@design-ai/cli 4.55.0 → 4.57.0

package/docs/ROADMAP.md

@@ -1,5 +1,2229 @@
 # Roadmap
 
+## Phase 755 — Local Retrieval Memory Release (v4.57.0) ✓ ready
+
+Groups the Phase 754 A+B retrieval work, the MCP ranked-search parameter, the `refs/` reference-page migration, and the docs-deploy retry hardening into one public package release. Scope and data boundaries are defined in [AI-LEARNING-PHASE2.md](AI-LEARNING-PHASE2.md); defaults are unchanged, so the release is additive and backward-compatible.
+
+### Changed
+- Shipped deterministic local retrieval (`design-ai index`, `search --ranked`, shared BM25 learning selection), Hangul-aware tokenization, and the version-2 source-keyed index sidecar.
+- Shipped the opt-in, local-only embedding rerank backend (`search --ranked --embeddings`, `~/.design-ai/config.json`) with graceful lexical degradation and no network path.
+- Added the MCP `design_ai_search` opt-in `ranked` parameter and the `design_ai_prompt` / `design_ai_pack` opt-in `withRecall` parameters, moved `refs/` links behind `docs/reference/` pages, and hardened `Deploy doc site` with a single automatic retry.
+- Added `prompt --with-recall` / `pack --with-recall`: brief-relevant corpus knowledge recalled into prompts/packs via the shared lexical scorer (opt-in; default output unchanged) — the first CLI-deepening increment from [NEXT-SURFACE-DECISION.md](NEXT-SURFACE-DECISION.md).
+- Added `design-ai learn --recall <query>`, the read-side companion completing the retrieval verb pair (`learn --recall` + `pack --with-recall`) named in [NEXT-SURFACE-DECISION.md](NEXT-SURFACE-DECISION.md): a read-only combined corpus + learning recall view.
+- Extended packed-tarball and public-registry smoke to cover the index build/status/verify round-trip, ranked-search determinism, and embeddings off-by-default fallback.
+
+### Verified
+- All 8 audits passed.
+- `npm run release:check` (unit tests, strict audits, whitespace, package contents, release metadata, release self-tests, packed-tarball smoke).
+- `npm run release:metadata`.
+- `git diff --check`.
+- Main-branch GitHub Actions (`Design-AI audit`, `Deploy doc site`) passed for the constituent Phase 754 commits.
+
+### Versions
+- `package.json` + `.claude-plugin/plugin.json`: 4.56.0 → 4.57.0.
+- `vscode-extension/package.json`: remains 0.4.1.
+
+### What this enables
+- Users get deterministic ranked corpus/learning retrieval (Korean included) and an optional local embedding rerank, with zero new dependencies and no behavior change unless they opt in.
+- Maintainers can publish a retrieval-capable release whose new CLI, MCP, and workspace surfaces are smoke-covered on both the packed tarball and the public registry.
+
+### What's still ahead
+- Push the v4.57.0 release commit and tag, then verify npm publish (provenance), GitHub Release, public `npm run registry:smoke`, and docs deployment workflows.
+- Refresh the Homebrew formula SHA-256 after the v4.57.0 tag tarball is available, then update [external-status.md](external-status.md) and the README distribution status from published `4.56.0` to `4.57.0`.
+
+## Phase 754 — AI learning depth: local retrieval memory (planning)
+
+Opens the deeper AI-learning phase chosen in Product Readiness ("Recommended next decision", option 2): deterministic local retrieval memory over the shipped knowledge corpus and the local learning profile, with an optional local-only embedding backend. Scope, data boundaries, and privacy constraints are defined in [AI-LEARNING-PHASE2.md](AI-LEARNING-PHASE2.md). Fine-tuning, training on user artifacts, external embedding APIs, background learning without explicit CLI commands, and telemetry stay out of scope; the README stance ("Not a model. Not a fine-tune.") is unchanged.
+
+### Plan
+- [x] Phase A: zero-dependency lexical retrieval index (tokenized BM25-style scoring) over `knowledge/`, `examples/`, `skills/`, `docs/`, `agents/`, `commands/`, and `learning.json`, built only by an explicit `design-ai index --build`.
+- [x] Phase A: opt-in `design-ai search --ranked` ranked corpus search with deterministic, fully ordered output.
+- [x] Phase A: shared lexical scorer behind `prompt --with-learning` / `pack --with-learning` selection, gated by `learn --eval --strict` checkpoints before and after. _Selection-drift callout: tf-aware BM25 ranking can reorder entries that the old containment scorer tied (text-relevant entries now outrank category-only matches); checkpoints record `ranker: "lexical"` so eval templates regenerate against the shared scorer._
+- [x] Phase A: `design-ai index --status/--verify` staleness and digest reporting, plus `workspace` readiness integration.
+- [x] Phase A: verification gates — `npm test` unit coverage, `npm run audit` 8/8, `npm run release:metadata`, and `release:check` packed-tarball smoke for index build/verify round-trip and ranked-search determinism.
+- [x] Phase B (optional, after Phase A review): opt-in local embedding rerank via a user-supplied local provider executable (`~/.design-ai/config.json` and/or `--provider`), never a default, never external HTTP, with graceful degradation to Phase A on any absence/failure/staleness.
+- [x] Phase B: deterministic stub-provider test coverage (`cli/lib/embedding-provider.test.mjs`, `cli/lib/embedding-index.test.mjs`, `cli/lib/embedding-rerank.test.mjs`, `cli/lib/search-ranked.test.mjs`) and `package:smoke` coverage proving off-by-default, missing-provider fallback, and a real stub-provider round-trip.
+- [x] Answer the open questions in [AI-LEARNING-PHASE2.md](AI-LEARNING-PHASE2.md) (ranked-search default, index location, route boundary, Korean tokenization, Phase B configuration home) before scheduling Phase B. _Answered in the Phase A implementation review; Hangul tokenization (FU-1) and index source-identity keying (FU-2) landed before Phase B._
+
+## Phase 753 — MCP Protocol Hardening Release (v4.56.0) ✓ ready
+
+The next public package release groups the post-v4.55 MCP protocol hardening work and the published MCP client evidence refresh into one versioned handoff.
+
+### Changed
+- Hardened JSON-RPC envelope, method, id, params, notification, initialize, and tools/call validation paths so malformed MCP requests fail before CLI execution.
+- Preserved existing Claude Code, Codex, Website Improvement, package smoke, and registry smoke contracts while making invalid inputs easier to diagnose.
+- Refreshed public npm MCP entrypoint evidence and local Claude/Codex MCP client connection evidence.
+
+### Verified
+- All 8 audits passed.
+- `npm run release:check`.
+- `npm run release:metadata`.
+- `npm run audit:strict`.
+- `git diff --check`.
+- Main-branch GitHub Actions passed for `Design-AI audit` and `Deploy doc site` after the MCP evidence refresh.
+
+### Versions
+- `package.json` + `.claude-plugin/plugin.json`: 4.55.0 → 4.56.0.
+- `vscode-extension/package.json`: remains 0.4.1.
+
+### What this enables
+- Claude Code and Codex users get deterministic MCP protocol errors for malformed requests while valid local stdio tool calls stay unchanged.
+- Maintainers can publish the hardened MCP server with current evidence for public npm and local client connection paths.
+
+### What's still ahead
+- Push the v4.56.0 release commit and tag, then verify npm publish, GitHub Release, registry smoke, and docs deployment workflows.
+- Refresh Homebrew formula SHA-256 after the v4.56.0 tag tarball is available.
+
+## Phase 752 — Published MCP Client Connection Evidence Refresh (unreleased)
+
+The external distribution status now includes current proof that the published npm MCP entrypoint and local Claude Code / Codex MCP client registrations are usable.
+
+### Changed
+- Refreshed npm registry and VS Code Marketplace status evidence for the current published versions.
+- Added public npm `design-ai-mcp` protocol-smoke evidence from a clean temporary directory.
+- Added local Claude Code and Codex MCP client status evidence for the clone-backed `design-ai` server.
+- Documented the npm one-shot verification gotcha when running from inside the package source checkout.
+
+### Impact
+- Operators can distinguish already-published package state from actions that would fail by republishing the same version.
+- Claude Code and Codex setup guidance now has current evidence for both client connection and public npm entrypoint behavior.
+- No runtime code, package version, public API, or release workflow behavior changed.
+
+### Verification Plan
+- `npm view @design-ai/cli version name time.modified dist-tags --json`
+- VS Code Marketplace Gallery API query for `sungjin.design-ai-vscode`
+- Public npm MCP protocol smoke with `npm exec --yes --package=@design-ai/cli@4.55.0 -- design-ai-mcp`
+- `codex mcp get design-ai`
+- `claude mcp list`
+- `git diff --check`
+
+## Phase 751 — design-ai MCP Initialize Blank Protocol Version Guard (unreleased)
+
+The MCP stdio server now rejects blank-only `initialize.params.protocolVersion` strings with JSON-RPC `-32602`, so version negotiation inputs must be meaningful strings when provided.
+
+### Changed
+- Added a blank-string guard for `initialize.params.protocolVersion`.
+- Added unit coverage for whitespace-only protocol versions.
+- Preserved fallback behavior for unsupported but non-empty protocol version strings.
+
+### Impact
+- Claude Code, Codex, and other MCP clients get clear invalid-params feedback for malformed initialize version negotiation.
+- Valid initialize requests, omitted initialize params, unsupported non-empty string protocol versions, notifications, ping, tools/list, tools/call, resources/list, prompts/list, package smoke, Website Improvement readiness behavior, and CLI runtime failure reporting stay unchanged.
+
+### Verification Plan
+- `node --check cli/lib/mcp-server.mjs`
+- `node --test cli/lib/mcp-server.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `npm run package:smoke:self-test`
+- `npm run release:check`
+- `git diff --check`
+
+## Phase 750 — design-ai MCP Blank String Argument Guard (unreleased)
+
+The MCP stdio server now rejects blank-only string arguments for schema fields that declare `minLength`, so required text inputs fail as JSON-RPC `-32602` before CLI invocation instead of surfacing later as tool execution errors.
+
+### Changed
+- Added whitespace-only validation for MCP string inputs with `minLength`.
+- Added unit coverage proving a blank `design_ai_route.brief` is rejected before the CLI runner is called.
+- Kept optional string fields without `minLength` on their existing CLI-mapping path.
+
+### Impact
+- Claude Code, Codex, and other MCP clients get immediate invalid-params feedback for blank required text fields.
+- Valid non-empty strings, optional string omission, initialize, notifications, ping, tools/list, tools/call success, resources/list, prompts/list, package smoke, Website Improvement readiness behavior, and CLI runtime failure reporting stay unchanged.
+
+### Verification Plan
+- `node --check cli/lib/mcp-server.mjs`
+- `node --test cli/lib/mcp-server.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `npm run package:smoke:self-test`
+- `npm run release:check`
+- `git diff --check`
+
+## Phase 749 — design-ai MCP Notification Namespace Guard (unreleased)
+
+The MCP stdio server now treats the whole `notifications/` method namespace consistently: notification messages must not include request ids, and valid id-less notifications are ignored without writing a response.
+
+### Changed
+- Added a small notification-method predicate for the `notifications/` namespace.
+- Generalized the request-id rejection from `notifications/initialized` to every `notifications/*` method.
+- Preserved silent handling for valid id-less notifications, including unsupported notification names that the local server does not need to act on.
+- Added unit and subprocess coverage for `notifications/progress` with and without ids.
+
+### Impact
+- Claude Code, Codex, and other MCP clients get deterministic `-32600` feedback when a notification is accidentally sent as a request.
+- Valid initialized notifications, progress-style notifications without ids, initialize, ping, tools/list, tools/call, resources/list, prompts/list, package smoke, Website Improvement readiness behavior, and CLI runtime failure reporting stay unchanged.
+
+### Verification Plan
+- `node --check cli/lib/mcp-server.mjs`
+- `node --test cli/lib/mcp-server.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `npm run package:smoke:self-test`
+- `npm run release:check`
+- `git diff --check`
+
+## Phase 748 — design-ai MCP Initialize Protocol Version Type Guard (unreleased)
+
+The MCP stdio server now rejects non-string `initialize.params.protocolVersion` values with JSON-RPC `-32602`, so malformed initialize payloads get the same explicit invalid-params handling as other MCP request parameter mistakes.
+
+### Changed
+- Added a focused initialize params validator that keeps optional params supported while checking `protocolVersion` when provided.
+- Added unit coverage for numeric `protocolVersion` values so the server no longer silently falls back when the client sends the wrong JSON type.
+
+### Impact
+- Claude Code, Codex, and other MCP clients get deterministic feedback when initialize version negotiation is malformed.
+- Valid initialize requests, omitted initialize params, unsupported string protocol versions, initialized notifications, ping, tools/list, tools/call, resources/list, prompts/list, package smoke, Website Improvement readiness behavior, and CLI runtime failure reporting stay unchanged.
+
+### Verification Plan
+- `node --check cli/lib/mcp-server.mjs`
+- `node --test cli/lib/mcp-server.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `npm run package:smoke:self-test`
+- `npm run release:check`
+- `git diff --check`
+
+## Phase 747 — design-ai MCP Tool Argument Invalid Params Guard (unreleased)
+
+The MCP stdio server now returns JSON-RPC `-32602` for schema-invalid `tools/call.params.arguments` payloads before CLI execution, so known-tool argument errors use the same invalid-params response shape as malformed `tools/call.params`.
+
+### Changed
+- Added pre-execution tool schema argument validation inside the `tools/call` request branch.
