v0.20.0 - First-party UI/UX closed loop (vNext)

v0.20.0 - First-party UI/UX closed loop (vNext)

Released 2026-08-15.

Major: implements the complete vNext capability matrix resolved by Wayfinder
map #23 — interactive requirement shaping, tool-neutral design decisions,
dual-track implementation review, evidence gates, recirculation, and
cross-run learning — delivered as slices S1-S6 in one unified release
(PR #42). Spec authority: docs/specs/ui-ux-vnext/.

Tiered run profiles (process weight scales with consequence)

• Every run is graded P1 point-fix / P2 standard / P3 full by
  declaration-touch surface, recorded in a mandatory run-profile block at
  the top of plan.md; promotion is automatic, demotion is user-only.
• P1 covers single-finding repairs (including line-level R2 backfills)
  without opening a shaping session; P3 carries the full gate matrix.

Rule registry and governance

• skills/design-playbook/references/rules.md: unified first-party
  registry (13 entries — CRAFT-01..08 migrated with zero ID breakage,
  A11Y-01/RESP-01 first-party, I18N/PERF/SEC placeholders), G8-validated at
  product level and (per tier) at run level.
• craft-guard detector tables migrate atomically to the seven-column
  three-state format; the historical N/A contradiction is fixed by splitting
  applicability from result. Historical run artifacts are never rewritten.
• rules-governance.jsonl: append-only governance log; promotion to
  machine-enforced and exemptions are user-only decisions.
• Learning candidates derive purely from run history
  (3 distinct runs, 2 task contexts, 0 unexplained false positives).

Interactive requirement shaping

• Stateful S0-S6 session model with batched confirmation points
  (question caps, forced-to-assume after two unproductive batches);
  assumed fields age into mandatory questions after two acknowledged runs.
• Session artifacts are append-only under <run>/shaping/ with an
  alive→archived→discardable lifecycle; the contract stays the sole
  persistent decision authority (zero-breaking v1 extension).

Dual-track review and evidence

• Product track (per-AC supported/unsupported/unverifiable), interaction
  track (seven dimensions with a strict fact/judgment split — judgment
  findings are advisory and never block without user escalation), and a
  cross-cutting applicability matrix (accessibility + responsive in the
  first version; placeholders never skip silently).
• Six-block point-back report with coverage statements (G11) and
  five-state×page sampling matrices (required at P3).
• Manifest entries carry optional method-semantics keys; human-subject
  evidence without population/ethics is quarantined and cannot support a
  pass.

Recirculation machine-checked

• Repair-round counting with two-round stop → escalated-stop narration
  (verdict values unchanged).
• G12 tier-boundary gate reuses the G7 contract diff; tier violations
  escalate the run instead of being waivable.
• R1-R5 routing, DD supersedes re-entry, and baseline-drift re-review
  narrate through run-status with minimal invalidated-evidence sets.

Design decisions (tool-neutral)

• decision-report deepens into versioned DD entry blocks behind the
  unchanged Fill-facing header; exploration is tiered record/compare/explore
  (R/C/E); providers are replaceable data-contract adapters whose assets
  are reference-only.

Breaking

• Severity legacy values high (blocking)|high|med|low are no longer
  legal (two-stage migration completed; ADR-0028). Write the axis values
  directly; re-checking pre-vNext run artifacts will correctly fail.

Verification

• Independent dual-track review (product + interaction + cross-cutting):
  verdict Pass, zero blocking findings; four advisory findings fixed in-PR.
• Full local matrix green (validate, doctor, run seam, 16 root suites,
  MCP stdio groups, doc links); CI Validate plugin pass; dogfood P3
  fixture passes the full chain under --strict.

v0.14.1 - Lockstep main and DSH publishing

v0.14.1 - Lockstep main and DSH publishing

Released 2026-08-14.

Patch: makes the public DSH bundle part of every stable release transaction.
One vX.Y.Z tag now drives both npm publishers and one shared GitHub Release.

