buildanything 2.1.1 → 2.2.0

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "buildanything",
-  "version": "2.1.1",
+  "version": "2.2.0",
   "description": "One command to build an entire product. 44 specialist agents orchestrated into a full engineering pipeline for Claude Code.",
   "bin": {
     "buildanything": "./bin/setup.js",

package/protocols/ios-phase-branches.md

@@ -134,11 +134,11 @@ Confirm all of: Xcode version OK, `.xcodeproj` exists, XcodeBuildMCP responds, a
 
 ## Phase 1 — Plan (iOS additions)
 
-Load phase-specific iOS skill bundle per `protocols/ios-context.md` §Phase 1. The Phase 1.1 research team is the same 4 agents as the web branch (`subagent_type: feature-intel`, `subagent_type: tech-feasibility`, `subagent_type: design-ux-researcher`, `subagent_type: business-model`) but each prompt must additionally check App Store category landscape, TestFlight constraints, and iOS 26 API availability (via apple-docs-mcp). For AI / Foundation Models prompts, additionally dispatch `subagent_type: ios-foundation-models-specialist`. Note: the Phase 2.3 sprint-breakdown `subagent_type: planner` is replaced by `subagent_type: ios-swift-architect` for iOS mode (see Phase 2 additions below).
+Load phase-specific iOS skill bundle per `protocols/ios-context.md` §Phase 1. The Phase 1.1 research team is the same 4 agents as the web branch (`agent_type: feature-intel` — `subagent_type: feature-intel`, `agent_type: tech-feasibility` — `subagent_type: tech-feasibility`, `agent_type: design-ux-researcher` — `subagent_type: design-ux-researcher`, `agent_type: business-model` — `subagent_type: business-model`) but each prompt must additionally check App Store category landscape, TestFlight constraints, and iOS 26 API availability (via apple-docs-mcp). For AI / Foundation Models prompts, additionally dispatch `agent_type: ios-foundation-models-specialist` — `subagent_type: ios-foundation-models-specialist`. Note: the Phase 2.3 sprint-breakdown `agent_type: planner` — `subagent_type: planner` is replaced by `agent_type: ios-swift-architect` — `subagent_type: ios-swift-architect` for iOS mode (see Phase 2 additions below).
 
 ## Phase 2 — Architecture (iOS additions)
 
-Load phase-specific iOS skill bundle per `protocols/ios-context.md` §Phase 2. Architecture agents must select iOS 26 APIs via apple-docs-mcp (verify availability, deprecations, minimum OS). Replace the web `subagent_type: engineering-backend-architect` / `subagent_type: engineering-frontend-developer` architecture dispatches with a single `subagent_type: ios-swift-architect` dispatch covering: (1) SwiftUI view hierarchy + navigation model, (2) SwiftData schema + CloudKit strategy, (3) Swift Concurrency / actor isolation plan, (4) iOS-specific security (Keychain, entitlements, ATS). Implementation blueprint lists Swift files + Xcode targets, not web modules. Security architecture stays on `subagent_type: engineering-security-engineer` (unchanged from web branch — the security engineer handles both stacks).
+Load phase-specific iOS skill bundle per `protocols/ios-context.md` §Phase 2. Architecture agents must select iOS 26 APIs via apple-docs-mcp (verify availability, deprecations, minimum OS). Replace the web `agent_type: engineering-backend-architect` — `subagent_type: engineering-backend-architect` / `agent_type: engineering-frontend-developer` — `subagent_type: engineering-frontend-developer` architecture dispatches with a single `agent_type: ios-swift-architect` — `subagent_type: ios-swift-architect` dispatch covering: (1) SwiftUI view hierarchy + navigation model, (2) SwiftData schema + CloudKit strategy, (3) Swift Concurrency / actor isolation plan, (4) iOS-specific security (Keychain, entitlements, ATS). Implementation blueprint lists Swift files + Xcode targets, not web modules. Security architecture stays on `agent_type: engineering-security-engineer` — `subagent_type: engineering-security-engineer` (unchanged from web branch — the security engineer handles both stacks).
 
 The iOS architect (and the iOS-context security architect) MUST also read `docs/plans/product-spec.md` — the Screen Inventory drives SwiftUI view hierarchy, per-feature Persona Constraints drive HIG navigation pattern choices (TabView vs NavigationStack vs sheets), and per-feature Permissions & Roles drive Keychain + entitlement scopes.
 
@@ -171,7 +171,7 @@ Before the Quality Gate 2 approval prompt in `commands/build.md` is rendered, di
 
 Call the Agent tool once:
 
-1. Description: "iOS visual direction preview" — subagent_type: `ios-swift-ui-design` — prompt: "[CONTEXT header above — phase: 2] Read `docs/plans/design-doc.md` (#persona, #scope, #voice), `docs/plans/phase1-scratch/findings-digest.md`, and `docs/plans/architecture.md`. Emit a 3-5 bullet DIRECTIONAL preview of the intended iOS visual direction — one-line brand read, then proposed leanings on: navigation pattern (TabView vs NavigationStack vs sheets), typography (Dynamic Type scale + tone), color (semantic + dark mode leaning), motion/material feel (Liquid Glass on iOS 26+ yes/no, haptic-forward yes/no), SF Symbol family vibe. NO rationale paragraphs, NO reference citations. Save to `docs/plans/visual-dna-preview.md` as a flat bullet list. Target 150 tokens, max 250."
+1. Description: "iOS visual direction preview" — agent_type: `ios-swift-ui-design` — subagent_type: `ios-swift-ui-design` — prompt: "[CONTEXT header above — phase: 2] Read `docs/plans/design-doc.md` (#persona, #scope, #voice), `docs/plans/phase1-scratch/findings-digest.md`, and `docs/plans/architecture.md`. Emit a 3-5 bullet DIRECTIONAL preview of the intended iOS visual direction — one-line brand read, then proposed leanings on: navigation pattern (TabView vs NavigationStack vs sheets), typography (Dynamic Type scale + tone), color (semantic + dark mode leaning), motion/material feel (Liquid Glass on iOS 26+ yes/no, haptic-forward yes/no), SF Symbol family vibe. NO rationale paragraphs, NO reference citations. Save to `docs/plans/visual-dna-preview.md` as a flat bullet list. Target 150 tokens, max 250."
 