+- Converted unknown tool arguments and wrong argument types from successful MCP tool-result errors to JSON-RPC invalid params errors.
+- Preserved CLI failure reporting as successful MCP tool results with `isError: true`.
+- Updated package and registry smoke assertions so publish-time MCP protocol checks verify the JSON-RPC invalid-params response shape.
+
+### Impact
+- Claude Code, Codex, and other MCP clients can handle known-tool schema mistakes through standard JSON-RPC invalid-params handling.
+- Valid initialize, id-less initialized notifications, ping, tools/list, resources/list, prompts/list, id-bearing tools/call, request id validation, output truncation, Website Improvement readiness behavior, external MCP call boundaries, target-repo mutation rules, learning write behavior, and CLI runtime failure reporting stay unchanged.
+- Package and registry smoke coverage now treats schema-invalid known-tool arguments as JSON-RPC invalid params instead of successful MCP tool-result errors.
+
+### Verification Plan
+- `node --check cli/lib/mcp-server.mjs`
+- `node --test cli/lib/mcp-server.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `npm run package:smoke:self-test`
+- `npm run release:check`
+- `git diff --check`
+
+## Phase 746 — design-ai MCP Response Request ID Guard (unreleased)
+
+The MCP stdio server now requires request ids for every response-bearing MCP method, so `initialize`, `ping`, `tools/list`, `tools/call`, `resources/list`, and `prompts/list` cannot be sent as silent notifications.
+
+### Changed
+- Added an explicit response-method set for MCP methods that must include a request id.
+- Reused the same request-id guard across initialize, ping, list methods, and tools/call.
+- Added unit and subprocess coverage for id-less response-bearing MCP methods while preserving silent handling for valid initialized notifications.
+
+### Impact
+- Claude Code, Codex, and other MCP clients get deterministic feedback when response-bearing MCP requests are accidentally sent without correlation ids.
+- Valid initialize, id-less initialized notifications, ping, tools/list, resources/list, prompts/list, id-bearing tools/call, request id validation, argument validation, output truncation, package smoke, Website Improvement readiness behavior, external MCP call boundaries, target-repo mutation rules, and learning write behavior stay unchanged.
+
+### Verification Plan
+- `node --check cli/lib/mcp-server.mjs`
+- `node --test cli/lib/mcp-server.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `npm run package:smoke:self-test`
+- `npm run release:check`
+- `git diff --check`
+
+## Phase 745 — design-ai MCP Tool Call Request ID Guard (unreleased)
+
+The MCP stdio server now rejects `tools/call` payloads that omit a request id before any CLI work starts, so request-shaped tool execution cannot run silently as a JSON-RPC notification.
+
+### Changed
+- Added a `tools/call` request-id requirement guard before tool lookup or CLI invocation.
+- Preserved normal `tools/call` behavior for valid string, number, and null ids.
+- Added unit and subprocess coverage proving id-less `tools/call` receives `-32600` feedback and does not execute the CLI.
+
+### Impact
+- Claude Code, Codex, and other MCP clients get deterministic feedback when a tool invocation is accidentally sent without a correlation id.
+- Valid initialize, id-less initialized notifications, ping, tools/list, resources/list, prompts/list, id-bearing tools/call, request id validation, argument validation, output truncation, package smoke, Website Improvement readiness behavior, external MCP call boundaries, target-repo mutation rules, and learning write behavior stay unchanged.
+
+### Verification Plan
+- `node --check cli/lib/mcp-server.mjs`
+- `node --test cli/lib/mcp-server.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `npm run package:smoke:self-test`
+- `npm run release:check`
+- `git diff --check`
+
+## Phase 744 — design-ai MCP Optional Params Container Guard (unreleased)
+
+The MCP stdio server now validates optional `params` containers for `ping`, `tools/list`, `resources/list`, and `prompts/list`, so malformed array, string, null, or boolean params receive deterministic `-32602` feedback instead of being silently accepted.
+
+### Changed
+- Added a shared optional object-params guard for no-body MCP methods that still accept future object-shaped params.
+- Preserved successful handling for omitted params and `{}` params on `ping` and `tools/list`.
+- Added unit coverage for malformed optional params on `ping`, `tools/list`, `resources/list`, and `prompts/list`.
+
+### Impact
+- Claude Code, Codex, and other MCP clients get clearer invalid-params feedback when list or ping requests carry malformed params.
+- Valid initialize, id-less initialized notifications, ping, tools/list, resources/list, prompts/list, tools/call, request id validation, argument validation, output truncation, package smoke, Website Improvement readiness behavior, external MCP call boundaries, target-repo mutation rules, and learning write behavior stay unchanged.
+
+### Verification Plan
+- `node --check cli/lib/mcp-server.mjs`
+- `node --test cli/lib/mcp-server.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `npm run package:smoke:self-test`
+- `npm run release:check`
+- `git diff --check`
+
+## Phase 743 — design-ai MCP Initialized Notification ID Guard (unreleased)
+
+The MCP stdio server now rejects `notifications/initialized` payloads that include an `id`, so request-shaped initialized notifications receive deterministic `-32600` feedback instead of being silently dropped.
+
+### Changed
+- Added a notification-shape guard for `notifications/initialized`.
+- Preserved silent handling for valid id-less initialized notifications.
+- Added unit and subprocess coverage for request-shaped initialized notification payloads.
+
+### Impact
+- Claude Code, Codex, and other MCP clients get a clear protocol error when an initialized notification is accidentally sent as a request.
+- Valid id-less notifications, initialize, tools/list, tools/call, request id validation, argument validation, output truncation, package smoke, Website Improvement readiness behavior, external MCP call boundaries, target-repo mutation rules, and learning write behavior stay unchanged.
+
+### Verification Plan
+- `node --check cli/lib/mcp-server.mjs`
+- `node --test cli/lib/mcp-server.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `npm run package:smoke:self-test`
+- `npm run release:check`
+- `git diff --check`
+
+## Phase 742 — design-ai MCP Request ID Guard (unreleased)
+
+The MCP stdio server now validates JSON-RPC request ids before dispatch, so malformed object, array, or boolean ids receive deterministic `-32600` feedback with `id: null` instead of being echoed back to clients.
+
+### Changed
+- Added a JSON-RPC request id guard for `string`, `number`, and `null` ids.
+- Normalized invalid request ids to `id: null` in protocol error responses.
+- Added unit and subprocess coverage for invalid object, array, and boolean id payloads while preserving explicit `id: null` requests.
+
+### Impact
+- Claude Code, Codex, and other MCP clients get clearer protocol feedback when request correlation ids are malformed.
+- Valid string, number, and null ids, notifications, initialize, tools/list, tools/call, argument validation, output truncation, package smoke, Website Improvement readiness behavior, external MCP call boundaries, target-repo mutation rules, and learning write behavior stay unchanged.
+
+### Verification Plan
+- `node --check cli/lib/mcp-server.mjs`
+- `node --test cli/lib/mcp-server.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `npm run package:smoke:self-test`
+- `npm run release:check`
+- `git diff --check`
+
+## Phase 741 — design-ai MCP JSON-RPC Envelope Guard (unreleased)
+
+The MCP stdio server now rejects requests without `jsonrpc: "2.0"` before method dispatch, so malformed clients cannot reach normal initialize/list/call handlers with an invalid JSON-RPC envelope.
+
+### Changed
+- Added a JSON-RPC envelope guard before MCP method validation.
+- Preserved deterministic `-32600` responses for id-less malformed requests.
+- Added unit and subprocess coverage for missing and unsupported `jsonrpc` payloads.
+
+### Impact
+- Claude Code, Codex, and other MCP clients get clearer protocol feedback when a malformed JSON-RPC envelope reaches the stdio server.
+- Valid JSON-RPC 2.0 initialize, notifications, tools/list, tools/call, argument validation, output truncation, package smoke, Website Improvement readiness behavior, external MCP call boundaries, target-repo mutation rules, and learning write behavior stay unchanged.
+
+### Verification Plan
+- `node --check cli/lib/mcp-server.mjs`
+- `node --test cli/lib/mcp-server.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `npm run package:smoke:self-test`
+- `npm run release:check`
+- `git diff --check`
+
+## Phase 740 — design-ai MCP Request Method Shape Guard (unreleased)
+
+The MCP stdio server now treats non-string or empty request methods as JSON-RPC invalid requests, so malformed notification-like payloads without ids still receive deterministic `-32600` feedback while valid notifications remain silent.
+
+### Changed
+- Added a request-method shape guard before MCP method dispatch.
+- Updated stdio response writing so id-less `-32600` invalid requests are emitted without also responding to valid unknown-method notifications.
+- Added unit and subprocess coverage for numeric and empty `method` payloads.
+
+### Impact
+- Claude Code, Codex, and other MCP clients get clearer protocol feedback when a malformed request method reaches the stdio server.
+- Valid notifications, unknown-method notifications without ids, initialize, tools/list, tools/call, argument validation, output truncation, package smoke, Website Improvement readiness behavior, external MCP call boundaries, target-repo mutation rules, and learning write behavior stay unchanged.
+
+### Verification Plan
+- `node --check cli/lib/mcp-server.mjs`
+- `node --test cli/lib/mcp-server.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `npm run package:smoke:self-test`
+- `npm run release:check`
+- `git diff --check`
+
+## Phase 739 — design-ai MCP Tool Arguments Container Guard (unreleased)
+
+The MCP server now rejects malformed `tools/call.params.arguments` containers before tool lookup or CLI execution, so omitted arguments still default to `{}` while explicit `null`, array, or primitive argument payloads return JSON-RPC `-32602` feedback.
+
+### Changed
+- Added a `tools/call.params.arguments` object-shape check to the MCP request boundary.
+- Preserved omitted `arguments` support for no-argument tools such as `design_ai_version`.
+- Added unit coverage for `arguments: null` and array argument payloads.
+
+### Impact
+- Claude Code, Codex, and other MCP clients get clearer protocol feedback when a tool call includes a malformed arguments container.
+- Normal `tools/call` execution, tool-schema argument validation, unknown-tool handling, initialize, notifications, tools/list, package smoke, Website Improvement readiness behavior, external MCP call boundaries, target-repo mutation rules, and learning write behavior stay unchanged.
+
+### Verification Plan
+- `node --check cli/lib/mcp-server.mjs`
+- `node --test cli/lib/mcp-server.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `npm run package:smoke:self-test`
+- `npm run release:check`
+- `git diff --check`
+
+## Phase 738 — design-ai MCP Invalid Request Stdio Response (unreleased)
+
+The MCP stdio server now distinguishes valid notifications from malformed request objects without ids, so invalid JSON-RPC payloads such as `{}`, `[]`, or `null` receive deterministic `-32600` responses with `id: null` while normal `notifications/initialized` messages remain silent.
+
+### Changed
+- Added a small response-write predicate for MCP stdio output.
+- Preserved silent handling for valid JSON-RPC notifications.
+- Added subprocess coverage for invalid request objects without ids.
+
+### Impact
+- Claude Code, Codex, and other MCP clients get clearer protocol feedback when a malformed payload reaches the stdio server.
+- Valid initialize requests, notifications, tools/list, tools/call, CLI argument mapping, output truncation, Website Improvement readiness behavior, external call boundaries, target-repo mutation rules, and learning write behavior stay unchanged.
+
+### Verification Plan
+- `node --check cli/lib/mcp-server.mjs`
+- `node --test cli/lib/mcp-server.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `git diff --check`
+
+## Phase 737 — design-ai MCP Request Param Guard (unreleased)
+
+The MCP server now rejects malformed `initialize` and `tools/call` request parameters at the protocol boundary, so Claude Code and Codex clients get clear JSON-RPC `-32602` errors instead of less specific request handling or internal exceptions.
+
+### Changed
+- Added a small MCP request-boundary guard for `tools/call` params.
+- Added a matching MCP request-boundary guard for `initialize` params.
+- Added unit coverage for malformed `initialize` params, `tools/call` `params: null`, and missing `params.name`.
+
+### Impact
+- Malformed MCP clients receive deterministic protocol errors instead of falling into less specific request handling.
+- Valid tool calls, tool schemas, CLI argument mapping, output truncation, protocol version fallback, Website Improvement readiness behavior, external call boundaries, target-repo mutation rules, and learning write behavior stay unchanged.
+
+### Verification Plan
+- `node --test cli/lib/mcp-server.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `git diff --check`
+
+## Phase 736 — design-ai MCP Test Subprocess Readability (unreleased)
+
+MCP stdio subprocess tests now share one small response collector, so parse-error and route-call smoke tests read as protocol scenarios instead of repeated stream plumbing.
+
+### Changed
+- Added a focused MCP subprocess test helper for stdout line buffering and JSON-RPC response collection.
+- Reused the helper in stdio parse-error and initialize/list/route subprocess tests.
+- Kept timeout cleanup explicit so failed subprocess tests do not leave a child process running.
+
+### Impact
+- MCP test code is easier to scan while preserving the same runtime behavior and assertions.
+- No MCP server implementation, protocol response shape, tool schema, CLI mapping, package contents, Website Improvement readiness output, external call boundary, or learning write behavior changes.
+
+### Verification Plan
+- `node --test cli/lib/mcp-server.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `git diff --check`
+
+## Phase 735 — design-ai MCP Negative Response Coverage (unreleased)
+
+The MCP server test suite now pins common client failure paths, so malformed Claude/Codex requests keep returning deterministic JSON-RPC errors without starting the design-ai CLI unnecessarily.
+
+### Changed
+- Added coverage for unknown MCP tool calls to prove the CLI runner is not invoked.