Fixed release group

• design-playbook and dsh-design-playbook publish at the same version.
• The DSH bundle depends on the matching ^X.Y.Z main-package line.
• Release, validation, and doctor diagnostics consume one shared lockstep
  policy and fail before publication when either invariant drifts.

Ordered publication and verification

• The DSH workflow waits for the exact main-package artifact before it
  publishes.
• The main workflow waits for the exact DSH artifact and verifies its npm
  provenance before creating the single GitHub Release.
• Both packages continue to use separate npm Trusted Publisher workflows with
  job-scoped GitHub OIDC and no long-lived npm token.

Recovery

• Fresh tag runs fail closed when an npm version already exists.
• Explicit manual recovery verifies an existing DSH package without
  republishing it.
• Shared GitHub Release recovery remains available after both registry
  artifacts exist.
• No new dsh-v* tag is created; historical dsh-v0.14.0 records remain
  unchanged.

dsh-design-playbook 0.14.0

dsh-design-playbook 0.14.0

Thin DSH bundle bridging design-playbook's preview + evidence MCP servers.
Depends on design-playbook@^0.14.0.

v0.14.0 - DSH integration + architecture deepening

v0.14.0 - DSH integration + architecture deepening

Released 2026-08-14.

Minor: DSH (DeepSeek Harness) plugin integration lands as a third install
surface alongside Claude Code and Codex. Six architecture deepenings
eliminate duplication across the preview and evidence runtimes.

DSH plugin integration (Issues #15-#20)

Three-phase integration: P0 skills softlink → P1 npm bundle → P2 MCP bridge

• commands. All three verified by real DSH boot smoke tests.

P1: npm bundle + skills provider (Issue #17)

• packages/design-playbook/lib/index.js — Cordis plugin entry; registers
  8 skills on ctx.skills via registerProvider() + __dirname (the
  !!js evaluation scope has no require, so customSkillDirs cannot
  point into package resources).
• packages/design-playbook/cordis.patch.yml — one bundle patch row
  mounting the plugin by package name.
• packages/design-playbook/package.json — adds dsh.bundle.patch +
  dsh-plugin/dsh-bundle keywords + main + lib in files[].

P2: MCP bridge (Issue #19)

• packages/dsh-design-playbook/ — thin npm bundle package depending on
  design-playbook. Its cordis.patch.yml bridges preview + evidence
  MCP servers via @deepseek-ai/dsh-mcp-client stdio.
• Path resolution uses process.getBuiltinModule('node:module') .createRequire(baseUrl).resolve(...) — DSH !!js scope has no
  require but exposes baseUrl (profile dir), so createRequire
  resolves design-playbook from the profile's node_modules.
• DSH smoke verified: both MCP tools register + tools/call round-trip
  passes (preview returns "summary is required"; evidence rejects
  missing schemaVersion per ADR-0018 fail-closed).

P2: commands (Issue #20)

• lib/index.js extended: inject grows to ['skills', 'commands'];
  apply() registers 6 slash commands (design-io, doctor,
  run-review, run-status, ui-review, ux-spec).
• Each handler loads commands/<name>.md, substitutes $ARGUMENTS,
  and injects via agent.followup() (not inject() — slash commands
  are explicit user actions that open a turn).
• DSH smoke verified: 6 commands register; /design-io <args> executes,
  handler calls agent.followup() with the substituted prompt; frontmatter
  stripped; $ARGUMENTS substituted; unknown commands rejected.
• Full interactive round-trip verified: preview server starts on fixed
  port, Chrome app window launches, HTTP POST auto-submits confirm form,
  transaction returns confirmed=true + floor_pass=true, decision entry
  written to disk.

Architecture deepening

Binding digest unification

• compute_binding_digest() moves to integrity.py as the single source
  for the binding shape (canonical JSON + SHA-256). transaction.py
  write side and load_entry() read side both import it; G5 gate
  _valid_decision_entry() consumes the same function. Three independent
  implementations collapse to one.