 Output: `docs/plans/visual-dna-preview.md` — surfaced by the orchestrator in the Gate 2 prompt. Phase 3.0 + Phase 3.2-ios together produce the full `DESIGN.md`; the preview is discarded after Gate 2 approval.
 
@@ -179,11 +179,11 @@ Output: `docs/plans/visual-dna-preview.md` — surfaced by the orchestrator in t
 
 Load phase-specific iOS skill bundle per `protocols/ios-context.md` §Phase 3. Do **NOT** build `/design-system` (web-only). The artifact is `DESIGN.md` at the repo root, same format as web — see `protocols/design-md-authoring.md` for the contract and §9 for iOS-specific rules.
 
-- **Step 3.0 iOS** — same dispatch as web: `subagent_type: design-brand-guardian` writes Pass 1 of `DESIGN.md` (Overview + 7-axis Brand DNA + Rationale + Locked At + References + Do's and Don'ts). Pass 2 sections present as placeholders. The Brand Guardian honors iOS-specific Material gating (Liquid Glass requires iOS 26+ target).
+- **Step 3.0 iOS** — same dispatch as web: `agent_type: design-brand-guardian` — `subagent_type: design-brand-guardian` writes Pass 1 of `DESIGN.md` (Overview + 7-axis Brand DNA + Rationale + Locked At + References + Do's and Don'ts). Pass 2 sections present as placeholders. The Brand Guardian honors iOS-specific Material gating (Liquid Glass requires iOS 26+ target).
 - **Step 3.0.idx iOS** — after `design-brand-guardian` returns and `DESIGN.md` is on disk, index it into the build graph. Slice 2 graph index — required for downstream agents. Run via the Bash tool: `node ${CLAUDE_PLUGIN_ROOT}/bin/graph-index.js DESIGN.md`. On exit 0, log success to `docs/plans/build-log.md` and continue. On non-zero exit, STOP — log the error to `docs/plans/build-log.md` and report the failure. Downstream agents require the graph.
-- **Step 3.1 iOS** — dispatch `subagent_type: visual-research` with the agent-browser (`-p desktop`) skill to harvest iOS UI references from **free** sources: screenlane.com (iOS screenshots), App Store web listings for top apps in the product category, Apple HIG pages, SF Symbols browser. No Mobbin (paid). Fallback: vibe-only design board if scraping blocked.
-- **Step 3.2-ios** — dispatch `subagent_type: ios-swift-ui-design` to write Pass 2 of `DESIGN.md`. Fills YAML front matter (`colors` with `-dark` pairs per §9.2; `typography` named after Dynamic Type roles; `rounded` for continuous corners; `spacing` on the HIG 4/8/16/20/24 scale; `components` covering at minimum the iOS vocabulary in §9.3 — nav-tab-bar, list-row, card-elevated, button-primary, input-text, sheet-modal, etc.) AND writes Pass 2 prose for `## Colors`, `## Typography`, `## Layout`, `## Elevation & Depth`, `## Shapes`, `## Components`. Pass 1 sections are READ-ONLY at this step. Grounded in Apple HIG + Liquid Glass (iOS 26+ when DNA Material = Glassy) + SF Symbols + the harvested references + the user's stated app vibe.
-- **Step 3.3-ios** — dispatch `subagent_type: design-ux-architect` to write `docs/plans/ux-architecture.md` + `docs/plans/page-specs/*.md` (one file per screen from product-spec Screen Inventory). Same agent as web — the agent already understands both platforms via skill gating. Reads:
+- **Step 3.1 iOS** — dispatch `agent_type: visual-research` — `subagent_type: visual-research` with the agent-browser (`-p desktop`) skill to harvest iOS UI references from **free** sources: screenlane.com (iOS screenshots), App Store web listings for top apps in the product category, Apple HIG pages, SF Symbols browser. No Mobbin (paid). Fallback: vibe-only design board if scraping blocked.
+- **Step 3.2-ios** — dispatch `agent_type: ios-swift-ui-design` — `subagent_type: ios-swift-ui-design` to write Pass 2 of `DESIGN.md`. Fills YAML front matter (`colors` with `-dark` pairs per §9.2; `typography` named after Dynamic Type roles; `rounded` for continuous corners; `spacing` on the HIG 4/8/16/20/24 scale; `components` covering at minimum the iOS vocabulary in §9.3 — nav-tab-bar, list-row, card-elevated, button-primary, input-text, sheet-modal, etc.) AND writes Pass 2 prose for `## Colors`, `## Typography`, `## Layout`, `## Elevation & Depth`, `## Shapes`, `## Components`. Pass 1 sections are READ-ONLY at this step. Grounded in Apple HIG + Liquid Glass (iOS 26+ when DNA Material = Glassy) + SF Symbols + the harvested references + the user's stated app vibe.
+- **Step 3.3-ios** — dispatch `agent_type: design-ux-architect` — `subagent_type: design-ux-architect` to write `docs/plans/ux-architecture.md` + `docs/plans/page-specs/*.md` (one file per screen from product-spec Screen Inventory). Same agent as web — the agent already understands both platforms via skill gating. Reads:
   - Product spec: `docs/plans/product-spec.md` (FULL — Screen Inventory is the screen list; per-feature sections define what each screen does, what data it shows, what states exist, persona constraints, business rules)
   - DESIGN.md `## Overview > ### Brand DNA` (Density axis drives layout — Airy uses generous safe-area margins; Dense uses HIG-minimum spacing. Character + Motion shape navigation transitions)
   - DESIGN.md YAML `components:` block (the iOS component vocabulary the wireframes compose from — nav-tab-bar, list-row, card-elevated, button-primary, etc.)
@@ -203,9 +203,9 @@ Load phase-specific iOS skill bundle per `protocols/ios-context.md` §Phase 3. D
 