+- Added request-level error coverage for missing and unsupported MCP methods.
+- Added stdio subprocess coverage for invalid JSON input and parse-error responses.
+
+### Impact
+- Claude Code and Codex integrations get stronger regression coverage for setup mistakes and malformed client messages.
+- The existing MCP runtime behavior, tool names, schemas, protocol version fallback, CLI command mappings, output truncation, and typed tool argument validation stay unchanged.
+- No package names, Website Improvement readiness outputs, external MCP calls, target-repo mutation rules, or learning writes change.
+
+### Verification Plan
+- `node --test cli/lib/mcp-server.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `git diff --check`
+- `npm run release:self-test`
+- `npm run release:check`
+
+## Phase 734 — design-ai MCP Public Registry Smoke Coverage (unreleased)
+
+Public registry smoke now verifies the published `design-ai-mcp` bin through `npm exec --package @design-ai/cli@<version>`, so the post-publish gate covers the same Claude Code and Codex stdio server path as the packed-tarball release smoke.
+
+### Changed
+- Moved the MCP protocol smoke input and response assertion into shared smoke helpers.
+- Added public registry `design-ai-mcp` protocol coverage after `npm exec` version metadata checks.
+- Added registry smoke self-test coverage for the MCP server name, expected tools, and typed invalid-argument response.
+
+### Impact
+- Public npm verification now catches a missing `design-ai-mcp` bin, broken stdio startup, missing design-ai tools, or lost typed validation after publish.
+- Package smoke and registry smoke share the same MCP protocol assertion for Claude/Codex setup paths.
+- No CLI command names, package names, Website Improvement readiness outputs, external MCP calls, target-repo mutation rules, or learning writes change.
+
+### Verification Plan
+- `python3 -m py_compile tools/audit/smoke_assertions.py tools/audit/package-smoke.py tools/audit/registry-smoke.py`
+- `python3 -B tools/audit/package-smoke.py --self-test`
+- `python3 -B tools/audit/registry-smoke.py --self-test`
+- `npm run package:smoke:self-test`
+- `npm run registry:smoke:self-test`
+- `npm run release:self-test`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+- `npm run release:check`
+
+## Phase 733 — design-ai MCP Package Entrypoint Smoke Coverage (unreleased)
+
+Packed-tarball release smoke now verifies the standalone `design-ai-mcp` bin after package installation and through one-shot `npm exec`, so Claude Code and Codex MCP setup cannot silently lose the stdio server entrypoint.
+
+### Changed
+- Added a package-smoke MCP protocol helper that sends `initialize`, `tools/list`, and invalid `tools/call` JSON-RPC messages over stdin.
+- Verified the installed `node_modules/.bin/design-ai-mcp` shim and the one-shot `npm exec --package <tarball> -- design-ai-mcp` path.
+- Added self-test coverage for the MCP protocol assertion so typed argument validation remains visible in package smoke failures.
+
+### Impact
+- Release smoke now covers the actual MCP server bin that Claude Code and Codex configuration examples use.
+- The smoke keeps the same protocol version, tool names, JSON-RPC response shape, and typed validation message checked by unit tests.
+- No CLI commands, package names, Website Improvement readiness outputs, external MCP calls, target-repo mutation rules, or learning writes change.
+
+### Verification Plan
+- `python3 -B tools/audit/package-smoke.py --self-test`
+- `npm run package:smoke:self-test`
+- `npm run package:check`
+- `npm run release:self-test`
+- `npm run release:check`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+
+## Phase 732 — design-ai MCP Tool Argument Validation (unreleased)
+
+The design-ai MCP server now validates tool arguments before invoking the CLI, so Claude Code and Codex get immediate, typed feedback for malformed tool calls instead of opaque downstream command failures.
+
+### Changed
+- Added MCP tool argument validation for required fields, unknown fields, string length, boolean values, and integer ranges.
+- Added MCP server tests proving invalid arguments are rejected before the CLI runner is called.
+- Added troubleshooting notes for common Claude/Codex MCP setup and tool-call failures.
+
+### Impact
+- `design-ai mcp` and `design-ai-mcp` keep the same tool names, schemas, protocol version, JSON-RPC response shape, and CLI command mappings.
+- Claude Code and Codex integrations now fail earlier and with clearer messages when a client sends malformed tool input.
+- No external MCP calls, target-repo mutation, learning writes, CLI flags, package bin names, or Website Improvement readiness contracts change.
+
+### Verification Plan
+- `node --check cli/lib/mcp-server.mjs`
+- `node --test cli/lib/mcp-server.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+- `npm run package:check`
+- `npm run release:self-test`
+- `npm run release:check`
+
+## Phase 731 — Website Improvement Bundle Handoff Next Stage Command Summary Snapshot Cleanup (unreleased)
+
+Website Improvement bundle handoff runbook next-stage command summary now groups stage classification, command metadata, and execution context into named internal snapshots before returning the same fields.
+
+### Changed
+- Added internal next-stage snapshots for stage classification, command metadata, and execution context.
+- Replaced the long inline `nextStageCommandSummary` assembly with ordered snapshot spreads while keeping the existing returned field names unchanged.
+
+### Impact
+- `buildBundleHandoffOperatorRunbook` keeps the same `nextStage*` command and execution contract.
+- Website Improvement handoff wrappers and tests can continue to consume the same next-stage kind, required flag, run policy, safety level, command lists, output files, command/manual state, and mutation boundary fields without migration.
+- No CLI flags, bundle files, command manifest fields, stage/action/evidence field names, validation status values, next-action keys, target-repo mutation rules, external-call boundaries, human runbook lines, or generated output files change.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-handoff-runbook.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+- `npm run package:check`
+- `npm run release:self-test`
+- `npm run release:check`
+
+## Phase 730 — Website Improvement Bundle Handoff Next Action Summary Snapshot Cleanup (unreleased)
+
+Website Improvement bundle handoff runbook next-action summary now groups action basics, evidence-capture metadata, payload mapping, validation state, field rules, counts, and run context into named internal snapshots before returning the same fields.
+
+### Changed
+- Added internal next-action snapshots for core action state, evidence-capture field metadata, payload templates and bindings, validation state, field rules, field counts, and run context.
+- Replaced the long inline `nextActionSummary` assembly with ordered snapshot spreads while keeping the existing returned field names unchanged.
+
+### Impact
+- `buildBundleHandoffOperatorRunbook` keeps the same `nextAction*` and `nextActionEvidenceCapture*` contract.
+- Website Improvement handoff wrappers and tests can continue to consume the same next-action status, dependency, evidence, evidence-capture, run policy, and safety fields without migration.
+- No CLI flags, bundle files, command manifest fields, stage/action/evidence field names, validation status values, next-action keys, target-repo mutation rules, external-call boundaries, human runbook lines, or generated output files change.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-handoff-runbook.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+- `npm run package:check`
+- `npm run release:self-test`
+- `npm run release:check`
+
+## Phase 729 — Website Improvement Bundle Handoff Next Stage Evidence Capture Snapshot Cleanup (unreleased)
+
+Website Improvement bundle handoff runbook next-stage evidence capture now groups field metadata, payload mapping, validation state, field rules, and field counts into named internal snapshots before returning the same fields.
+
+### Changed
+- Added internal next-stage evidence-capture snapshots for field metadata, payload templates and bindings, validation state, field rules, and field counts.
+- Replaced the long inline `nextStageEvidenceCapture` assembly with ordered snapshot spreads while keeping the existing returned field names unchanged.
+
+### Impact
+- `buildBundleHandoffOperatorRunbook` keeps the same `nextStageActionEvidenceCapture*` contract.
+- Website Improvement handoff wrappers and tests can continue to consume the same next-stage evidence-capture fields without migration.
+- No CLI flags, bundle files, command manifest fields, stage/action/evidence field names, validation status values, next-action keys, target-repo mutation rules, external-call boundaries, human runbook lines, or generated output files change.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-handoff-runbook.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+- `npm run package:check`
+- `npm run release:self-test`
+- `npm run release:check`
+
+## Phase 728 — Website Improvement Bundle Handoff Next Stage Human Line Summary Snapshot Cleanup (unreleased)
+
+Website Improvement bundle handoff runbook next-stage identity now prepares the next-stage human-line summary as a named internal snapshot before returning the same identity fields.
+
+### Changed
+- Added an internal `nextStageHumanLineSummary` snapshot for the next-stage line, evidence-progress status, progress label, and first unchecked evidence label.
+- Replaced the nested inline `nextStageHumanLineSummary` assembly inside `nextStageIdentity` with a direct snapshot reference while keeping the existing returned field names unchanged.
+
+### Impact
+- `buildBundleHandoffOperatorRunbook` keeps the same `nextStageHumanLineSummary` and `nextStageIdentity` contract.
+- Website Improvement handoff wrappers and tests can continue to consume the same next-stage human-line summary fields without migration.
+- No CLI flags, bundle files, command manifest fields, stage/action/evidence field names, validation status values, next-action keys, target-repo mutation rules, external-call boundaries, human runbook lines, or generated output files change.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-handoff-runbook.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+- `npm run package:check`
+- `npm run release:self-test`
+- `npm run release:check`
+
+## Phase 727 — Website Improvement Bundle Handoff Human Line Summary Snapshot Cleanup (unreleased)
+
+Website Improvement bundle handoff runbook human-line summary now groups count, evidence-progress, and first-value fields into named internal snapshots before returning the same summary fields.
+
+### Changed
+- Added internal human-line summary snapshots for base counts, evidence-progress counts, and first stage/line values.
+- Replaced the long inline `stageHumanLineSummary` assembly with snapshot spreads while keeping the existing summary field names unchanged.
+
+### Impact
+- `buildBundleHandoffOperatorRunbook` keeps the same `stageHumanLineSummary` contract.
+- Website Improvement handoff wrappers and tests can continue to consume the same human-line count, progress, and first-value summary fields without migration.
+- No CLI flags, bundle files, command manifest fields, stage/action/evidence field names, validation status values, next-action keys, target-repo mutation rules, external-call boundaries, human runbook lines, or generated output files change.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-handoff-runbook.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+- `npm run package:check`
+- `npm run release:self-test`
+- `npm run release:check`
+
+## Phase 726 — Website Improvement Bundle Handoff Human Line Display Row Summary Snapshot Cleanup (unreleased)
+
+Website Improvement bundle handoff runbook human-line display row summary now groups count, action-status, evidence-progress, and first-key fields into named internal snapshots before returning the same summary fields.
+
+### Changed
+- Added internal display-row summary snapshots for base counts, action-status counts, evidence-progress counts, and first matching row keys.
+- Replaced the long inline `stageHumanLineDisplayRowSummary` assembly with snapshot spreads while keeping the existing summary field names unchanged.
+
+### Impact
+- `buildBundleHandoffOperatorRunbook` keeps the same `stageHumanLineDisplayRowSummary` contract.
+- Website Improvement handoff wrappers and tests can continue to consume the same count, status, progress, and first-key summary fields without migration.
+- No CLI flags, bundle files, command manifest fields, stage/action/evidence field names, validation status values, next-action keys, target-repo mutation rules, external-call boundaries, human runbook lines, or generated output files change.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-handoff-runbook.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+- `npm run package:check`
+- `npm run release:self-test`
+- `npm run release:check`
+
+## Phase 725 — Website Improvement Bundle Handoff Human Line Display Row Snapshot Cleanup (unreleased)
+
+Website Improvement bundle handoff runbook human-line display rows now group row identity, execution context, action status, and evidence progress fields into named internal snapshots before returning the same display-row fields.
+
+### Changed
+- Added internal display-row snapshots for stage identity, execution context, action status, and evidence progress metadata.
+- Replaced the long inline `stageHumanLineDisplayRows` return assembly with snapshot spreads while keeping the existing display-row field names unchanged.
+
+### Impact
+- `buildBundleHandoffOperatorRunbook` keeps the same `stageHumanLineDisplayRows[*]` contract.
+- Website Improvement handoff wrappers can continue to consume the same stage line, required/manual, command count, action status, and evidence-progress display fields without migration.
+- No CLI flags, bundle files, command manifest fields, stage/action field names, evidence field names, validation status values, next-action keys, target-repo mutation rules, external-call boundaries, human runbook lines, or generated output files change.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-handoff-runbook.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+- `npm run package:check`
+- `npm run release:self-test`
+- `npm run release:check`
+
+## Phase 724 — Website Improvement Bundle Handoff Action Row Snapshot Cleanup (unreleased)
+
+Website Improvement bundle handoff runbook action rows now group action identity, status, dependency, evidence, evidence-capture, and execution fields into named internal snapshots before returning the same row fields.
+
+### Changed
+- Added internal action-row snapshots for action identity, status, dependency/completion metadata, and execution metadata.
+- Replaced the long inline action-row return assembly with snapshot spreads while keeping the existing action-row field names unchanged.
+
+### Impact
+- `buildBundleHandoffOperatorRunbook` keeps the same `stageActionRows[*]` contract.
+- Website Improvement handoff wrappers can continue to consume the same action identity, status, dependency, evidence, evidence-capture, command, safety, and boundary fields without migration.
+- No CLI flags, bundle files, command manifest fields, stage/action field names, evidence field names, validation status values, next-action keys, target-repo mutation rules, external-call boundaries, human runbook lines, or generated output files change.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-handoff-runbook.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+- `npm run package:check`
+- `npm run release:self-test`
+- `npm run release:check`
+
+## Phase 723 — Website Improvement Bundle Handoff Stage Snapshot Cleanup (unreleased)
+
+Website Improvement bundle handoff runbook stage assembly now groups stage identity, command metadata, and safety metadata into named internal snapshots before returning the same public stage fields.