Prototype resolution unification

• _resolve_prototype() extracts the path/html resolution + digest
  computation that was duplicated 3× in transaction.py. Pre-lock
  binding and in-lock TOCTOU re-validation share one function.

Action dispatch registry

• _run_actions 75-line if/elif chain → _ACTION_HANDLERS dict mapping
  do → handler function. Each handler owns its validation + Playwright
  call. Adding an action type adds a function + one entry, not an elif.

Evidence runtime-object validation

• _validate_runtime_object() extracts url/type/state/actions validation
  from execute_capture_plan so the handler reads cleanly.

Preview server argument validation

• _validate_preview_args() extracts the 6-field validation from
  handle_preview_prototype.

Dead code + duplication cleanup

• _log() duplication removed (capture_runtime.py imports from
  util.py). Dead _require_str() removed. type: ignore replaced
  with assert.

v0.13.0 - Architecture deepening (C1-C6)

v0.13.0 - Architecture deepening (C1-C6)

Released 2026-08-09.

Refactor-only minor: no new commands (inventory stays 6), no layout change —
six architecture deepenings that make the bundled runtimes, gates, and
imports single-homed. Behavior changes are narrow, sanctioned corrections
(see below); everything else is movement, not semantics.

Contract v1 lands atomically (C2, ADR-0018)

• Capture contract rules move into one package-internal module
  mcp/evidence/capture_contract.py: write-side parse_capture_contract,
  read-side validate_capture_snapshot, and the contract-fields JSON Schema
  fragment the provider composes (const/enum/required share one source, so
  schema and parser cannot drift).
• G6 now validates bound manifest request snapshots through the same read
  authority instead of hand-written partial checks. Sanctioned correction:
  malformed viewport shape or missing freeze fail closed (was lax).
  New rule IDs G6.capture_viewport_shape / G6.capture_freeze; existing
  G6.capture_schema / G6.capture_viewport keep wording.

Preview integrity deepens (C1, ADR-0008/0013)

• One read-only mcp/preview/integrity.py owns prototype digest, feedback
  floor, current round, and confirm validity; G5 and run status consume the
  same snapshot. run status aligns with G5 on Preview occurrence.

Stage registry (C4, ADR-0021)

• scripts/stages.py is the single home for named stage specs (SKILL.md step
  mirror), regular resume actions, and the shared artifact names
  (evidence/, manifest.jsonl, point-back.md, decision-report.md,
  spec.md); run status and G6 share them, so the table and the names cannot
  disagree.

Closed-loop gate split (C3, ADR-0023)

• validate_run.py (1262 → 272 lines) is a thin orchestrator over focused
  gate modules: g1_spec, g2_g4_pointback, g5_preview,
  g6_evidence, g6_warnings, with shared ledger/manifest parsing in
  g6_records. Rule IDs, messages, ordering, and exit codes remain unchanged.

Real import seam (C5, ADR-0022)

• Cross-runtime sys.path adapters (31) collapse to one package-root
  bootstrap + absolute design_playbook.* imports; mcp/ and scripts/
  become importable packages. The duplicated root scripts/run_status.py
  dev copy is removed (one copy remains, in the package).

Preview versions collapse (C6, ADR-0024)

• versions.py no longer reaches through transaction's private seam: the
  shared primitives (atomic write, directory lock, entry load, log render)
  are transaction's public API, and the lock policy is one source
  (DIRECTORY_LOCK_*) — no second lock namespace to desync.

Release-boundary notes

• Per ADR-0018, capture contract v1 ships as one merge/release boundary;
  old unversioned run-local evidence requires recapture (no dual-read path).
• main stays the stable channel (ADR-0015); this tag is the first surface
  carrying the deepening branch.

v0.12.0 - vNext contracts + shallow preview control

v0.12.0 - vNext contracts + shallow preview control

Released 2026-08-08.