   DESIGN.md Pass 2 (exact spacing values, typography ramp YAML) already exists at this point — Step 3.2-ios produced it. Page-specs reference DESIGN.md token names, not raw values; the SwiftUI translator at Step 4.0.b emits the constants.
 - **Step 3.3.idx iOS** — after `design-ux-architect` returns and `docs/plans/page-specs/` is populated with one .md file per screen, index the directory into the build graph. Slice 3 graph index — required for downstream agents. Run via the Bash tool: `node ${CLAUDE_PLUGIN_ROOT}/bin/graph-index.js docs/plans/page-specs/`. On exit 0, log success to `docs/plans/build-log.md` and continue. On non-zero exit, STOP — log the error to `docs/plans/build-log.md` and report the failure. Downstream agents require the graph — do not proceed without a successful index.
-- **Step 3.3b-ios** — dispatch `subagent_type: design-ux-researcher` to validate the iOS UX flows against persona/JTBD. Reads `docs/plans/ux-architecture.md`, `docs/plans/page-specs/`, `docs/plans/product-spec.md`, `docs/plans/design-doc.md`, DESIGN.md `### Brand DNA`. Walk each user flow as the target persona on an iPhone — narrate the steps, flag friction points, check HIG conformance (gesture discoverability, tap target ≥44pt, navigation depth), check critical tasks reachable in minimum taps, check Dynamic Type at xxxLarge doesn't break flows. Writes `docs/plans/ux-flow-validation.md`. Critical issues route back to Step 3.3-ios.
+- **Step 3.3b-ios** — dispatch `agent_type: design-ux-researcher` — `subagent_type: design-ux-researcher` to validate the iOS UX flows against persona/JTBD. Reads `docs/plans/ux-architecture.md`, `docs/plans/page-specs/`, `docs/plans/product-spec.md`, `docs/plans/design-doc.md`, DESIGN.md `### Brand DNA`. Walk each user flow as the target persona on an iPhone — narrate the steps, flag friction points, check HIG conformance (gesture discoverability, tap target ≥44pt, navigation depth), check critical tasks reachable in minimum taps, check Dynamic Type at xxxLarge doesn't break flows. Writes `docs/plans/ux-flow-validation.md`. Critical issues route back to Step 3.3-ios.
 - **Skip Step 3.3** (Living Style Guide) — no web route. The visual-design surface is the SwiftUI Preview captures from Step 3.4-ios.
-- **Step 3.4-ios** — Per `protocols/metric-loop.md` Step 0.5, extract scoring criteria from `DESIGN.md` (HIG values from spacing/typography YAML, navigation pattern from components, color tokens, SF Symbol choices from prose) into the Scoring Criteria Checklist. Extraction is **mechanical** — `DESIGN.md` has structured YAML + named prose sections. Persist to `active_metric_loop.scoring_criteria_checklist` in `.build-state.json`. Visual QA loop uses XcodeBuildMCP SwiftUI Preview captures (not Playwright screenshots). The loop runs `subagent_type: ios-swift-ui-design` as the generator (Preview tweaks AND DESIGN.md token re-tunes per critic findings) paired with `subagent_type: design-critic` as the critic. Critic receives the checklist + fresh Preview captures each iteration (NOT the full `DESIGN.md`). Generator re-invocation on iteration 2+ follows the lean context rule (top issue + file paths + relevant checklist values only). Exit criterion = user-approved pass/fail (not a 0-100 rubric). **Max 3 iterations** (tighter than web's 5). On stall or max iterations, present the score history to the user.
+- **Step 3.4-ios** — Per `protocols/metric-loop.md` Step 0.5, extract scoring criteria from `DESIGN.md` (HIG values from spacing/typography YAML, navigation pattern from components, color tokens, SF Symbol choices from prose) into the Scoring Criteria Checklist. Extraction is **mechanical** — `DESIGN.md` has structured YAML + named prose sections. Persist to `active_metric_loop.scoring_criteria_checklist` in `.build-state.json`. Visual QA loop uses XcodeBuildMCP SwiftUI Preview captures (not Playwright screenshots). The loop runs `agent_type: ios-swift-ui-design` — `subagent_type: ios-swift-ui-design` as the generator (Preview tweaks AND DESIGN.md token re-tunes per critic findings) paired with `agent_type: design-critic` — `subagent_type: design-critic` as the critic. Critic receives the checklist + fresh Preview captures each iteration (NOT the full `DESIGN.md`). Generator re-invocation on iteration 2+ follows the lean context rule (top issue + file paths + relevant checklist values only). Exit criterion = user-approved pass/fail (not a 0-100 rubric). **Max 3 iterations** (tighter than web's 5). On stall or max iterations, present the score history to the user.
 - **Step 3.4.idx iOS** — after `ios-swift-ui-design` completes the visual QA loop (which may re-tune DESIGN.md tokens), re-run the indexer on DESIGN.md. The CLI dispatch detects Pass 2 content and writes `slice-3-tokens.json` alongside the existing `slice-2-dna.json`. Run via the Bash tool: `node ${CLAUDE_PLUGIN_ROOT}/bin/graph-index.js DESIGN.md`. On exit 0, log success to `docs/plans/build-log.md` and continue. On non-zero exit, STOP — log the error to `docs/plans/build-log.md` and report the failure. Downstream agents require the graph — do not proceed without a successful index.
 - **Step 3.8 iOS lint** — same lint hook as web (`hooks/design-md-lint`). Broken-refs is a hard fail and routes back to Step 3.2-ios. Warnings logged. Three iOS-specific post-process checks per §9.5 (dark-pair rule, Dynamic Type role check, iOS 26 gating) layer on top of the vendored linter — codepath in `hooks/design-md-lint.ts` gated on `project_type=ios`.
 