+
+### Changed
+- Added internal stage snapshots for stage identity, command summary, and safety summary inside `buildStage`.
+- Replaced the long inline stage return assembly with snapshot spreads while keeping the existing stage field names unchanged.
+
+### Impact
+- `buildBundleHandoffOperatorRunbook` keeps the same top-level `stages[*]` contract.
+- Website Improvement handoff wrappers can continue to consume the same stage identity fields, command lists, command counts, run policies, safety levels, output files, and safety flags without migration.
+- No CLI flags, bundle files, command manifest fields, stage/action field names, evidence field names, validation status values, next-action keys, target-repo mutation rules, external-call boundaries, human runbook lines, or generated output files change.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-handoff-runbook.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+- `npm run package:check`
+- `npm run release:self-test`
+- `npm run release:check`
+
+## Phase 722 — Website Improvement Bundle Handoff Command Manifest Return Snapshot Cleanup (unreleased)
+
+Website Improvement bundle handoff command manifest return assembly now groups command counts, selected task ids, and strict task command keys into named internal snapshots before returning the same public fields.
+
+### Changed
+- Added internal return snapshots for command manifest counts, task selection ids, and strict task command keys.
+- Replaced the long inline command manifest return assembly with snapshot spreads while keeping the existing returned field names unchanged.
+
+### Impact
+- `buildBundleHandoffCommandManifest` keeps the same top-level command manifest contract.
+- Website Improvement bundle handoff wrappers can continue to consume the same command counts, task ids, strict task command keys, and `commands` array without migration.
+- No CLI flags, bundle files, command entries, command scopes, run policies, safety flags, task selection behavior, output-file selection, target-repo mutation rules, external-call boundaries, or generated output files change.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-handoff-runbook.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+- `npm run package:check`
+- `npm run release:self-test`
+- `npm run release:check`
+
+## Phase 721 — Website Improvement Bundle Handoff Runbook Return Snapshot Cleanup (unreleased)
+
+Website Improvement bundle handoff runbook return assembly now groups count, stage identity, human-line, action, evidence-capture, execution, and next-step fields into named internal snapshots before returning the same public fields.
+
+### Changed
+- Added internal return snapshots for runbook counts, stage identity maps, human-line maps, action status maps, action dependency maps, action evidence maps, evidence-capture maps, stage execution maps, and next-step summary fields.
+- Replaced the long inline operator runbook return assembly with snapshot spreads while keeping the existing returned field names unchanged.
+
+### Impact
+- `buildBundleHandoffOperatorRunbook` keeps the same top-level operator runbook contract.
+- Website Improvement handoff wrappers can continue to consume the same stage maps, action maps, evidence-capture maps, `actionSummary`, next-stage fields, next-command fields, and `stages` array without migration.
+- No CLI flags, bundle files, command manifest fields, stage/action field names, evidence field names, validation status values, next-action keys, target-repo mutation rules, external-call boundaries, human runbook lines, or generated output files change.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-handoff-runbook.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+- `npm run package:check`
+- `npm run release:self-test`
+- `npm run release:check`
+
+## Phase 720 — Website Improvement Bundle Handoff Runbook Action Summary Snapshot Cleanup (unreleased)
+
+Website Improvement bundle handoff runbook action summary now groups action count, dependency, evidence, evidence-capture, validation, human-line, lookup, and boundary aggregates into named internal snapshots.
+
+### Changed
+- Added internal action summary snapshots for action counts, dependency/completion metadata, evidence target counts, evidence-capture field counts, payload counts, validation counts, initial validation counts, checklist counts, human-line display counts, validated field counts, run-policy counts, first-action lookups, and boundary flags.
+- Replaced the long inline `actionSummary` assembly with snapshot spreads while keeping the existing action summary field names unchanged.
+
+### Impact
+- `buildBundleHandoffOperatorRunbook` keeps the same returned `actionSummary` contract.
+- Website Improvement handoff wrappers can continue to consume the same action count, evidence-capture, validation, next-action, first-action lookup, and boundary fields without migration.
+- No CLI flags, bundle files, command manifest fields, stage/action field names, evidence field names, validation status values, next-action keys, target-repo mutation rules, external-call boundaries, or human runbook lines change.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-handoff-runbook.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+- `npm run package:check`
+- `npm run release:self-test`
+- `npm run release:check`
+
+## Phase 719 — Website Improvement Bundle Handoff Runbook Action Row Evidence Snapshot Cleanup (unreleased)
+
+Website Improvement bundle handoff runbook action rows now group evidence and evidence-capture assembly into named internal snapshots before returning the same row fields.
+
+### Changed
+- Added an internal `actionEvidenceSummary` snapshot for evidence requirements, evidence-required state, and evidence target metadata.
+- Added an internal `actionEvidenceCaptureSummary` snapshot for evidence-capture fields, field metadata, payload templates, validation metadata, checklist summaries, and capture counts.
+- Replaced the long inline `stageActionRows` evidence assembly with snapshot spreads while keeping the existing action-row field names unchanged.
+
+### Impact
+- `buildBundleHandoffOperatorRunbook` keeps the same `stageActionRows[*].actionEvidence*` and `stageActionRows[*].actionEvidenceCapture*` contract.
+- Downstream Website Improvement handoff wrappers can continue to consume the same action-row evidence fields without migration.
+- No CLI flags, bundle files, command manifest fields, stage/action field names, evidence field names, validation status values, next-action keys, target-repo mutation rules, external-call boundaries, or human runbook lines change.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-handoff-runbook.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+- `npm run package:check`
+- `npm run release:self-test`
+- `npm run release:check`
+
+## Phase 718 — Website Improvement Bundle Handoff Runbook Action Summary Next Snapshot Cleanup (unreleased)
+
+Website Improvement bundle handoff runbook action summary now groups the internal `nextAction*` assembly into a named snapshot object before returning the same public fields.
+
+### Changed
+- Added a grouped internal `nextActionSummary` snapshot for next-action status, dependency/completion metadata, evidence metadata, evidence-capture metadata, and stage linkage metadata.
+- Replaced the long inline `actionSummary` assembly for `nextAction*` fields with the grouped snapshot spread.
+- Kept the existing next-action accessor helpers and public field names unchanged.
+
+### Impact
+- `buildBundleHandoffOperatorRunbook` keeps the same returned `actionSummary.nextAction*` contract.
+- Website Improvement handoff wrappers can continue to consume the same runbook next-action summary fields without migration.
+- No CLI flags, bundle files, command manifest fields, stage/action field names, evidence field names, validation status values, next-action keys, target-repo mutation rules, external-call boundaries, or human runbook lines change.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-handoff-runbook.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+- `npm run package:check`
+- `npm run release:self-test`
+- `npm run release:check`
+
+## Phase 717 — Website Improvement Bundle Handoff Runbook Next Snapshot Cleanup (unreleased)
+
+Website Improvement bundle handoff runbook next-stage output now groups the internal `nextStage*`, `nextStageAction*`, evidence-capture, and `nextCommand*` assembly into named snapshot objects before returning the same public fields.
+
+### Changed
+- Renamed generic next-action and next-stage accessors so scalar, list, object, number, and boolean reads are explicit at the call site.
+- Added grouped internal snapshots for next-stage identity, action status, dependency/completion metadata, evidence metadata, evidence-capture metadata, command-stage metadata, and next-command metadata.
+- Replaced the long inline return assembly for `nextStage*` and `nextCommand*` fields with those grouped snapshots.
+
+### Impact
+- `buildBundleHandoffOperatorRunbook` keeps the same returned `nextStage*`, `nextStageAction*`, `nextCommand*`, evidence-capture, command, safety, and stage fields.
+- Website Improvement handoff wrappers can continue to consume the same runbook next-step contract without migration.
+- No CLI flags, bundle files, command manifest fields, stage/action field names, evidence field names, validation status values, next-action keys, target-repo mutation rules, external-call boundaries, or human runbook lines change.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-handoff-runbook.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+- `npm run release:check`
+
+## Phase 716 — Website Improvement Bundle Handoff Runbook Evidence Predicate Cleanup (unreleased)
+
+Website Improvement bundle handoff evidence-capture summaries now use named predicates for evidence field input type, value shape, payload mapping, validation state, display state, and checklist item state.
+
+### Changed
+- Added evidence-capture field predicates for input type, value shape, multi-value behavior, empty values, placeholder, ARIA label, help text, section membership, payload mapping, and validation rules.
+- Added shared required/optional and multi-value predicates for evidence payload bindings, validation specs, and checklist items.
+- Added initial validation state, display row, and checklist item predicates for valid/invalid, blocking/non-blocking, checked/unchecked, and completion-blocking summary counts.
+- Reused those predicates in evidence-capture summary counts and first evidence-capture action lookups.
+
+### Impact
+- `buildBundleHandoffOperatorRunbook` keeps the same evidence-capture summary fields and first-action lookup fields.
+- Website Improvement handoff wrappers can continue to consume the same runbook evidence summary contract without migration.
+- No CLI flags, bundle files, command manifest fields, stage fields, action field names, evidence field names, evidence targets, validation status values, first-action keys, target-repo mutation rules, external-call boundaries, or human runbook lines change.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-handoff-runbook.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+- `npm run release:check`
+
+## Phase 715 — Website Improvement Bundle Handoff Runbook Action Predicate Cleanup (unreleased)
+
+Website Improvement bundle handoff action summaries now use named predicates for action enabled state, prerequisites, evidence targets, evidence-capture presence, and first-action lookups.
+
+### Changed
+- Added action-row predicates for enabled, disabled, manual-disabled, prerequisite, dependency, blocking, completion-criteria, and evidence-required states.
+- Added action evidence-target predicates for local command output, local output file, target repo working tree, and handoff evidence record.
+- Added evidence-capture predicates for field presence, required/optional field presence, multiple sections, multiple payload namespaces, and payload template presence.
+- Reused those predicates in action summary counts and first-action lookup fields.
+
+### Impact
+- `buildBundleHandoffOperatorRunbook` keeps the same action summary fields and first-action lookup fields.
+- Website Improvement handoff wrappers can continue to consume the same runbook action summary contract without migration.
+- No CLI flags, bundle files, command manifest fields, stage fields, action field names, evidence field names, evidence targets, first-action keys, target-repo mutation rules, external-call boundaries, or human runbook lines change.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-handoff-runbook.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+- `npm run release:check`
+
+## Phase 714 — Website Improvement Bundle Handoff Runbook Display Row Predicate Cleanup (unreleased)
+
+Website Improvement bundle handoff display-row summaries now use named predicates for required/manual rows, command rows, action status, and evidence-progress status.
+
+### Changed
+- Added display-row predicates for required, optional, command-backed, and manual rows.
+- Added display-row predicates for ready, optional, manual, and blocked action statuses.
+- Added display-row predicates for evidence-progress presence plus blocked and ready evidence-progress states.
+- Reused those predicates in display-row status key maps, summary counts, and first-row lookups.
+
+### Impact
+- `buildBundleHandoffOperatorRunbook` keeps the same display-row summary fields.
+- Website Improvement handoff wrappers can continue to consume the same human-line display row contract without migration.
+- No CLI flags, bundle files, command manifest fields, stage fields, action fields, evidence field names, status labels, evidence-progress values, target-repo mutation rules, external-call boundaries, or human runbook lines change.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-handoff-runbook.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+- `npm run release:check`
+
+## Phase 713 — Website Improvement Bundle Handoff Runbook Scope and Safety Predicate Cleanup (unreleased)
+
+Website Improvement bundle handoff command and operator summaries now use named predicates for command scope, safety flags, required stages, and external/target-repo boundary counts.
+
+### Changed
+- Added command-manifest predicates for source-bundle and task-handoff scope counts.
+- Added command-manifest safety predicates for external calls, target-repo mutation, clean-workspace requirements, and review-before-mutation requirements.
+- Added runbook-stage predicates for required/optional stages and external/target-repo boundary checks.
+- Reused those predicates in manifest counts, action summaries, first-stage lookups, returned stage counts, and top-level boundary booleans.
+
+### Impact
+- `buildBundleHandoffCommandManifest` keeps the same command count and safety count fields.
+- `buildBundleHandoffOperatorRunbook` keeps the same stage count, action summary, first-stage key, and boundary fields.
+- Website Improvement handoff wrappers can continue to consume the same manifest and runbook contracts without migration.
+- No CLI flags, bundle files, command manifest field names, target-repo mutation rules, external-call boundaries, evidence field names, action counts, first-stage keys, or human runbook lines change.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-handoff-runbook.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+- `npm run release:check`
+
+## Phase 712 — Website Improvement Bundle Handoff Runbook Run Policy Predicate Cleanup (unreleased)
+
+Website Improvement bundle handoff command and operator summaries now use named predicates for read-only and local-output run policy counts.
+
+### Changed
+- Added command-manifest run policy predicates for read-only and local-output command counts.
+- Added a runbook-stage `usesReadOnlyRunPolicy` predicate beside the existing local-output predicate.
+- Reused those predicates for manifest counts, action summary counts, and returned command-stage summary counts.
+
+### Impact
+- `buildBundleHandoffCommandManifest` keeps the same command count fields.
+- `buildBundleHandoffOperatorRunbook` keeps the same read-only and local-output summary fields.
+- Website Improvement handoff wrappers can continue to consume the same manifest and runbook contracts without migration.