New commands and run machinery

• Packaged run-status and doctor commands (inventory grows from 4 → 6; ADR-0015 minor, not patch).
• Persistent design contract v1 + structured G1–G6 diagnostics on the validate surface.
• G7 contract drift gate against the bound contract snapshot.
• Capture plan requires schemaVersion: 1 (fail-closed without it).
• Orchestrator skill routes: load-map, first-run, a11y, and repair pointers (ADR-0016–0020).

Preview control interaction (Scheme A′)

Trusted parent control chrome (mcp/preview/control.*):

• Abort uses an explicit second-confirm popover (replaces 4s arm): fixed viewport positioning, Tab trap, Cancel / Esc / outside dismiss.
• Pill Revise is a real /decide submit (not open-drawer only).
• Pill Confirm keeps arm → confirm mis-tap protection (no one-click Approve; no fake "LGTM" inject); Esc / outside click / timeout undo the arm.
• Pill quick 1-line feedback (wide layouts only) shares state with the drawer textarea.
• Status chip focuses the feedback field; Ctrl/Cmd+Enter always routes confirm through isSubstantive (ADR-0008 floor).
• Visual tokens: control heights unified; float notes 999, bar 1000, drawer 1001; drawer overflow no longer clips the abort popover.

G5 sandbox / iframe pin bridge / decision token paths are unchanged.

Dogfood

• Live HITL design-io dogfood 2026-08-08-vnext-live pass (confirm-round-2 + HTTP fill-host observe v1; G5/G6/G7 green).

Distribution

• Installable inventory: 8 model skills, 6 commands, 2 bundled MCP servers.
• COMMAND_INVENTORY admits the 0.12 command set explicitly (stable main channel).

Validation

• python scripts/validate.py — VALIDATION PASSED
• Seam + contract v1 + G7 + vNext integration + vnext_live_dogfood.py preflight — green
• python packages/design-playbook/tests/test_floor_frontend.py — FRONTEND FLOOR TEST PASSED (incl. A′ scenarios)
• Preview MCP suites (test_browser_control, test_server_stdio) green

v0.11.1 - trusted publishing verification

v0.11.1 - trusted publishing verification

Released 2026-08-07.

Release automation

• npm publishing now runs from .github/workflows/release.yml through GitHub Actions OIDC and the package-scoped npm Trusted Publisher.
• The workflow binds a stable SemVer tag to the exact main commit and package version before running release gates, the full test suite, and package inspection.
• Registry availability and npm provenance are verified before the matching GitHub Release is created.

Security posture

• The GitHub npm environment is restricted to v* release tags, and protected release tags cannot be deleted or retargeted.
• No npm automation token is configured in GitHub; the trusted connection grants only the npm publish action.

Canvas interaction reliability

• New-anchor comment focus now completes synchronously, so a delayed callback cannot steal focus after the user moves to another control.
• The browser regression test deterministically covers immediate focus handoff followed by Ctrl/Cmd+Z anchor undo.

Package compatibility

• Public contracts remain compatible with v0.11.0; this patch updates distribution metadata and fixes the Canvas focus race.
• Installable inventory remains 8 model skills, 4 commands, and 2 bundled MCP servers.

Validation

• Clean origin/main baseline: python -m pytest -q passed 190 tests plus 83 subtests.
• npm pack --dry-run: design-playbook@0.11.1, 53 files, 122.8 kB packed / 460.9 kB unpacked.

v0.11.0 — local canvas versions

v0.11.0 - local canvas versions

Released 2026-08-07.

New: local canvas version control

• Preview runs can create immutable named versions from committed decisions.
• timeline() combines decisions and named versions in timestamp order, while state_at() replays historical state without mutating the active run.
• fork() derives an independent linear run from a selected historical round so exploration does not rewrite the source history.
• Anchor v2 records stable node identity and structural features for more reliable feedback targeting.

Transaction integrity

• Preview writes are serialized across processes for the full transaction, including projection updates.
• Confirmation binds to the exact rendered candidate and fails closed when state changes or validation cannot be completed.
• Append-only decision and version records use atomic replacement and explicit committed-versus-projection failure semantics, preventing unsafe retries after durable commits.
• Prototype HTML and path modes are explicit, and browser-side undo preserves native comment editing while correctly removing canvas anchors.

