missing inputs, warnings, visible-file manifest, and recommended next command under state/room/.
• Strengthened beat boards with causality fields: causal_link, choice, consequence, and turn, so room material is easier to use as story pressure instead of loose idea inventory.
• Added the scene.turn typed review sensor and panel routes for checking movement, pressure, turn, and consequence after prose exists.
• Updated mlab help, status/report artifact discovery, install-anywhere smoke coverage, Pi prompts, and public docs so diagnosis, causal beat boards, and scene-turn review are first-class package surfaces.

1.2.0 - 2026-06-17

• Reframed Chorus as a prose line lab by default: chorus run now samples beat-level candidates and writes CONTACT_SHEET.md, per-beat contact sheets, plan-quality.json, metrics, and reports without assembling prose.
• Kept the pick-and-assemble path available through explicit --assemble, chorus judge, and chorus assemble, with reports warning against wholesale merges.
• Tightened Chorus beat plans with sensory/object targets, continuity limits, stricter no-new-canon guardrails, plan-quality warnings, and usage/cost metrics for model-backed candidates.
• Updated mlab help, status/report artifact links, installed-package smoke coverage, Pi prompts, and public docs so Room, Chorus, and candidate loops present as first-class but distinct workflows.

1.1.0 - 2026-06-17

• Added a writers' room workflow research memo mapping industry room practices to future Manuscript Lab commands, prompts, and durable artifacts.
• Added the room protocol command for blue-sky role cards, showrunner decisions, beat-board materialization, table-read packets, and room reports under state/room/.
• Added the room.table_read typed review sensor, Lightning/OpenRouter panel routes, Pi room prompts, project scaffold support for state/room/, and root-aware room command smoke coverage.
• Added the first Chorus prose ensemble MVP: npm run chorus / mlab chorus can plan beat-level voice experiments, sample local/mock/model-backed candidates, pick provisional beat prose, assemble state/chorus/ output, and report runs without modifying draft/.
• Promoted Room and Chorus into first-class package surfaces: status and report now expose recent runs and artifact links, project lifecycle commands mount/archive their state, generated next-step hints are mlab-aware in installed workspaces, Pi prompt coverage includes report/plan entrypoints, and installed-package smoke tests assert the cockpit/report integration.

1.0.3 - 2026-06-17

• Cleaned public changelog wording so the v1 history stays focused on fresh config-first installed projects.

1.0.2 - 2026-06-17

• Aligned public release docs with the published npm package and GitHub release: added npm/GitHub/CI badges, deleted the old v1 branch plan, and removed stale pre-publish language from install docs.

1.0.1 - 2026-06-17

• Fixed project-free mlab doctor --no-network so public one-off registry smokes in blank directories report missing project/git context as diagnostics instead of failing the command before mlab init.

1.0.0 - 2026-06-17

• Set the v1/npm-publishing path around fresh config-first installed projects while preserving template-clone compatibility.
• Added v1 release scope notes, stronger config validation, project-free validation hints, project-local install smoke coverage, and local one-off npm exec package smoke coverage.
• Added public CLI aliases for review, revision, comparison, merge, and audit workflows while preserving compatibility command names.
• Added deterministic export-ready gate templates and broadened manuscript-ready / done:no-export gate integration with persisted gate artifacts and final project filesystem sync.
• Broadened evidence reporting with richer claim/source normalization, risk/status-aware issue output, citation resolution, source validation, and claims list filters.
• Added a deterministic broken-whitepaper fixture for demonstrating useful not-ready output without model or network calls.
• Excluded generated example .doccheck artifacts from package contents and extended packlist coverage.

0.9.0 - 2026-06-17

• Tightened public contributor guidance with the full verification gate, package-boundary checks, current install-anywhere examples, and done export-format command documentation.
• Added mlab version / mlab --version, project-free mlab doctor --no-project, and packed-tarball temporary-prefix global install smoke coverage for help/version/doctor/init/validate/gate/refusal behavior.

0.8.0 - 2026-06-17

• Made mlab review:run root-aware in packed install-anywhere workspaces, loading bundled review suites, model panels, and prompts from the package while writing review artifacts and issue-ledger updates under the configured manuscript root.
• Added installed-tarball smoke coverage for review:run --dry-run from the workspace root, manuscript root, and nested draft/ directory, plus a non-network mocked review run that saves artifacts and imports ledger issues.
• Made model-backed revision commands root-aware in install-anywhere workspaces: revise:candidates, compare:candidates, taste:arbiter, and diff:audit now load workspace/project .env files before provider setup and default model-call audit ledgers to the configured manuscript root.
• Added non-network installed-tarball smoke coverage for the model-shaped revision chain: candidate generation, comparison, taste gate, and diff audit from a nested draft/ directory.
• Made mlab done configurable for install-anywhere release gates with --export-formats, --export-slug, --export-out, and --include-todo-exports, while keeping the default reader-export expectation at Markdown, HTML, EPUB, and PDF.
• Added installed-tarball smoke coverage for running full done with Markdown/HTML exports from the workspace root, manuscript root, and nested draft/ directory.
• Guarded template-only wrapper commands (init without --profile, project:*, story:*, and related aliases) so they refuse outside the template clone root instead of creating confusing legacy project workspaces in install-anywhere repositories.