@@ -217,7 +217,7 @@ Load phase-specific iOS skill bundle per `protocols/ios-context.md` §Phase 4
 
 Phase 4 in the iOS branch contains the Step 4.0 Scaffold work (iOS project bootstrap follow-up, entitlements, Info.plist, XcodeBuildMCP folder structure, SwiftUI design tokens, Maestro flow stubs). Per-task implementation (Step 4.1+) is handled inline in the "Phase 4 — Build per-task flow (iOS branch)" section below.
 
-Dispatch the `ios-entitlements-generator` skill (Info.plist + entitlements based on features: push, background, HealthKit, etc.) and the `ios-info-plist-hardening` skill (ATS config, privacy usage strings, URL schemes). Both live under `skills/ios/` and are loaded as skill bundles, not agents — they inherit the active implementer's `subagent_type` rather than being dispatched standalone. The active implementer for Phase 4 scaffold work is `subagent_type: engineering-senior-developer` (inheriting the `ios-context.md` persona).
+Dispatch the `ios-entitlements-generator` skill (Info.plist + entitlements based on features: push, background, HealthKit, etc.) and the `ios-info-plist-hardening` skill (ATS config, privacy usage strings, URL schemes). Both live under `skills/ios/` and are loaded as skill bundles, not agents — they inherit the active implementer's `subagent_type` rather than being dispatched standalone. The active implementer for Phase 4 scaffold work is `agent_type: engineering-senior-developer` — `subagent_type: engineering-senior-developer` (inheriting the `ios-context.md` persona).
 
 - **Step 4.0.a (iOS):** Scaffolding is already done by Phase -1 Bootstrap. Instead, create the app target's folder structure (`Views/`, `Models/`, `Services/`, `Resources/`) via XcodeBuildMCP.
 - **Step 4.0.b (iOS):** Implement iOS-native design tokens from `DESIGN.md` (YAML `colors`, `typography`, `rounded`, `spacing` blocks). Write `Sources/<target>/DesignTokens.swift` per the SwiftUI translator template in `protocols/design-md-authoring.md` §9.4 — emits `Color` extensions (Asset Catalog–backed; the `-dark` color pairs populate the dark appearance), `Font` extensions (Dynamic Type roles map to `Font.TextStyle`), `Spacing` and `Radius` enums (CGFloat constants; use radius with `.continuous` corner style). Also create `Resources/Assets.xcassets` color set entries — one per `colors:` token, with the `-dark` variant populating the Dark appearance slot. Component tokens (`button-primary`, `card-elevated`, etc.) are applied via SwiftUI view modifiers in per-screen views — NOT translated to Swift directly. NOT web CSS.
@@ -227,7 +227,7 @@ Dispatch the `ios-entitlements-generator` skill (Info.plist + entitlements based
 
 **Step 4.0.d build-fix dispatch (iOS):** When the scaffold-health metric loop hits an `xcodebuild` failure, the orchestrator MUST dispatch the Swift build resolver rather than re-running the scaffolder blindly.
 
-Call the Agent tool — description: "Swift build fix (scaffold)" — subagent_type: `swift-build-resolver` — mode: "bypassPermissions" — prompt: "[CONTEXT header above — phase: 4] xcodebuild failed with this error: [paste]. Apply the minimal diff to fix the specific error. No architectural edits, no dependency changes, no refactors. Confirm green before returning."
+Call the Agent tool — description: "Swift build fix (scaffold)" — agent_type: `swift-build-resolver` — subagent_type: `swift-build-resolver` — mode: "bypassPermissions" — prompt: "[CONTEXT header above — phase: 4] xcodebuild failed with this error: [paste]. Apply the minimal diff to fix the specific error. No architectural edits, no dependency changes, no refactors. Confirm green before returning."
 
 If the resolver returns `status: blocked` (architectural change required), the orchestrator returns to Step 4.0.a/4.0.b with the blocker surfaced — the resolver is NOT permitted to restructure foundation types.
 