Distribution

• Installable inventory remains 8 model skills, 4 commands, and 2 bundled MCP servers.
• The Preview MCP package now includes local version projections and the hardened transaction path.
• CI requires Chromium-backed canvas version-control end-to-end coverage.

Validation

• Release functional gates passed: version consistency, static validation, run seam, adapter floor self-check, and tag availability.
• python -m pytest -q: 183 tests passed, plus 67 subtests.
• npm pack --dry-run: design-playbook@0.11.0, 53 files, 122.8 kB packed / 461.0 kB unpacked.

v0.10.0 — run review

v0.10.0 — run review

Released 2026-08-04.

New: cross-run run-review

• New fourth command, run-review, reviews multiple user-side .scratch/<run>/ Design I/O runs without becoming another step in a single run.
• Markdown contract starts with run-review/v1: inclusion manifest, per-run gate table, repeat-blocker frequency table, point-back citations, then row-derived rollup.
• Runs without point-back.md are listed as skipped; fewer than two included runs refuse cleanly.
• Gate status comes only from a real validate_run.py exit when that script is available; otherwise the report says not checked.

Integrity boundaries

• Repeat blockers use observed text after casefold + whitespace collapse, then char-for-char equality. Literal differences remain separate.
• _none_ is a normal result. The command must not loosen normalization to manufacture repeats.
• No new run ledger, prose lessons, baseline writeback, or semantic clustering.
• tests/test_normalize_lockstep.py keeps repo-side aggregate_runs.normalize() aligned with the shipped command wording, while OPP-21 continues to count distinct runs rather than ledger rows.

Distribution

• Installable inventory: 8 model skills, 4 commands, 2 bundled MCP servers.
• Claude marketplace instructions now use the explicit public HTTPS URL. This avoids Claude Code 2.1.220 selecting SSH for owner/repo shorthand on hosts without a GitHub SSH key.
• v0.9.2 stable-main identity and version-to-command inventory enforcement remain intact.

Validation

• scripts/validate.py VALIDATION PASSED; scripts/doctor.py DOCTOR PASSED; claude plugin validate --strict passed.
• Run seam passed; 54 repo unit tests passed.
• MCP gates passed: 27 shared transport/Preview tests, 8 Evidence pure-path tests, and 6 real-browser Evidence capture tests; frontend floor passed.
• npm pack --dry-run: design-playbook@0.10.0, 50 files, 110.7 kB packed / 388.6 kB unpacked.

v0.9.2 — release-identity marker (stable main)

v0.9.2 — release-identity marker (stable main)

Released 2026-08-04.

Version-marker patch: no changes to the installable surface — skills,
commands, and MCP adapters are byte-identical to v0.9.1 (only the version
fields move). Cut so the version line carries a released marker for the
ADR-0015 stable-main restoration and the gate hardening that followed it.

Marked (repo-side, not in the npm/plugin tarball)

• ADR-0015 — stable main channel. main is the public install
  surface; its version and installable inventory always equal the latest
  formal release. The prematurely landed v0.10 run-review command was
  moved back to feature/v0.10-run-review and main restored to the
  v0.9.1 inventory (e5c9ed7).
• OPP-01 — version-to-inventory invariant. scripts/validate.py now
  fails whenever the shipped command set differs from what the plugin
  version admits; the policy map lives in shared scripts/_checks.py and
  scripts/doctor.py derives its expectations from the same source.
• OPP-21 — repeat-blocker counting. scripts/aggregate_runs.py
  counts a repeat blocker per distinct run, not per ledger row.
• Release-checklist gate-1 count correction and ADR/ticket cleanups.

Validation

• scripts/release.py gate green (tree / versions / validate / seam /
  adapter floor); scripts/doctor.py DOCTOR PASSED; full local suites
  green; Ubuntu CI green on the source commit (b91879e).