0.7.1 - 2026-06-17

• Loaded local .env values in mlab doctor so provider-key diagnostics match model-backed command behavior without printing secret values.
• Made mlab model:calls and mlab report honor MODEL_CALL_AUDIT_DIR when inspecting model-call ledgers from calibration or external workspaces.

0.7.0 - 2026-06-17

• Sharpened the README demo payoff with a concrete report excerpt and an inspection trail for the technical-whitepaper fixture.
• Made mlab issues, mlab revise:candidates --dry-run, mlab compare:candidates --dry-run, mlab taste:arbiter --dry-run, and mlab merge:winner root-aware in packed install-anywhere workspaces, with smoke coverage from the workspace root, manuscript root, and nested draft/ directory.
• Added packed install smoke coverage for merge:winner --apply --audit --static-only, keeping candidate and revision-audit artifacts under the configured manuscript root.

0.6.0 - 2026-06-16

• Added exports/manifest.json generation to mlab export / npm run export, including export ID, source commit when available, input hashes, output hashes, file sizes, formats, git dirty state, and chapter metadata.
• Added installed-mode smoke coverage to ensure export manifests stay under the configured manuscript root and appear in mlab report.
• Added mlab report / npm run report for text, JSON, and HTML readiness summaries that combine status, evidence, gates, review runs, revision trails, model-call counts, exports, blockers, and suggested next steps.
• Added installed-mode smoke coverage for report --json and report --write, with generated reports staying under the configured manuscript root.
• Made the technical-whitepaper fixture config-first so it can be validated and reported in place without changing the active template project.
• Added reports/ to private/generated path hygiene checks and package exclusions.

0.5.0 - 2026-06-16

• Added shared protocol path helpers for separating package assets, workspace roots, manuscript roots, configured state directories, and configured export directories.
• Made status, compose, check, done:no-export, export --formats md,html, and review:report work from packed install-anywhere workspaces.
• Split doccheck package assets from project files so installed workspaces no longer need copied checks/, reviews/, .pi/, docs, or templates.
• Added installed-tarball smoke coverage for running commands from the workspace root, manuscript root, and nested draft/ directory.
• Added installed-mode export coverage that keeps generated runtime packets, .doccheck artifacts, and Markdown/HTML exports under the configured manuscript root instead of the package directory.

0.4.0 - 2026-06-16

• Added config-first mlab init --profile whitepaper --root manuscript for the install-anywhere alpha.
• Added a neutral whitepaper scaffold with manuscript-lab.config.json, section contracts, taste/style docs, source and claim registers, issue ledgers, and truth-state placeholders.
• Preserved legacy template init for bare init, project:init, and story:init.
• Added an installed-tarball smoke test that packs the package, installs it into a disposable npm project, initializes a workspace, and runs validate, claims, citations, and gate commands.
• Added a packlist assertion to keep private/generated project files out of package contents.

0.3.0 - 2026-06-16

• Added mlab validate / npm run validate for deterministic file-protocol discovery and validation across template-first and config-first workspaces.
• Added deterministic evidence commands for listing unsupported claims, checking citation markers, reporting evidence state, and adding local source records.
• Added mlab gate / npm run gate for initial section, citation, and manuscript readiness gates with optional JSON artifacts.
• Expanded CI fresh-project smoke coverage for validate, evidence, citations, and gate commands.

0.2.0 - 2026-06-16

• Added product strategy documentation for the "local CI for prose" direction.
• Added a Codex skill adapter, validator, and installer for Manuscript Lab shipping work.
• Added active GitHub Actions CI for tests, audits, smoke checks, and package dry-runs.
• Added draft protocol/install/gate/evidence design docs and the public technical whitepaper tutorial fixture.
• Fixed doctor ignore checks so fresh CI checkouts validate private/generated path rules correctly.

0.1.1

• Added npm run doctor and manuscript-lab doctor for environment and release-health diagnostics.
• Documented doctor in onboarding and CI guidance.
• Updated public-readiness notes now that the doctor gap is closed.

0.1.0

Initial public-ready release checkpoint.

• Extracted reusable writing harness from a live document workspace.
• Added runtime context packets, static and model-backed checks, typed reviews, issue-ledger revisions, candidate arenas, taste gates, diff audits, and exports.
• Added optional Pi skills and prompt commands.
• Added public onboarding, architecture, security, contribution, and release readiness docs.
• Added a lightweight manuscript-lab / mlab command wrapper.
• Documented the intended CI workflow for tests, audits, package dry-run, and a fresh project smoke test.
• Defaulted generated project files, drafts, exports, logs, and state to ignored paths so private writing work stays out of the harness repo.