@@ -239,7 +239,7 @@ find maestro -name '*.yaml' -type f | wc -l
 
 The result MUST be `>= 2`. If less than 2, re-dispatch the Maestro stub scaffolder ONCE (max 1 retry) via:
 
-Call the Agent tool — description: "Maestro stub scaffold (retry)" — subagent_type: `engineering-senior-developer` — mode: "bypassPermissions" — prompt: "[CONTEXT header above — phase: 4] Step 4.0.c regression: Maestro flow stubs were not scaffolded. Load the `skills/ios/ios-maestro-flow-author/` skill bundle and scaffold at least 2 .yaml flow stubs in `maestro/` per the sprint-tasks.md Behavioral Test fields. Do NOT touch other files."
+Call the Agent tool — description: "Maestro stub scaffold (retry)" — agent_type: `engineering-senior-developer` — subagent_type: `engineering-senior-developer` — mode: "bypassPermissions" — prompt: "[CONTEXT header above — phase: 4] Step 4.0.c regression: Maestro flow stubs were not scaffolded. Load the `skills/ios/ios-maestro-flow-author/` skill bundle and scaffold at least 2 .yaml flow stubs in `maestro/` per the sprint-tasks.md Behavioral Test fields. Do NOT touch other files."
 
 After the retry, re-run the `find` command. If still `< 2`, HALT the build with directive: "Step 4.0.c regression: Maestro flow stubs were not scaffolded. Return to Step 4.0.c before proceeding." Do NOT advance to Step 4.0.d / Step 4.0.e / Step 4.1+ per-task flow until the assertion passes.
 
@@ -262,7 +262,7 @@ Load full iOS skill bundle per `protocols/ios-context.md` §Phase 4 — Build (S
 
 ### Step 4.1 — Implement (iOS)
 
-Call the Agent tool — description: "[task name]" — subagent_type: `[from BO brief]` — mode: "bypassPermissions" — prompt: "[CONTEXT header above — phase: 4] [COMPLEXITY: S/M/L from sprint-tasks.md].
+Call the Agent tool — description: "[task name]" — agent_type: `[from BO brief]` — subagent_type: `[from BO brief]` — mode: "bypassPermissions" — prompt: "[CONTEXT header above — phase: 4].
 
 TASK: [task description from BO brief]
 
@@ -296,7 +296,7 @@ Return deviation_row or null. Do NOT write decisions.jsonl directly.
 
 Implement fully with real code and tests. Commit: 'feat: [task]'."
 
-Implementation agents edit Swift files directly and build/diagnose via XcodeBuildMCP. Set `[COMPLEXITY: S/M/L]` based on the task's Size from sprint-tasks.md.
+Implementation agents edit Swift files directly and build/diagnose via XcodeBuildMCP.
 
 **Agent selection table for Step 4.1 (keyed on `ios_features.*` + task kind):**
 
@@ -317,16 +317,16 @@ Precedence rule: if a task matches multiple rows, the most specific (top-down) w
 
 After every Step 4.1 implementer returns (and before Step 4.2 Metric Loop / verify), run a Swift-specific review pass to catch concurrency / SwiftUI / protocol-DI issues the generic code-reviewer misses. Run in parallel with the generic code-reviewer + silent-failure-hunter pair from `commands/build.md` per-task review block.
 
-Call the Agent tool — description: "Swift review: [task name]" — subagent_type: `swift-reviewer` — prompt: "[CONTEXT header above — phase: 4] Review the Swift changes in this task. Task: [name]. Files changed: [list]. Walk the CRITICAL → HIGH → MEDIUM checklist for Swift concurrency 6.2, SwiftUI patterns, protocol DI testability, and Foundation Models integration. Confidence-filter at 80%."
+Call the Agent tool — description: "Swift review: [task name]" — agent_type: `swift-reviewer` — subagent_type: `swift-reviewer` — prompt: "[CONTEXT header above — phase: 4] Review the Swift changes in this task. Task: [name]. Files changed: [list]. Walk the CRITICAL → HIGH → MEDIUM checklist for Swift concurrency 6.2, SwiftUI patterns, protocol DI testability, and Foundation Models integration. Confidence-filter at 80%."
 
-For auth / PII / Keychain / credential tasks, also dispatch `subagent_type: security-reviewer` per the build.md per-task review block.
+For auth / PII / Keychain / credential tasks, also dispatch `agent_type: security-reviewer` — `subagent_type: security-reviewer` per the build.md per-task review block.
 
 ### Step 4.1c — Cleanup (iOS)
 
 Run the code-simplifier + refactor-cleaner pair from `commands/build.md` per-task cleanup block against the Swift changeset. Swift dead-code detection relies on SwiftLint / xcodebuild warnings rather than `knip` / `depcheck` — the refactor-cleaner runs in a Swift-aware mode.
 
