GCW v1.5.0 - Quality Gates and Continuity

What's Changed

• Made Design DNA mandatory for every teardown depth and kept Web Shader Extractor mandatory for evidenced GPU targets, with explicit protection against unrelated skill substitution.
• Added evidence-before-candidate rules and repeated ready-state source captures, with phase-sensitive handling for dynamic regions.
• Added a formal quality-gate.json: open P0/P1/P2 blocks REVIEW_GATE, while disclosed P3 may pass.
• Added bounded continuity for large or multi-session work through one compact progress ledger, state validation, and user-authorized local checkpoints.
• Added base, teardown, and GPU environment profiles with Python 3.10+ validation and hermetic cross-platform tests.
• Added an internal dogfood record template for timing, discovery quality, intervention, and verification evidence.

Verification

• npm run verify: 26/26 smoke tests passed.
• npm run check and npm run check:gpu: passed.
• Branch CI passed on Ubuntu, Windows, and macOS: run 29184830042.
• Tag CI passed on Ubuntu, Windows, and macOS: run 29185226155.
• npm audit --registry=https://registry.npmjs.org: 0 vulnerabilities.

Full Changelog: v1.4.0...v1.5.0

GCW v1.4.0 - Editable Delivery Contracts

What's Changed

• Added an explicit A/B/C final-deliverable contract before build work, persisted in run-state schema v5.
• Added formal editability gates for maintainable source entrypoints, content replacement maps, controlled content changes, and runtime independence.
• Restricted ARTIFACT_REPLAY to oracle-only use for editable B/C delivery.
• Renamed the maintainable recovery strategy to MAINTAINABLE_REBUILD, with audited migration from legacy EDITABLE_REBUILD state.
• Made choice B a durable accepted checkpoint that can upgrade to C during review or resume Creative after completion.
• Closed delivery-contract mismatch, teardown-completion, and Creative-closeout gate bypasses.

Verification

• npm run verify: 25/25 smoke tests passed.
• Windows, Ubuntu, and macOS GitHub Actions verification passed on the release commit.
• npm audit: 0 vulnerabilities.

Full Changelog: v1.3.0...v1.4.0

GCW v1.3.0 — Evidence Capture and Offline Replay

What's changed

Added

• Source Map evidence from response headers, sourceMappingURL comments, and conventional probes, with bounded reads and semantic v3 validation.
• Explicit opt-in SPA HAR recording with credential and request-body redaction, origin rebasing, deterministic offline replay, and runtime-independence evidence.
• Review-gated interaction-state detection for hover, focus, and common aria-expanded controls, including before/after screenshots.
• Deterministic asset manifest drafts generated from inventory data without overwriting reviewed files; downloads remain blocked until human approval.

Fixed

• Source Map probes now reject SPA fallback HTML, incomplete JSON, malformed regular maps, and invalid index-map sections.

Internal

• Added macOS to the Windows/Ubuntu CI matrix.
• Upgraded pinned GitHub Actions to Node 24 runtimes.
• Expanded the release smoke suite to 21 tests.

Full Changelog: v1.2.1...v1.3.0

GCW v1.2.1 — Reliability and Performance Hardening

What's changed

Added

• Teardown depth profiles: minimal, standard, and deep.
• Machine-validated interaction-state schema, screenshot decoding, companion schema anchors, and review-decision history.
• Configurable inventory viewport plus generated route-map and network evidence.
• Public ADR for the staged workflow and conditional gates.

Changed

• Image comparison now uses Pillow-native operations and one shared implementation.
• Capture scenarios use one canonical template and default to a light color scheme.
• Environment reporting separates base runtime readiness from teardown companion readiness.
• Skill commands resolve through <skill-root> and recognize both AGENTS.md and CLAUDE.md workspace instructions.