+- No CLI flags, bundle files, command manifest field names, target-repo mutation rules, external-call boundaries, evidence field names, action counts, or human runbook lines change.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-handoff-runbook.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+- `npm run release:check`
+
+## Phase 711 — Website Improvement Bundle Handoff Runbook Output Predicate Cleanup (unreleased)
+
+Website Improvement bundle handoff operator stage summaries now use named predicates for local-output run policy and output-file checks.
+
+### Changed
+- Added local `usesLocalOutputRunPolicy` for `writes-local-file` run-policy checks.
+- Added local `hasOutputFile` for output-file presence checks.
+- Reused those predicates in action summary and returned command-stage summary counts.
+
+### Impact
+- `buildBundleHandoffOperatorRunbook` keeps the same local-output and output-file count fields.
+- Website Improvement handoff wrappers can continue to consume the same runbook summary contract without migration.
+- No CLI flags, bundle files, command manifest fields, target-repo mutation rules, external-call boundaries, evidence field names, action counts, or human runbook lines change.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-handoff-runbook.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+- `npm run release:check`
+
+## Phase 710 — Website Improvement Bundle Handoff Runbook Display Row Summary Cleanup (unreleased)
+
+Website Improvement bundle handoff operator action summaries now reuse the prepared human-line display-row summary for display-row counts.
+
+### Changed
+- Reused `stageHumanLineDisplayRowSummary.count` for `humanLineDisplayRowCount`.
+- Reused `stageHumanLineDisplayRowSummary.byKeyCount` for `humanLineDisplayRowByKeyCount`.
+- Removed repeated direct display-row length and by-key object counting inside `actionSummary`.
+
+### Impact
+- `buildBundleHandoffOperatorRunbook` keeps the same human-line display-row action summary fields.
+- Website Improvement handoff wrappers can continue to consume the same action summary contract without migration.
+- No CLI flags, bundle files, command manifest fields, target-repo mutation rules, external-call boundaries, evidence field names, action counts, or human runbook lines change.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-handoff-runbook.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+- `npm run release:check`
+
+## Phase 709 — Website Improvement Bundle Handoff Runbook Initial Validation Summary Cleanup (unreleased)
+
+Website Improvement bundle handoff operator action summaries now route initial validation summary and checklist summary aggregations through named local helpers.
+
+### Changed
+- Added local `countActionsBySummary` and `sumSummaryValues` helpers for action-row summary objects.
+- Added local initial-validation summary wrappers for field-count, status, completion, and blocking totals.
+- Added local initial-checklist summary wrappers for item-count, status, completion, checked, unchecked, and blocking-unchecked totals.
+- Replaced repeated direct `actionEvidenceCaptureInitialValidationSummary` and `actionEvidenceCaptureInitialValidationChecklistSummary` accesses inside `actionSummary`.
+
+### Impact
+- `buildBundleHandoffOperatorRunbook` keeps the same initial validation summary and checklist summary fields.
+- Website Improvement handoff wrappers can continue to consume the same action summary contract without migration.
+- No CLI flags, bundle files, command manifest fields, target-repo mutation rules, external-call boundaries, evidence field names, action counts, or human runbook lines change.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-handoff-runbook.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+- `npm run release:check`
+
+## Phase 708 — Website Improvement Bundle Handoff Runbook Action Item Count Cleanup (unreleased)
+
+Website Improvement bundle handoff operator action summaries now name repeated item-count and item-value aggregations before assembling evidence-capture totals.
+
+### Changed
+- Added local `countActionsWithItems` for action rows that contain at least one item in a named list field.
+- Added local `sumActionItems` for total item counts across action rows.
+- Added local `sumActionItemValues` for numeric totals derived from list item values.
+- Replaced repeated `.length > 0`, `.length`, and nested `reduce` summary calculations with the named helpers.
+
+### Impact
+- `buildBundleHandoffOperatorRunbook` keeps the same payload binding, validation spec, initial validation state, display metadata, checklist, and minimum-length summary fields.
+- Website Improvement handoff wrappers can continue to consume the same action summary fields without migration.
+- No CLI flags, bundle files, command manifest fields, target-repo mutation rules, external-call boundaries, evidence field names, action counts, or human runbook lines change.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-handoff-runbook.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+- `npm run release:check`
+
+## Phase 707 — Website Improvement Bundle Handoff Runbook Payload Summary Cleanup (unreleased)
+
+Website Improvement bundle handoff operator action summaries now name the payload template and unique-list summary calculations before returning the runbook payload.
+
+### Changed
+- Added local `uniqueActionListValueCount` for unique action-list summary counts.
+- Added local `payloadTemplatePathCount` for payload flat-template path counts.
+- Replaced inline `flatMap`/`Object.keys(...).length` summary calculations with the named helpers.
+
+### Impact
+- `buildBundleHandoffOperatorRunbook` keeps the same payload template path counts, unique section counts, unique payload namespace counts, action summary fields, and returned runbook contract.
+- Website Improvement handoff wrappers can continue to consume the same summary fields without migration.
+- No CLI flags, bundle files, command manifest fields, target-repo mutation rules, external-call boundaries, evidence field names, action counts, or human runbook lines change.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-handoff-runbook.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+- `npm run release:check`
+
+## Phase 706 — Website Improvement Bundle Handoff Runbook Action Item Summary Cleanup (unreleased)
+
+Website Improvement bundle handoff operator action summaries now use shared local item-summary helpers for repeated evidence item counts and first-action lookups.
+
+### Changed
+- Added local `countActionItems` and `maxActionItemValue` helpers near the `actionSummary` assembly.
+- Reused the item-summary helpers for evidence capture fields, payload bindings, validation specs, initial validation state rows, display rows, and checklist items.
+- Added `firstActionWithEvidenceCaptureField` for first-action evidence field lookups.
+
+### Impact
+- `buildBundleHandoffOperatorRunbook` keeps the same action summary counts, first-action keys, evidence capture metadata, and returned runbook contract.
+- Website Improvement handoff wrappers can continue to consume the same summary fields without migration.
+- No CLI flags, bundle files, command manifest fields, target-repo mutation rules, external-call boundaries, evidence field names, action counts, or human runbook lines change.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-handoff-runbook.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+- `npm run release:check`
+
+## Phase 705 — Website Improvement Bundle Handoff Runbook Evidence Field Value Cleanup (unreleased)
+
+Website Improvement bundle handoff operator action rows now read evidence-capture field values through one local helper before assembling the row payload.
+
+### Changed
+- Added a local `evidenceFieldValues` helper inside the `stageActionRows` mapper.
+- Replaced repeated `evidenceCaptureFields.map(...)` field extraction calls with `evidenceFieldValues`.
+- Reused the same helper for evidence capture section labels and payload namespaces before returning action row metadata.
+
+### Impact
+- `buildBundleHandoffOperatorRunbook` keeps the same evidence-capture field arrays, section metadata, payload namespace metadata, validation metadata, and action row contract.
+- Website Improvement handoff wrappers can continue to consume the same stage action evidence fields without migration.
+- No CLI flags, bundle files, command manifest fields, target-repo mutation rules, external-call boundaries, evidence field names, action counts, or human runbook lines change.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-handoff-runbook.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+- `npm run release:check`
+
+## Phase 704 — Website Improvement Bundle Handoff Runbook Manual Stage Predicate Cleanup (unreleased)
+
+Website Improvement bundle handoff operator stages now use named predicates for command-backed and manual stage checks instead of repeating `commandCount` comparisons inline.
+
+### Changed
+- Added local `hasCommands` and `isManualStage` predicates inside `buildBundleHandoffOperatorRunbook`.
+- Replaced repeated `stage.commandCount > 0` checks with `hasCommands`.
+- Replaced repeated `stage.commandCount === 0` checks with `isManualStage`.
+
+### Impact
+- `buildBundleHandoffOperatorRunbook` keeps the same command-stage counts, manual-stage counts, lookup maps, next-action summary fields, and stage lists.
+- Website Improvement handoff wrappers can continue to consume the same stage and action fields without migration.
+- No CLI flags, bundle files, command manifest fields, target-repo mutation rules, external-call boundaries, evidence fields, action counts, or human runbook lines change.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-handoff-runbook.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+- `npm run release:check`
+
+## Phase 703 — Website Improvement Bundle Handoff Runbook Stage Command Safety Cleanup (unreleased)
+
+Website Improvement bundle handoff operator stages now read first-command run policy and safety metadata through named local values before assembling each stage record.
+
+### Changed
+- Added local `firstCommand`, `firstCommandSafety`, and `commandHasSafetyFlag` values inside `buildStage`.
+- Replaced repeated `stageCommands[0]?.*` lookups with the named first-command values.
+- Replaced repeated stage-command safety flag predicates with `commandHasSafetyFlag`.
+
+### Impact
+- `buildBundleHandoffOperatorRunbook` keeps the same stage run policy, safety level, local-output, external-call, and target-repo mutation metadata.
+- Website Improvement handoff wrappers can continue to consume the same stage and command fields without migration.
+- No CLI flags, bundle files, command manifest fields, target-repo mutation rules, external-call boundaries, evidence fields, action counts, or human runbook lines change.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-handoff-runbook.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+- `npm run release:check`
+
+## Phase 702 — Website Improvement Bundle Handoff Command Manifest Fallback Cleanup (unreleased)
+
+Website Improvement bundle handoff command manifest entries now read source-bundle and task command metadata through small local fallback helpers before assembling each command record.
+
+### Changed
+- Added a local `sourceValue` helper inside `buildBundleHandoffCommandManifest` for source-bundle command, args, run-policy, and safety metadata.
+- Read source-bundle safety once per source command so `strict` comes from the same object that is returned in the command manifest.
+- Added a local `taskValue` helper inside `pushTaskCommand` and read task safety once before deriving strict task command metadata and output file fallback.
+
+### Impact
+- `buildBundleHandoffCommandManifest` keeps the same command keys, task ids, output file fallback, strict flags, safety metadata, and manifest JSON contract.
+- Website Improvement handoff wrappers can continue to consume the same command manifest fields without migration.
+- No CLI flags, bundle files, target-repo mutation rules, external-call boundaries, operator runbook fields, evidence fields, action counts, or human runbook lines change.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-handoff-runbook.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+- `npm run release:check`
+
+## Phase 701 — Website Improvement Bundle Handoff Runbook Next Stage Return Cleanup (unreleased)
+
+Website Improvement target-repo handoff runbook next-stage and next-command return metadata now use the same local fallback style as the next-action fields.
+
+### Changed
+- Added local `nextStageList`, `nextStageFlag`, `nextCommandValue`, and `nextCommandList` helpers inside `buildBundleHandoffOperatorRunbook`.
+- Precomputed the next-stage human line, display row, command count, and command safety before assembling the returned runbook payload.
+- Replaced the remaining repeated `nextStage?.*` and `nextCommandEntry?.*` fallback expressions in the return object with named values or helper calls.
+
+### Impact
+- `buildBundleHandoffOperatorRunbook` keeps the same next-stage labels, command metadata, safety metadata, human-line summary, and JSON contract.
+- Target-repo handoff wrappers can continue to consume the same next-stage and next-command fields without migration.
+- No CLI flags, bundle files, target-repo mutation rules, external-call boundaries, lookup maps, evidence fields, action counts, or human runbook lines change.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-handoff-runbook.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+- `npm run release:check`
+
+## Phase 700 — Website Improvement Bundle Handoff Runbook Next Action Fallback Cleanup (unreleased)
+
+Website Improvement target-repo handoff runbook next-action metadata now uses small local fallback helpers instead of repeating the same optional-chaining defaults in `actionSummary` and the returned runbook payload.
+
+### Changed
+- Added local `nextActionValue`, `nextActionList`, `nextActionObject`, `nextActionCount`, `nextActionFlag`, `nextStageValue`, and `nextStageCommandList` helpers inside `buildBundleHandoffOperatorRunbook`.
+- Replaced repeated `nextAction*` and `nextStageAction*` fallback expressions with helper calls while keeping every exported field name unchanged.
+- Read the nested next-stage evidence progress summary once before writing the human-line summary fields.
+
+### Impact
+- `buildBundleHandoffOperatorRunbook` keeps the same next-action, next-stage action, evidence progress, command metadata, and JSON contract.
+- Target-repo handoff wrappers can continue to consume the same next-action and next-stage fields without migration.
+- No CLI flags, bundle files, target-repo mutation rules, external-call boundaries, lookup maps, evidence fields, action counts, or human runbook lines change.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-handoff-runbook.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+
+## Phase 699 — Website Improvement Bundle Handoff Runbook Stage Command Lookup Cleanup (unreleased)
+
+Website Improvement target-repo handoff runbook stage command lookup maps now use named helpers instead of repeating command-list mapping expressions inline.
+
+### Changed
+- Added local `commandListByKey` and `stageKeysBy` helpers inside `buildBundleHandoffOperatorRunbook`.
+- Replaced repeated stage command label, command string, command args, run policy, and safety-level lookup map construction with `commandListByKey`.
+- Replaced command/manual stage key list construction with `stageKeysBy` while keeping the exported arrays unchanged.
+
+### Impact
+- `buildBundleHandoffOperatorRunbook` keeps the same command lookup maps, stage key arrays, action summaries, and JSON contract.
+- Target-repo handoff wrappers can continue to consume the same stage command metadata without migration.
+- No CLI flags, bundle files, target-repo mutation rules, external-call boundaries, evidence fields, action counts, or human runbook lines change.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-handoff-runbook.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+
+## Phase 698 — Website Improvement Bundle Handoff Runbook Human Line Summary Cleanup (unreleased)
+
+Website Improvement target-repo handoff runbook human-line display summaries and first-action key fields now use small named helpers instead of repeating row and action scans inline.