-1. Call the Agent tool — description: "Simplify [task name]" — subagent_type: `code-simplifier` — mode: "bypassPermissions" — prompt: "[CONTEXT header above — phase: 4] Simplify changed Swift files from [task]. Remove dead code, unused imports, redundant abstractions. Do NOT add features. Do NOT change architecture. Do NOT touch files outside the changeset. Files: [list]."
-2. Call the Agent tool — description: "Refactor-clean [task name]" — subagent_type: `refactor-cleaner` — mode: "bypassPermissions" — prompt: "[CONTEXT header above — phase: 4] Clean dead code from changed Swift files in [task]. Run SwiftLint / xcodebuild warning sweep and remove orphaned helpers, unused types, dead functions. Same scope rules — changeset only. Files: [list]."
+1. Call the Agent tool — description: "Simplify [task name]" — agent_type: `code-simplifier` — subagent_type: `code-simplifier` — mode: "bypassPermissions" — prompt: "[CONTEXT header above — phase: 4] Simplify changed Swift files from [task]. Remove dead code, unused imports, redundant abstractions. Do NOT add features. Do NOT change architecture. Do NOT touch files outside the changeset. Files: [list]."
+2. Call the Agent tool — description: "Refactor-clean [task name]" — agent_type: `refactor-cleaner` — subagent_type: `refactor-cleaner` — mode: "bypassPermissions" — prompt: "[CONTEXT header above — phase: 4] Clean dead code from changed Swift files in [task]. Run SwiftLint / xcodebuild warning sweep and remove orphaned helpers, unused types, dead functions. Same scope rules — changeset only. Files: [list]."
 
 ### Step 4.2 — Metric Loop (iOS)
 
@@ -336,7 +336,7 @@ Metric loop uses XcodeBuildMCP SwiftUI Preview captures for UI verification (not
 
 **Build-fix dispatch (iOS):** When `xcodebuild` fails during the metric loop (or during Step 4.1 implementer return), the orchestrator MUST spawn the Swift build resolver rather than asking the generic implementer to guess at the error.
 
-Call the Agent tool — description: "Swift build fix" — subagent_type: `swift-build-resolver` — mode: "bypassPermissions" — prompt: "[CONTEXT header above — phase: 4] xcodebuild failed with this error: [paste]. Apply the minimal diff to fix the specific error. No architectural edits, no dependency changes, no refactors. Confirm green before returning."
+Call the Agent tool — description: "Swift build fix" — agent_type: `swift-build-resolver` — subagent_type: `swift-build-resolver` — mode: "bypassPermissions" — prompt: "[CONTEXT header above — phase: 4] xcodebuild failed with this error: [paste]. Apply the minimal diff to fix the specific error. No architectural edits, no dependency changes, no refactors. Confirm green before returning."
 
 If the resolver returns `status: blocked` (architectural or dependency change required), the orchestrator hands back to the Step 4.1 implementer with the resolver's `blocking_error` payload so the implementer can make an informed architectural fix — the resolver is NOT permitted to restructure types.
 
@@ -364,15 +364,15 @@ Phase 5 runs in three layers matching the web structure: Track A (engineering re
 
 Call the Agent tool 5 times in one message:
 
-1. **API Contract** — subagent_type: `testing-api-tester` — Run network integration tests via XcodeBuildMCP test runner. Validate URLSession/networking layer against architecture.md API contracts. Evidence: `docs/plans/evidence/track-a/api-contract.json`
+1. **API Contract** — agent_type: `testing-api-tester` — subagent_type: `testing-api-tester` — Run network integration tests via XcodeBuildMCP test runner. Validate URLSession/networking layer against architecture.md API contracts. Evidence: `docs/plans/evidence/track-a/api-contract.json`
 
-2. **Performance** — subagent_type: `testing-performance-benchmarker` — iOS-adapted: app launch time (cold/warm via XcodeBuildMCP), memory footprint, binary size budget, scroll jank. Use `xcodebuild -showBuildTimingSummary`. Compare against `quality-targets.json`. Evidence: `docs/plans/evidence/track-a/performance.json`
+2. **Performance** — agent_type: `testing-performance-benchmarker` — subagent_type: `testing-performance-benchmarker` — iOS-adapted: app launch time (cold/warm via XcodeBuildMCP), memory footprint, binary size budget, scroll jank. Use `xcodebuild -showBuildTimingSummary`. Compare against `quality-targets.json`. Evidence: `docs/plans/evidence/track-a/performance.json`
 
-3. **Accessibility** — subagent_type: `a11y-architect` — Load `swift-accessibility` skill (Mode 3: audit pass). XcodeBuildMCP `describe_ui` for accessibility tree inspection. VoiceOver labels, Dynamic Type at all sizes, contrast ratios, hit targets ≥44pt. Evidence: `docs/plans/evidence/track-a/accessibility.json`
+3. **Accessibility** — agent_type: `a11y-architect` — subagent_type: `a11y-architect` — Load `swift-accessibility` skill (Mode 3: audit pass). XcodeBuildMCP `describe_ui` for accessibility tree inspection. VoiceOver labels, Dynamic Type at all sizes, contrast ratios, hit targets ≥44pt. Evidence: `docs/plans/evidence/track-a/accessibility.json`
 
-4. **Security** — subagent_type: `engineering-security-engineer` — Load `swift-security-expert` skill (audit mode). Keychain usage, CryptoKit, ATS exceptions, privacy manifest, entitlements, hardcoded secrets, `swift package audit`. Evidence: `docs/plans/evidence/track-a/security.json`
+4. **Security** — agent_type: `engineering-security-engineer` — subagent_type: `engineering-security-engineer` — Load `swift-security-expert` skill (audit mode). Keychain usage, CryptoKit, ATS exceptions, privacy manifest, entitlements, hardcoded secrets, `swift package audit`. Evidence: `docs/plans/evidence/track-a/security.json`
 
-5. **Brand Drift** — subagent_type: `design-brand-guardian` — Capture every screen via XcodeBuildMCP simulator screenshots. Score against DESIGN.md DNA axes (Character, Density, Material, Motion, Type). Save screenshots to `docs/plans/evidence/brand-drift/`. Findings to `docs/plans/evidence/brand-drift.md`. Drift check only — no pass/fail verdict.
+5. **Brand Drift** — agent_type: `design-brand-guardian` — subagent_type: `design-brand-guardian` — Capture every screen via XcodeBuildMCP simulator screenshots. Score against DESIGN.md DNA axes (Character, Density, Material, Motion, Type). Save screenshots to `docs/plans/evidence/brand-drift/`. Findings to `docs/plans/evidence/brand-drift.md`. Drift check only — no pass/fail verdict.
 