Fixed

• Workflow transitions no longer fabricate a creative brief and now enforce CLONE_REPORT.md plus explicit A/B/C decisions.
• Teardown finalization validates all inputs before writing final state.
• Runtime-independence checks normalize default ports and inspect both network evidence and final build output.
• Asset downloads report per-item failures and reject every cross-origin redirect hop.
• Installed visual-regression CI invokes Playwright from .gcw and warns when non-Vite build commands need adaptation.

Full Changelog: v1.2.0...v1.2.1

GCW v1.2.0 — Evidence-Orchestrated Staged Workflow

What changed

• Replaced the four parallel modes with a staged workflow: TEARDOWN_PHASE -> FAITHFUL_CLONE -> REVIEW_GATE -> CREATIVE_REBUILD.
• Kept PRODUCTION_RECOVERY as an ownership-gated recovery profile rather than a competing mode.
• Made teardown useful as a standalone research outcome, without requiring a clone or rebuild.

Evidence orchestration

• Runs design-dna for every teardown, before SITE_SPEC.md is finalized.
• Requires web-shader-extractor when Canvas, WebGL, WebGPU, or shader signals are found; otherwise records an evidence-backed N/A.
• Preserves each sibling skill's native artifacts while GCW coordinates them through SITE_SPEC.md, teardown-manifest.json, and evidence-index.json.
• Adds ownership and SHA-256 tracking so evidence remains attributable and verifiable without duplicating schemas.

Truth and fidelity gates

• Validates SOURCE, PARTIAL, and GUESS claims against cited evidence.
• Requires subsystem fidelity decisions before teardown finalization.
• Requires GPU investigations to reach TARGET_LOCKED and REPLAY_READY before the specification can be finalized.
• Keeps SITE_SPEC.md as the single human-readable implementation specification and the manifest as the machine gate.

Documentation and compatibility

• Updated the Skill, bilingual READMEs, references, templates, scripts, and CI fixtures for v1.2.0.
• Existing Design DNA and shader-analysis artifacts remain native and are referenced rather than rewritten, including compatibility validation against the Learnhaoqi evidence set.
• Existing v1.1 project inputs remain supported through normalized defaults and migration guidance.

Verification

• 12/12 full verification tests passed locally.
• Official Skill validation passed.
• Windows and Ubuntu GitHub Actions jobs passed.
• Production dependency audit reported 0 vulnerabilities.

Full Changelog: v1.1.0...v1.2.0

GCW v1.1.0 — Evidence-Driven Website Cloning

What changed

• Repositioned GCW around website learning, technical teardown, faithful cloning, creative reconstruction, and production recovery.
• Added four explicit modes: TEARDOWN, FAITHFUL_CLONE, CREATIVE_REBUILD, and PRODUCTION_RECOVERY.
• Kept ordinary static, content, and multi-page websites as first-class use cases while preserving advanced WebGL, Canvas, shader, motion, and deterministic visual comparison workflows.
• Moved production recovery tiers and baked GLTF/GLB text replacement into focused, opt-in guidance.

Reliability and safety

• Rejects credential-bearing URLs and sensitive query parameters.
• Blocks document redirects and route navigation that leave the configured origin.
• Redacts sensitive URLs from manifests and error output.
• Detects Canvas contexts without actively probing or replacing an existing rendering context.
• Validates screenshot phase timing, scenario ID collisions, image thresholds, and non-destructive CI installation.
• Pins Playwright, Pillow, and GitHub Actions for reproducible runs.

Verification

• 7/7 release smoke tests passed with real Chromium.
• Windows and Ubuntu GitHub Actions jobs passed.
• Official Skill validation passed.
• Production dependency audit reported 0 vulnerabilities.
• Worktree and Git history secret scans reported 0 matches.

Visual assets

The bilingual READMEs now include the updated GCW cloning banner and workflow illustration. A compressed 2:1 Social Preview is also available in assets/social-preview.jpg.

Optional capability

Blender is only required for the specialized GLTF/GLB baked-text replacement workflow and was not part of this release verification.