+
+### Changed
+- Added local display-row count, key-list, first-key, and action first-key helpers inside `buildBundleHandoffOperatorRunbook`.
+- Replaced repeated human-line display row status counts, evidence-progress counts, and first-row lookups with named helper calls.
+- Replaced repeated first action key lookups in `actionSummary` with the shared `firstActionKey` helper while keeping every exported field name unchanged.
+
+### Impact
+- `buildBundleHandoffOperatorRunbook` keeps the same human-line summary, display-row summary, first-action key fields, and JSON contract.
+- Target-repo handoff wrappers can continue to consume the same stage human-line, evidence-progress, and first-action metadata without migration.
+- No CLI flags, bundle files, target-repo mutation rules, external-call boundaries, lookup maps, evidence fields, action counts, or human runbook lines change.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-handoff-runbook.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+
+## Phase 697 — Website Improvement Bundle Handoff Runbook Action Summary Count Cleanup (unreleased)
+
+Website Improvement target-repo handoff runbook action summary count and aggregate calculations now use named local helpers instead of repeating filter, reduce, max, and evidence-capture traversal expressions.
+
+### Changed
+- Added local `countActions`, `sumActions`, `maxActionValue`, and evidence-capture aggregate helpers inside `buildBundleHandoffOperatorRunbook`.
+- Replaced repeated action summary count, sum, max, payload binding, validation spec, validation state, display metadata, and checklist aggregate expressions with named helper calls.
+- Kept the action summary object keys and computed values unchanged.
+
+### Impact
+- `buildBundleHandoffOperatorRunbook` keeps the same `actionSummary` JSON contract while the summary section reads as a set of named aggregate calculations.
+- Target-repo handoff wrappers can continue to consume the same summary fields without migration.
+- No CLI flags, bundle files, target-repo mutation rules, external-call boundaries, lookup maps, evidence fields, or human runbook lines change.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-handoff-runbook.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+
+## Phase 696 — Website Improvement Bundle Handoff Runbook Lookup Assembly Cleanup (unreleased)
+
+Website Improvement target-repo handoff runbook lookup map assembly now uses a small shared helper instead of repeating the same `Object.fromEntries` pattern through the assembly body.
+
+### Changed
+- Added local `byKey`, `fieldByKey`, `stageFieldByKey`, and `actionFieldByKey` helpers inside `buildBundleHandoffOperatorRunbook`.
+- Replaced repeated stage/action lookup map construction with named helper calls while keeping every exported lookup object and key unchanged.
+- Kept nested command lookup maps explicit where they derive arrays from stage commands.
+
+### Impact
+- `buildBundleHandoffOperatorRunbook` keeps the same lookup maps, action rows, summaries, next-action fields, and JSON contract.
+- Target-repo handoff wrappers can continue to read the same `stage*ByKey` and `stageAction*ByKey` objects without migration.
+- No CLI flags, bundle files, target-repo mutation rules, external-call boundaries, evidence fields, or human runbook lines change.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-handoff-runbook.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+
+## Phase 695 — Website Improvement Bundle Handoff Runbook Stage Metadata Split (unreleased)
+
+Website Improvement target-repo handoff runbook stage dependency and requirement metadata now lives outside the runbook assembly body.
+
+### Changed
+- Moved stage prerequisite, blocked-stage, dependency reason, completion criteria, and evidence requirement helpers into `cli/lib/site-bundle-handoff-runbook-stage-metadata.mjs`.
+- Kept `cli/lib/site-bundle-handoff-runbook.mjs` focused on building stages, action rows, lookup maps, summaries, and the final operator runbook payload.
+- Reworked action row assembly to compute prerequisite, blocking, completion, evidence requirement, and evidence capture values once per stage before writing the public row fields.
+
+### Impact
+- `buildBundleHandoffOperatorRunbook` keeps the same stage action dependency, completion criteria, evidence requirement, and evidence capture JSON contract.
+- Target-repo handoff wrappers can continue to read the same prerequisite keys, blocked-stage labels, dependency reason codes, completion criteria, evidence requirements, payload templates, and validation metadata.
+- No CLI flags, bundle files, target-repo mutation rules, external-call boundaries, or human runbook lines change.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-handoff-runbook-stage-metadata.mjs && node --check cli/lib/site-bundle-handoff-runbook.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+
+## Phase 694 — Website Improvement Bundle Handoff Runbook Evidence Field Split (unreleased)
+
+Website Improvement target-repo handoff runbook stage-specific evidence field definitions now live outside the runbook assembly body.
+
+### Changed
+- Moved stage-specific evidence capture field definitions and field enrichment into `cli/lib/site-bundle-handoff-runbook-evidence-fields.mjs`.
+- Kept `cli/lib/site-bundle-handoff-runbook.mjs` focused on assembling stage action rows, summaries, lookup maps, and the final operator runbook payload.
+- Reused the existing action metadata and generic evidence helper modules so evidence targets, field shapes, payload paths, sections, and validation metadata stay consistent.
+
+### Impact
+- `buildBundleHandoffOperatorRunbook` keeps the same evidence capture fields and JSON contract while the runbook assembly reads as a shorter flow.
+- Target-repo handoff wrappers can continue to consume the same field keys, placeholders, examples, validation hints, payload paths, and section metadata.
+- No CLI flags, bundle files, target-repo mutation rules, or external-call boundaries change.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-handoff-runbook-evidence-fields.mjs && node --check cli/lib/site-bundle-handoff-runbook.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+
+## Phase 693 — Website Improvement Bundle Handoff Runbook Evidence Helper Split (unreleased)
+
+Website Improvement target-repo handoff runbook evidence capture helper logic now lives outside the runbook assembly body.
+
+### Changed
+- Moved evidence capture field shape, section, payload binding, validation spec, and initial checklist/summary helpers into `cli/lib/site-bundle-handoff-runbook-evidence.mjs`.
+- Kept stage-specific evidence field selection in `cli/lib/site-bundle-handoff-runbook.mjs`, where it still has access to the current runbook stage context.
+- Preserved existing evidence capture payload paths, validation labels, section keys, checklist summaries, and next-action metadata.
+
+### Impact
+- `buildBundleHandoffOperatorRunbook` keeps the same evidence capture JSON contract while the generic evidence helper code has a focused module boundary.
+- Target-repo handoff wrappers can continue to consume the same payload template, flat template, validation specs, checklist rows, and summary fields.
+- No CLI flags, bundle files, target-repo mutation rules, or external-call boundaries change.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-handoff-runbook-evidence.mjs && node --check cli/lib/site-bundle-handoff-runbook.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+
+## Phase 692 — Website Improvement Bundle Handoff Runbook Action Metadata Split (unreleased)
+
+Website Improvement target-repo handoff runbook action metadata now lives outside the runbook assembly body.
+
+### Changed
+- Moved stage action type, labels, instructions, button labels, affordances, status metadata, disabled reasons, and evidence target labels into `cli/lib/site-bundle-handoff-runbook-actions.mjs`.
+- Kept `cli/lib/site-bundle-handoff-runbook.mjs` focused on command manifest input, stage assembly, evidence capture rows, and the final runbook payload.
+- Preserved the existing runbook JSON keys and human output by importing the same helper behavior from the new action metadata module.
+
+### Impact
+- `buildBundleHandoffOperatorRunbook` keeps its current public contract while the stage action wording and UI-oriented metadata have a focused module boundary.
+- Target-repo handoff wrappers can still consume the same action fields, status labels, evidence targets, and disabled reason codes.
+- No bundle files, CLI flags, target-repo mutation rules, or external-call boundaries change.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-handoff-runbook-actions.mjs && node --check cli/lib/site-bundle-handoff-runbook.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+
+## Phase 691 — Website Improvement Bundle Handoff Runbook Formatter Split (unreleased)
+
+Website Improvement target-repo handoff runbook line formatting now lives outside the runbook assembly module.
+
+### Changed
+- Moved operator runbook human line formatting into `cli/lib/site-bundle-handoff-runbook-format.mjs`.
+- Kept `cli/lib/site-bundle-handoff-runbook.mjs` responsible for command manifest and operator runbook assembly.
+- Re-exported the existing public formatter from the runbook module so current imports keep working.
+
+### Impact
+- `buildBundleHandoffOperatorRunbook` keeps the same JSON contract and stage metadata.
+- Human runbook lines remain available through the same public helper while formatting logic has a smaller module boundary.
+- Target-repo handoff command manifest, runbook stages, evidence capture metadata, and existing CLI output contracts are unchanged.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-handoff-runbook-format.mjs && node --check cli/lib/site-bundle-handoff-runbook.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+
+## Phase 690 — Website Improvement Bundle Repair Report Split (unreleased)
+
+Website Improvement bundle repair report orchestration now lives outside the public site helper surface.
+
+### Changed
+- Moved bundle repair preview, repair bundle creation, applied repair report assembly, and JSON/human repair formatters into `cli/lib/site-bundle-repair-report.mjs`.
+- Kept `cli/lib/site.mjs` as a public compatibility surface by re-exporting repair report helpers.
+- Avoided a circular dependency with `cli/lib/site-bundle-check.mjs` by keeping repair guidance helpers in `cli/lib/site-bundle-repair.mjs` and repair orchestration in a separate module.
+
+### Impact
+- `cli/lib/site.mjs` now acts as a compact Website Improvement public export surface instead of owning bundle workflow bodies.
+- `design-ai site <bundle-dir> --bundle-repair` continues to expose the same preview/apply report contracts.
+- Bundle check, repair guidance, and repair report orchestration remain separated by dependency direction.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-repair-report.mjs && node --check cli/lib/site.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+
+## Phase 689 — Website Improvement Bundle Handoff Report Split (unreleased)
+
+Website Improvement target-repo handoff report construction now lives outside the main site workflow module.
+
+### Changed
+- Moved target-repo handoff workspace loading, prompt assembly, task selection handling, command manifest/runbook wiring, and JSON/human formatters into `cli/lib/site-bundle-handoff.mjs`.
+- Kept `cli/lib/site.mjs` as the public compatibility surface by re-exporting bundle handoff helpers.
+- Removed local imports from `cli/lib/site.mjs` that were only needed by handoff report construction.
+
+### Impact
+- `cli/lib/site.mjs` no longer owns the target-repo handoff report body.
+- `design-ai site <bundle-dir> --bundle-handoff` continues to expose the same prompt and JSON report contracts.
+- Handoff report assembly now sits beside the handoff summary and operator runbook helpers instead of remaining in the workflow orchestration module.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-handoff.mjs && node --check cli/lib/site.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+
+## Phase 688 — Website Improvement Bundle Compare Report Split (unreleased)
+
+Website Improvement handoff bundle comparison report construction now lives with the compare helpers.
+
+### Changed
+- Moved `buildSiteBundleCompareReport` into `cli/lib/site-bundle-compare.mjs`.
+- Kept `cli/lib/site.mjs` as the public compatibility surface by re-exporting the compare report builder.
+- Preserved compare issue ids, digest comparison, metadata diffing, changed file reporting, and JSON/human formatter contracts.
+
+### Impact
+- `cli/lib/site.mjs` no longer owns bundle comparison report assembly.
+- `design-ai site <bundle-dir> --bundle-compare other-bundle-dir` continues to use the same report shape and formatter outputs.
+- Bundle comparison logic is grouped with its checksum/metadata diff helpers instead of remaining in the workflow orchestration module.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-compare.mjs && node --check cli/lib/site.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+
+## Phase 687 — Website Improvement Bundle Check Module Split (unreleased)
+
+Website Improvement handoff bundle validation now lives outside the main site workflow module.
+
+### Changed
+- Moved bundle directory validation, summary/checksum verification, MCP recomputation checks, generated contract checks, and bundle check formatting into `cli/lib/site-bundle-check.mjs`.
+- Kept `cli/lib/site.mjs` as the public compatibility surface by re-exporting bundle check helpers.
+- Removed local imports that were only needed by the moved bundle check implementation.
+
+### Impact
+- `cli/lib/site.mjs` no longer owns the large bundle check validation body.
+- Bundle repair, bundle compare, and target-repo handoff flows continue to consume the same `buildSiteBundleCheckReport` shape.
+- Bundle validation now sits with the bundle-specific helper modules instead of being embedded in the remaining workflow orchestration file.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-check.mjs && node --check cli/lib/site.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+
+## Phase 686 — Website Improvement Generated Contract Helper Split (unreleased)
+
+Website Improvement generated bundle contract helpers now live outside the main site workflow module.
+
+### Changed
+- Moved generated contract construction, drift issue creation, and drift summary formatting into `cli/lib/site-bundle-contract.mjs`.
+- Kept `cli/lib/site.mjs` focused on bundle check and handoff orchestration by importing the generated contract helpers.
+- Preserved generated bundle contract comparison behavior, digest formatting, repair guidance inputs, and bundle check report shape.
+
+### Impact
+- `cli/lib/site.mjs` no longer owns the generated bundle contract helper bodies.
+- Bundle check, bundle repair, bundle compare, and target-repo handoff flows continue to receive the same `generatedContract` data.
+- Generated contract logic now sits beside bundle file/checksum helpers instead of being embedded in the main workflow module.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-contract.mjs && node --check cli/lib/site.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+
+## Phase 685 — Website Improvement Handoff Bundle Builder Split (unreleased)
+
+Website Improvement handoff bundle assembly now lives outside the main site workflow module.
+
+### Changed
+- Moved `buildSiteHandoffBundle` into `cli/lib/site-bundle-build.mjs`.
+- Kept `cli/lib/site.mjs` as the public compatibility surface by re-exporting the bundle builder.