 Post-5.1: Index brand drift screenshots into graph (Slice 5) via `graph-index.js`.
 
@@ -390,9 +390,9 @@ Post-5.2: Index Track B evidence into graph.
 
 1. **Maestro E2E (3 iterations)** — INTERNAL inline — Generate multi-feature journey Maestro flows (login→browse→buy, signup→onboarding→first-action). Run 3x for flakiness detection. Multi-device: iPhone SE, iPhone 16 Pro, iPad. Quarantine flaky tests. Pass criteria: 95%+ pass rate. Evidence: `docs/plans/evidence/e2e/iter-3-results.json`.
 
-2. **iOS Dogfood** — subagent_type: `testing-evidence-collector` — Load `ios-debugger-agent` skill. Use XcodeBuildMCP to systematically explore: `describe_ui` to discover all tappable elements, navigate every screen, tap every button, fill every form. Capture console logs via `start_sim_log_cap`. Spec-blind exploratory testing. Evidence: `docs/plans/evidence/dogfood/findings.md` + `docs/plans/evidence/dogfood/findings.json`.
+2. **iOS Dogfood** — agent_type: `testing-evidence-collector` — subagent_type: `testing-evidence-collector` — Load `ios-debugger-agent` skill. Use XcodeBuildMCP to systematically explore: `describe_ui` to discover all tappable elements, navigate every screen, tap every button, fill every form. Capture console logs via `start_sim_log_cap`. Spec-blind exploratory testing. Evidence: `docs/plans/evidence/dogfood/findings.md` + `docs/plans/evidence/dogfood/findings.json`.
 
-3. **iOS Fake-Data Detector** — subagent_type: `silent-failure-hunter` — mode: "bypassPermissions" — Run `protocols/ios-fake-data-detector.md`. Static: grep for UUID() in business paths, hardcoded arrays as mock responses, Task.sleep faking async, #Preview data leaking into production, placeholder strings, hardcoded URLs. Evidence: `docs/plans/evidence/fake-data-audit.md`.
+3. **iOS Fake-Data Detector** — agent_type: `silent-failure-hunter` — subagent_type: `silent-failure-hunter` — mode: "bypassPermissions" — Run `protocols/ios-fake-data-detector.md`. Static: grep for UUID() in business paths, hardcoded arrays as mock responses, Task.sleep faking async, #Preview data leaking into production, placeholder strings, hardcoded URLs. Evidence: `docs/plans/evidence/fake-data-audit.md`.
 
 ### Post-audit evidence verification
 
@@ -412,11 +412,11 @@ Ship pipeline is **optional** (simulator-only is a valid end-state — no Apple
 
 If the user opts to ship: run the iOS `asc-*` pipeline. The per-agent wiring for Phase 7 lives in `commands/build.md` §Phase 7 — the iOS branch here only names the role slots:
 
-- App Store Connect listing + keywords + description → `subagent_type: marketing-app-store-optimizer` (dispatch lives in `commands/build.md` Phase 7). The `asc-metadata-generator`, `asc-screenshot-generator`, and `asc-privacy-manifest` items below are skill bundles the optimizer pulls in, not standalone agents.
+- App Store Connect listing + keywords + description → `agent_type: marketing-app-store-optimizer` — `subagent_type: marketing-app-store-optimizer` (dispatch lives in `commands/build.md` Phase 7). The `asc-metadata-generator`, `asc-screenshot-generator`, and `asc-privacy-manifest` items below are skill bundles the optimizer pulls in, not standalone agents.
 - `asc-metadata-generator` (skill — App Store Connect listing + keywords + description, loaded by marketing-app-store-optimizer)
 - `asc-screenshot-generator` (skill — generate App Store screenshots via XcodeBuildMCP at all required device sizes)
 - `asc-privacy-manifest` (skill — PrivacyInfo.xcprivacy)
-- iOS app review sanity check → `subagent_type: ios-app-review-guardian` before TestFlight upload — catches rejection risks (IAP rules, HIG violations, entitlement issues, metadata problems).
-- Code signing + TestFlight upload → `subagent_type: engineering-devops-automator` with `fastlane` as the underlying tool.
+- iOS app review sanity check → `agent_type: ios-app-review-guardian` — `subagent_type: ios-app-review-guardian` before TestFlight upload — catches rejection risks (IAP rules, HIG violations, entitlement issues, metadata problems).
+- Code signing + TestFlight upload → `agent_type: engineering-devops-automator` — `subagent_type: engineering-devops-automator` with `fastlane` as the underlying tool.
 