+- Preserved bundle summary, file ordering, checksums, README generation, MCP artifacts, and prompt artifacts without changing output contracts.
+
+### Impact
+- `cli/lib/site.mjs` no longer owns the handoff bundle file assembly body.
+- Existing imports from `cli/lib/site.mjs`, `design-ai site --bundle`, bundle repair, generated-contract comparison, and bundle handoff flows remain behavior-compatible.
+- Bundle construction is now grouped with its deterministic artifact dependencies while bundle check/repair orchestration remains in the main workflow module.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-build.mjs && node --check cli/lib/site.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+
+## Phase 684 — Website Improvement Task and Workflow Graph Helper Split (unreleased)
+
+Website Improvement generated task planning and portable workflow graph rendering now live outside the main site workflow module.
+
+### Changed
+- Moved deterministic refactor task generation into `cli/lib/site-tasks.mjs`.
+- Moved workflow graph node/edge construction plus JSON/Markdown graph formatters into `cli/lib/site-workflow-graph.mjs`.
+- Kept `cli/lib/site.mjs` as the public compatibility surface by re-exporting task and graph helpers.
+
+### Impact
+- `cli/lib/site.mjs` no longer owns generated task prompt bodies or graph rendering bodies.
+- Existing imports from `cli/lib/site.mjs`, `design-ai site --tasks`, and `design-ai site --graph` remain behavior-compatible.
+- Task planning can now be reused by graph and bundle orchestration without forcing those deterministic helpers to stay in the main workflow module.
+
+### Verification Plan
+- `node --check cli/lib/site-tasks.mjs && node --check cli/lib/site-workflow-graph.mjs && node --check cli/lib/site.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+
+## Phase 683 — Website Improvement Analysis Helper Split (unreleased)
+
+Website Improvement workspace validation, summary generation, JSON loading, and report formatting now live outside the main site workflow module.
+
+### Changed
+- Moved raw workspace validation, issue status reduction, summary counts, `analyzeSiteWorkspace`, `loadSiteWorkspaceInput`, `buildSiteReport`, and `formatSiteJson` into `cli/lib/site-analysis.mjs`.
+- Kept `cli/lib/site.mjs` as the public compatibility surface by re-exporting the existing analysis/report helper API.
+- Imported `addIssue`, `statusFromIssues`, `analyzeSiteWorkspace`, and `loadSiteWorkspaceInput` back into `cli/lib/site.mjs` for bundle repair/check/handoff orchestration.
+
+### Impact
+- `cli/lib/site.mjs` no longer owns schema validation, report input loading, or workspace summary construction bodies.
+- Existing imports from `cli/lib/site.mjs`, report JSON output, strict workspace validation, and bundle consistency checks remain behavior-compatible.
+- Analysis rules now sit beside workspace normalization while bundle orchestration remains in the main site workflow module.
+
+### Verification Plan
+- `node --check cli/lib/site-analysis.mjs && node --check cli/lib/site.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+
+## Phase 682 — Website Improvement Workspace Builder Split (unreleased)
+
+Website Improvement workspace initialization, intake parsing, and normalization now live outside the main site workflow module.
+
+### Changed
+- Moved real-project `--init` workspace construction, filled intake Markdown parsing, and workspace normalization into `cli/lib/site-workspace.mjs`.
+- Kept `cli/lib/site.mjs` as the public compatibility surface by re-exporting workspace builder helpers.
+- Imported only `normalizeSiteWorkspace` and `normalizeObject` back into `cli/lib/site.mjs` for report validation and bundle summary orchestration.
+
+### Impact
+- `cli/lib/site.mjs` no longer owns workspace creation, intake table parsing, or schema normalization bodies.
+- Existing imports from `cli/lib/site.mjs`, `design-ai site --init`, and `design-ai site --from-intake` remain behavior-compatible.
+- Workspace construction is grouped with its defaults and normalization rules, leaving the main site module focused on analysis, reports, graphs, and bundle orchestration.
+
+### Verification Plan
+- `node --check cli/lib/site-workspace.mjs && node --check cli/lib/site.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+
+## Phase 681 — Website Improvement Starter Artifact Helper Split (unreleased)
+
+Website Improvement intake template and sample workspace builders now live outside the main site workflow module.
+
+### Changed
+- Moved intake template Markdown/JSON rendering and sample workspace construction into `cli/lib/site-starter.mjs`.
+- Kept `cli/lib/site.mjs` as the public compatibility surface by re-exporting the starter helper API.
+- Preserved the sample workspace as the normalization fallback by importing it from the starter module.
+
+### Impact
+- `cli/lib/site.mjs` no longer owns starter artifact bodies for `--intake-template` and `--sample`.
+- Intake template output, sample workspace output, CLI import paths, and normalization fallback behavior remain behavior-compatible.
+- Starter artifacts now live with their static content dependencies instead of mixing with workspace analysis and bundle orchestration.
+
+### Verification Plan
+- `node --check cli/lib/site-starter.mjs && node --check cli/lib/site.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+
+## Phase 680 — Website Improvement Bundle README Helper Split (unreleased)
+
+Website Improvement bundle README and strict handoff guidance builders now live outside the main site workflow module.
+
+### Changed
+- Moved bundle handoff guidance construction and bundle README Markdown assembly into `cli/lib/site-bundle-readme.mjs`.
+- Kept bundle generation orchestration, checksum creation, and artifact ordering in `cli/lib/site.mjs`.
+- Reused the existing target-repo execution checklist contract from `cli/lib/site-bundle-handoff-summary.mjs`.
+
+### Impact
+- `cli/lib/site.mjs` no longer owns bundle README copy or strict/draft handoff command guidance.
+- Bundle contents, README text, handoff readiness metadata, and existing bundle check behavior remain behavior-compatible.
+- The remaining bundle generation flow is easier to scan because artifact orchestration is separated from README prose assembly.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-readme.mjs && node --check cli/lib/site.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+
+## Phase 679 — Website Improvement CLI Argument Parser Split (unreleased)
+
+Website Improvement CLI argument parsing now lives outside the main site workflow module.
+
+### Changed
+- Moved `SITE_OPTIONS`, option value parsing, and `parseSiteArgs` into `cli/lib/site-args.mjs`.
+- Exported shared Website Improvement option enums from `cli/lib/site-options.mjs` so parsing, intake, normalization, and validation use the same catalogs.
+- Kept `cli/lib/site.mjs` as the public compatibility surface by re-exporting the parser API.
+
+### Impact
+- `cli/lib/site.mjs` no longer owns command-line parsing rules.
+- `design-ai site` argument behavior, error messages, and existing imports from `cli/lib/site.mjs` remain behavior-compatible.
+- The remaining site workflow module is easier to scan because CLI parsing is separated from workspace construction and report generation.
+
+### Verification Plan
+- `node --check cli/lib/site-args.mjs && node --check cli/lib/site-options.mjs && node --check cli/lib/site.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+
+## Phase 678 — Website Improvement Prompt Helper Consolidation (unreleased)
+
+Website Improvement prompt template formatters and bundle implementation prompt fallback now live with the prompt builders.
+
+### Changed
+- Moved prompt template JSON/human formatters and bundle implementation prompt fallback rendering into `cli/lib/site-prompts.mjs`.
+- Kept `cli/lib/site.mjs` as the public re-export surface and bundle orchestration caller for existing imports.
+- Reused the existing prompt builder helpers for the no-task bundle fallback instead of keeping prompt body composition in the site workflow module.
+
+### Impact
+- `cli/lib/site.mjs` no longer owns prompt-template display or bundle implementation prompt body construction.
+- Prompt list CLI output, prompt bundle generation, and handoff bundle `codex-implementation.md` content remain behavior-compatible.
+- Prompt-related rendering is grouped with the rest of the prompt/handoff report code, making the remaining site workflow module easier to scan.
+
+### Verification Plan
+- `node --check cli/lib/site-prompts.mjs && node --check cli/lib/site.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+
+## Phase 677 — Website Improvement Bundle Handoff Summary Helper Split (unreleased)
+
+Website Improvement bundle handoff task catalog and source summary helpers now live outside the main site workflow module.
+
+### Changed
+- Moved target-repo execution checklist, bundle task catalog summaries, selected-task summary, handoff issue/task catalog line formatting, handoff boundaries, and source bundle command summary into `cli/lib/site-bundle-handoff-summary.mjs`.
+- Kept bundle workspace loading, task selection orchestration, handoff prompt construction, and final handoff report assembly in `cli/lib/site.mjs`.
+- Reused the extracted summary helpers in bundle readme/check validation and target-repo handoff output paths without changing the JSON or Markdown contracts.
+
+### Impact
+- `cli/lib/site.mjs` no longer owns task catalog shaping or source bundle command summary construction.
+- Bundle check validation, handoff prompt checklist text, selected-task command metadata, and command manifest inputs remain behavior-compatible.
+- This makes the remaining handoff flow easier to scan because catalog/source summary data is separated from report orchestration.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-handoff-summary.mjs && node --check cli/lib/site.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+
+## Phase 676 — Website Improvement Bundle Handoff Runbook Helper Split (unreleased)
+
+Website Improvement bundle handoff command manifest and operator runbook helpers now live outside the main site workflow module.
+
+### Changed
+- Moved bundle handoff command manifest construction, operator runbook assembly, and runbook Markdown line formatting into `cli/lib/site-bundle-handoff-runbook.mjs`.
+- Kept bundle validation, task selection, prompt construction, and final handoff report assembly in `cli/lib/site.mjs`.
+- Reused the extracted runbook helpers from the existing bundle handoff report path without changing JSON or human-readable output contracts.
+
+### Impact
+- `cli/lib/site.mjs` no longer owns the large deterministic operator runbook payload builder.
+- Bundle handoff JSON snapshots, task command manifests, evidence capture metadata, and target-repo prompt output remain behavior-compatible.
+- This leaves the remaining handoff flow easier to read because report orchestration is separated from runbook payload construction.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-handoff-runbook.mjs && node --check cli/lib/site.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+
+## Phase 675 — Website Improvement Bundle Compare Helper Split (unreleased)
+
+Website Improvement bundle compare summary, diff, and formatter helpers now live outside the main site workflow module.
+
+### Changed
+- Moved bundle compare summaries, metadata/file diff builders, and JSON/Markdown compare formatters into `cli/lib/site-bundle-compare.mjs`.
+- Kept bundle check orchestration and compare report assembly in `cli/lib/site.mjs` because compare report creation still depends on strict bundle validation.
+- Re-exported compare formatters from `cli/lib/site.mjs` so existing imports keep their public API.
+
+### Impact
+- `cli/lib/site.mjs` no longer owns deterministic compare formatting or checksum diff construction.
+- Bundle compare JSON and human-readable output keep the existing output contract.
+- This narrows the remaining bundle orchestration surface before larger check/handoff boundaries are split.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-compare.mjs && node --check cli/lib/site.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+
+## Phase 674 — Website Improvement Bundle Repair Helper Split (unreleased)
+
+Website Improvement bundle repair guidance helpers now live outside the main site workflow module.
+
+### Changed
+- Moved bundle repair command guidance, repair guidance Markdown lines, and before/after repair check summaries into `cli/lib/site-bundle-repair.mjs`.
+- Kept repair preview, repair bundle regeneration, applied repair verification, and bundle check orchestration in `cli/lib/site.mjs`.
+- Reused the extracted repair guidance in bundle check reports, repair reports, and target-repo handoff prompts.
+
+### Impact
+- `cli/lib/site.mjs` no longer owns repair command construction or repair check summary formatting.
+- Bundle repair preview/apply, strict bundle check, and handoff prompt repair sections keep their existing JSON/Markdown output contract.
+- This leaves repair execution in the orchestrator while moving deterministic repair guidance into a smaller module.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-repair.mjs && node --check cli/lib/site.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+
+## Phase 673 — Website Improvement Bundle File Helper Split (unreleased)
+
+Website Improvement bundle file, digest, and lightweight validation helpers now live outside the main site workflow module.
+
+### Changed
+- Moved SHA-256 digest creation, bundle checksum manifest hashing, bundle file reads, JSON parsing, Markdown fragment checks, short digest rendering, and array equality checks into `cli/lib/site-bundle-files.mjs`.
+- Kept bundle generation, generated-contract comparison, repair, check, compare, and target-repo handoff report assembly in `cli/lib/site.mjs`.
+- Reused the extracted helpers for bundle creation checksums, strict bundle verification, generated file drift reporting, and handoff prompt context reads.
+
+### Impact
+- `cli/lib/site.mjs` no longer owns low-level bundle file parsing or checksum mechanics.
+- Bundle check, repair, compare, handoff, and checksum validation keep the existing JSON/Markdown output contract.
+- This keeps the next bundle split safer because filesystem and digest helpers are now isolated from workspace analysis and report assembly.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-files.mjs && node --check cli/lib/site.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+
+## Phase 672 — Website Improvement Bundle Command Helper Split (unreleased)
+
+Website Improvement bundle handoff command and safety helpers now live outside the main site workflow module.
+
+### Changed
+- Moved bundle task handoff command args, check/handoff command args, shell quoting, output-file naming, and command safety policy helpers into `cli/lib/site-bundle-commands.mjs`.
+- Kept bundle check, repair, compare, and target-repo handoff report assembly in `cli/lib/site.mjs` so this split stays limited to deterministic command construction.
+- Reused the same helper outputs for task catalogs, operator runbooks, repair guidance, and source bundle verification prompts.
+
+### Impact
+- `cli/lib/site.mjs` no longer owns low-level bundle command string construction or local/read-only safety policy payloads.
+- Bundle repair guidance, strict bundle checks, task-specific handoff commands, and target-repo handoff prompts keep the existing command and JSON contract.
+- This prepares the remaining bundle check/repair/handoff logic for smaller future module boundaries without changing external behavior.
+
+### Verification Plan
+- `node --check cli/lib/site-bundle-commands.mjs && node --check cli/lib/site.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+
+## Phase 671 — Website Improvement Prompt Module Split (unreleased)
+
+Website Improvement prompt and handoff report builders now live outside the main site workflow module.
+
+### Changed
+- Moved Codex/Claude prompt rendering, task selection, ordered refactor-task sorting, and Website Improvement handoff report Markdown into `cli/lib/site-prompts.mjs`.
+- Moved audit category catalog and category lookup into `cli/lib/site-options.mjs` so prompts, workflow graphs, task normalization, and reports share one source.
+- Moved default implementation risk normalization into `cli/lib/site-evidence.mjs` so handoff reports and workspace normalization share the same evidence defaults.
+- Re-exported prompt and handoff helpers from `cli/lib/site.mjs` so existing imports keep working.
+
+### Impact
+- `cli/lib/site.mjs` no longer owns Codex/Claude prompt body composition or final handoff report rendering.
+- Bundle creation, bundle handoff, workflow graph, and prompt CLI flows keep their existing output contracts and local-only safety boundaries.
+- This keeps the Website Improvement CLI moving toward smaller, readable modules without weakening verification, evidence, or MCP readiness tracking.
+
+### Verification Plan
+- `node --check cli/lib/site-options.mjs && node --check cli/lib/site-evidence.mjs && node --check cli/lib/site-prompts.mjs && node --check cli/lib/site.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+
+## Phase 670 — Website Improvement Next Actions Module Split (unreleased)
+
+Website Improvement next-action reports now live outside the main site workflow module.
+
+### Changed
+- Moved standard, init, and intake next-action report builders plus next-action JSON/human formatting into `cli/lib/site-next-actions.mjs`.
+- Moved implementation evidence key/count helpers into `cli/lib/site-evidence.mjs` so next-action reports, bundles, and bundle checks share the same evidence counting rules.
+- Re-exported next-action helpers from `cli/lib/site.mjs` so existing imports keep working.
+
+### Impact
+- `cli/lib/site.mjs` no longer owns operator next-action prioritization or setup command rendering.
+- Website Improvement init, intake, readiness, task generation, prompt, handoff, and bundle guidance keep their existing command and output contracts.
+- This continues the split toward smaller modules while preserving verification, handoff evidence, and local-only safety boundaries.
+
+### Verification Plan
+- `node --check cli/lib/site-evidence.mjs && node --check cli/lib/site-next-actions.mjs && node --check cli/lib/site.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+
+## Phase 669 — Website Improvement MCP Report Module Split (unreleased)
+
+Website Improvement MCP readiness reports and action plans now live outside the main site workflow module.
+
+### Changed
+- Moved MCP status combination, MCP key normalization, task/MCP gap detection, readiness check report formatting, and MCP action plan rendering into `cli/lib/site-mcp-report.mjs`.
+- Moved shared MCP item and priority option catalogs into `cli/lib/site-options.mjs` so `site.mjs` and MCP report helpers can share them without circular imports.
+- Re-exported the public MCP report/action-plan helpers from `cli/lib/site.mjs` so existing imports keep working.
+
+### Impact
+- `cli/lib/site.mjs` no longer owns MCP readiness report construction or action-plan Markdown rendering.
+- Website Improvement MCP check, probe-aware MCP check, MCP action plan, next-actions, workflow graph, bundle, and handoff flows keep the same output contract.
+- This leaves `site.mjs` closer to orchestration while MCP handoff logic is grouped in a dedicated module for Claude/Codex MCP use.
+
+### Verification Plan
+- `node --check cli/lib/site-options.mjs && node --check cli/lib/site-mcp-report.mjs && node --check cli/lib/site.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+
+## Phase 668 — Website Improvement String Helper Split (unreleased)
+
+Website Improvement string normalization and Markdown rendering helpers now live outside the main site workflow module.
+
+### Changed
+- Moved `normalizeStringArray`, `markdownTable`, and `markdownList` into `cli/lib/site-strings.mjs`.
+- Kept the existing Markdown output shape for MCP action plans, workflow graphs, bundle READMEs, handoff reports, and prompt bundles.
+- Left browser-console helpers in `docs/website-console/app.js` untouched because this change only affects the CLI module boundary.
+
+### Impact
+- `cli/lib/site.mjs` no longer owns low-level string cleanup and Markdown table/list rendering.
+- Future Website Improvement action-plan and bundle modules can share the same small formatting helpers without circular imports.
+- Existing CLI JSON/human/Markdown output contracts remain unchanged.
+
+### Verification Plan
+- `node --check cli/lib/site-strings.mjs && node --check cli/lib/site.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+
+## Phase 667 — Website Improvement MCP Command Helper Split (unreleased)
+
+Website Improvement MCP readiness command strings now live outside the main site workflow module.
+
+### Changed
+- Moved MCP probe command generation, next-action command generation, and stdin command-target normalization into `cli/lib/site-mcp-commands.mjs`.
+- Kept the existing command names and output paths unchanged for `--mcp-check`, `--mcp-plan`, `--tasks`, `--prompt`, `--report`, and `--bundle` flows.
+- Left report assembly in `cli/lib/site.mjs` so this split stays limited to deterministic command construction.
+
+### Impact
+- `cli/lib/site.mjs` no longer owns repeated MCP/readiness command string construction.
+- MCP readiness reports, action plans, next-actions output, and bundle handoff guidance keep their current CLI command contract.
+- This continues the Website Improvement module split toward a cleaner Claude/Codex MCP handoff surface without changing runtime behavior.
+
+### Verification Plan
+- `node --check cli/lib/site-mcp-commands.mjs && node --check cli/lib/site.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+
+## Phase 666 — Website Improvement MCP Readiness Helper Split (unreleased)
+
+Website Improvement MCP readiness scoring now lives outside the main site workflow module.
+
+### Changed
+- Moved MCP readiness evidence collection, readiness item construction, and aggregate check status scoring into `cli/lib/site-mcp-readiness.mjs`.
+- Kept task/MCP gap detection in `cli/lib/site.mjs` because it still depends on normalized task metadata and local workspace structures.
+- Left the existing `buildSiteMcpCheckReport` JSON and human report contract unchanged.
+
+### Impact
+- `cli/lib/site.mjs` is smaller and now focuses on command/report assembly rather than low-level MCP readiness rules.
+- MCP readiness, probe output, task gap warnings, and generated next actions keep the same behavior for existing Website Improvement workspaces.
+- This continues the incremental Website Improvement module split while preserving MCP handoff evidence and verification surfaces.
+
+### Verification Plan
+- `node --check cli/lib/site-mcp-readiness.mjs && node --check cli/lib/site.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `python3 -B tools/audit/local-ci.py --docs-only`
+- `git diff --check`
+
+## Phase 665 — Website Improvement MCP Probe Helper Split (unreleased)
+
+Website Improvement read-only MCP probe logic now lives outside the main site workflow module.
+
+### Changed
+- Moved URL parsing, GitHub/Figma reference parsing, local repo directory probe checks, probe item construction, and `buildSiteMcpProbeReport` into `cli/lib/site-mcp-probes.mjs`.
+- Re-exported `buildSiteMcpProbeReport` from `cli/lib/site.mjs` so existing tests and consumers keep the same helper import surface.
+- Kept broader MCP readiness scoring and task/MCP gap logic in `cli/lib/site.mjs` for now because it still depends on workspace normalization and task ranking helpers.
+
+### Impact
+- `cli/lib/site.mjs` is smaller and its MCP readiness section now focuses on report assembly rather than low-level URL/path probe mechanics.
+- The probe report remains read-only, makes no external MCP calls, and keeps the same JSON/human output contract.
+- This continues the incremental Website Improvement module split without changing CLI flags, bundle checks, or handoff evidence surfaces.
+
+### Verification Plan
+- `node --check cli/lib/site-mcp-probes.mjs && node --check cli/lib/site.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `git diff --check`
+
+## Phase 664 — Website Improvement Intake Template Content Split (unreleased)
+
+Website Improvement intake template Markdown now lives with the static content catalog instead of the main workflow module.
+
+### Changed
+- Moved English and Korean intake template Markdown plus the intake section list into `cli/lib/site-content.mjs`.
+- Kept `buildSiteIntakeTemplateMarkdown` and `formatSiteIntakeTemplateJson` in `cli/lib/site.mjs` so command behavior and public helper names remain unchanged.
+
+### Impact
+- `cli/lib/site.mjs` no longer carries the long intake template bodies inline, making the executable workflow code easier to scan.
+- The generated English/Korean intake Markdown, JSON metadata, CLI flags, and tests remain unchanged.
+- This continues the incremental split of static Website Improvement content without reducing validation, bundle, or handoff evidence surfaces.
+
+### Verification Plan
+- `node --check cli/lib/site-content.mjs && node --check cli/lib/site.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `git diff --check`
+
+## Phase 663 — Website Improvement Static Content Module Split (unreleased)
+
+Website Improvement prompt and bundle catalogs now live outside the main site workflow module.
+
+### Changed
+- Moved Website Improvement prompt template ids, prompt template metadata, bundle file manifest, and bundle checksum manifest into `cli/lib/site-content.mjs`.
+- Re-exported the public constants from `cli/lib/site.mjs` so existing tests and consumers keep the same import surface.
+- Kept the prompt template ordering assertion next to the catalog data it protects.
+
+### Impact
+- `cli/lib/site.mjs` is smaller and its executable workflow logic is easier to scan.
+- No CLI behavior, JSON contract, bundle manifest, prompt catalog, or test import path changes.
+- This is the first low-risk split toward reducing the oversized Website Improvement module without reducing verification or evidence surfaces.
+
+### Verification Plan
+- `node --check cli/lib/site-content.mjs && node --check cli/lib/site.mjs`
+- `node --test cli/lib/site.test.mjs`
+- `npm test`
+- `npm run audit:strict`
+- `npm run release:metadata`
+- `git diff --check`
+
+## Phase 662 — VS Code Extension Dev Dependency Audit Hygiene (unreleased)
+
+VS Code extension development dependencies now install with a clean npm audit report.
+
+### Changed
+- Pinned the Mocha integration-test harness at `11.3.0` to avoid the vulnerable newer `diff` range while keeping the extension e2e runner on the current major.
+- Added npm `overrides` for `js-yaml` and `serialize-javascript` so transitive Mocha audit findings resolve without changing extension runtime dependencies.
+- Refreshed `vscode-extension/package-lock.json` from the updated dependency policy.
+
+### Impact
+- `npm audit --prefix vscode-extension --audit-level=moderate` now reports zero vulnerabilities.
+- The published VSIX runtime surface remains unchanged because the extension still has no production dependencies.
+- The integration test harness still uses Mocha and remains covered by the VS Code extension e2e test.
+
+### Verification Plan
+- `npm audit --prefix vscode-extension --audit-level=moderate`
+- `npm run compile --prefix vscode-extension`
+- `npm test --prefix vscode-extension`
+- `npm run test:e2e --prefix vscode-extension`
+- `npm test`
+- `npm run audit:strict`
+- `git diff --check`
+
 ## Phase 661 — Website Console Revalidation Gate JSON Action Labels (unreleased)
 
 Website Console revalidation gate action buttons now identify that they produce compact gate JSON.
@@ -19474,18 +21698,18 @@ Coverage:
 - [x] Component spec coverage push: 11 → 19 worked specs (5% → 8.5%). New: Alert, Tooltip, Form-controls (Switch/Checkbox/Radio combined), Skeleton, Progress, Avatar, Breadcrumb, Accordion.
 - [ ] Optional embedding index if knowledge base exceeds 100K tokens. _(Deferred — base is currently ~13K lines / well under threshold.)_
 
-## Phase 4 — Multi-tool
+## Phase 4 — Multi-tool ✓ shipped
 
-- [ ] Codex CLI: real-world session against this repo, captured as a worked example.
-- [ ] Cursor `.cursorrules` overlay.
-- [ ] Aider configuration.
+- [x] Codex CLI: real-world session against this repo, captured as a worked example. _([docs/integrations/codex-walkthrough.md](integrations/codex-walkthrough.md))_
+- [x] Cursor `.cursorrules` overlay. _([docs/integrations/cursor-walkthrough.md](integrations/cursor-walkthrough.md))_
+- [x] Aider configuration. _([docs/integrations/aider-walkthrough.md](integrations/aider-walkthrough.md))_
 
-## Phase 5 — Maturity
+## Phase 5 — Maturity (public site / packaging shipped)
 
-- [ ] Versioned knowledge files (semver headers).
+- [x] Versioned knowledge files (semver headers). _(frontmatter `version` field on every knowledge file, enforced by the frontmatter audit)_
 - [ ] CHANGELOG that summarizes upstream-source updates affecting `refs/`.
-- [ ] Public site (knowledge/ as a browsable doc site).
-- [ ] Plugin packaging — install design-ai as a Claude Code plugin / VS Code extension.
+- [x] Public site (knowledge/ as a browsable doc site). _(GitHub Pages: https://sungjin9288.github.io/design-ai/)_
+- [x] Plugin packaging — install design-ai as a Claude Code plugin / VS Code extension. _(`.claude-plugin/plugin.json` + VS Code Marketplace `sungjin.design-ai-vscode` v0.4.1)_
 
 ## Out of scope