 This is SEPARATE from the web ship pipeline — do NOT run web README/deployment steps. Documentation = README with simulator run instructions + TestFlight invite link (if shipped). Skip Step 7.1 web docs and web deployment notes.

package/protocols/state-schema.json

@@ -1,6 +1,6 @@
 {
   "$schema": "http://json-schema.org/draft-07/schema#",
-  "$comment": "Schema version migration table: schema_version 1 = Stages 1-3; schema_version 2 = Stage 4 (adds backward_routing_count, backward_routing_count_by_target_phase, in_flight_backward_edge, mode_transitions); schema_version 3 = Stage 5 (adds lrr_cycle_state); schema_version 4 = Stage 6 (adds current_sprint_context_hash), schema_version 5 = Stage 7 (adds feature_delegation_plan_path, current_wave, completed_features, feature_acceptance, feature_briefs). --- Runtime validation rules (not encodable in JSON Schema, require code): Rule 5 — step prefix must match current phase number; Rule 6 — mode/autonomous consistency (mode==='autonomous' iff autonomous===true); Rule 7 — iOS fields gating (app_name, bundle_id, xcodeproj_path, ios_features, phase_progress.phase_minus_1 exist iff project_type==='ios'); Rule 10 — pending/in-progress disjoint (in_progress_task.task_id not in pending_tasks or completed_tasks); Rule 11 — resume_point.phase/step must not be ahead of top-level phase/step; Rule 12 — timestamps monotonic (session_last_saved >= session_started).",
+  "$comment": "Schema version migration table: schema_version 1 = Stages 1-3; schema_version 2 = Stage 4 (adds backward_routing_count, backward_routing_count_by_target_phase, in_flight_backward_edge, mode_transitions); schema_version 3 = Stage 5 (adds lrr_cycle_state); schema_version 4 = Stage 6 (adds current_sprint_context_hash), schema_version 5 = Stage 7 (adds feature_delegation_plan_path, current_wave, completed_features, feature_acceptance, feature_briefs), schema_version 6 = Stage 8 (adds phase_summaries). --- Runtime validation rules (not encodable in JSON Schema, require code): Rule 5 — step prefix must match current phase number; Rule 6 — mode/autonomous consistency (mode==='autonomous' iff autonomous===true); Rule 7 — iOS fields gating (app_name, bundle_id, xcodeproj_path, ios_features, phase_progress.phase_minus_1 exist iff project_type==='ios'); Rule 10 — pending/in-progress disjoint (in_progress_task.task_id not in pending_tasks or completed_tasks); Rule 11 — resume_point.phase/step must not be ahead of top-level phase/step; Rule 12 — timestamps monotonic (session_last_saved >= session_started).",
   "title": ".build-state.json",
   "description": "Typed source of truth for BuildAnything build state. Validated by the PreToolUse schema lint hook (W2-2). additionalProperties: false enforces fail-closed per A8 SSOT rule.",
   "type": "object",
@@ -199,6 +199,20 @@
         "session_id": { "type": ["string", "null"] },
         "timestamp":  { "type": "string", "format": "date-time" }
       }
+    },
+
+    "phase_summary": {
+      "type": "object",
+      "additionalProperties": false,
+      "required": ["phase", "completed_at", "artifacts", "decisions", "status"],
+      "properties": {
+        "phase": { "type": "integer", "enum": [-1, 0, 1, 2, 3, 4, 5, 6, 7] },
+        "completed_at": { "type": "string", "format": "date-time" },
+        "artifacts": { "type": "array", "items": { "type": "string" } },
+        "decisions": { "type": "string", "maxLength": 300 },
+        "status": { "type": "string", "enum": ["approved", "approved_with_concerns", "auto_approved"] },
+        "carry_forward": { "type": "string", "maxLength": 200 }
+      }
     }
   },
 
@@ -229,7 +243,7 @@
     "schema_version": {
       "type": "integer",
       "minimum": 1,
-      "maximum": 5,
+      "maximum": 6,
       "description": "Currently 5 (Stage 7). Bumped to 2 at Stage 4, 3 at Stage 5, 4 at Stage 6, 5 at Stage 7."
     },
     "project_type": {
@@ -418,6 +432,13 @@
       "additionalProperties": { "type": "string" },
       "$comment": "Stage 7+ (schema_version >= 5). Written by orchestrator after each briefing-officer dispatch (Step 4.2.a).",
       "description": "Stage 7+ (schema_version >= 5). Map of feature name to feature brief file path (docs/plans/feature-briefs/{feature}.md)."
+    },
+
+    "phase_summaries": {
+      "type": "array",
+      "items": { "$ref": "#/$defs/phase_summary" },
+      "$comment": "Stage 8+ (schema_version >= 6). Written by orchestrator at each phase boundary per the Context Budget Protocol.",
+      "description": "Stage 8+ (schema_version >= 6). Structured carry-forward summaries from completed phases. Max ~500 tokens per entry."
     }
   }
 }

package/protocols/state-schema.md

@@ -17,6 +17,7 @@
 | 3 | Stage 5 | `lrr_cycle_state` (object; interior fields loose-typed pending Stage 5 iteration — see "Fields added at v3" below) | `BUILDANYTHING_SDK_LRR=false` reverts to markdown aggregator; `lrr_cycle_state` becomes an ignored field on the orchestrator read path (additive-only, no data loss on downgrade) |
 | 4 | Stage 6 | `current_sprint_context_hash` | `BUILDANYTHING_SDK_SPRINT_CONTEXT=false` (web) and/or `BUILDANYTHING_SDK_SPRINT_CONTEXT_IOS=false` (iOS parity gate) reverts Phase 4 to per-task refs re-send; `current_sprint_context_hash` becomes an ignored field on the orchestrator read path (additive-only, no data loss on downgrade) |
 | 5 | Stage 7 | `feature_delegation_plan_path`, `current_wave`, `completed_features`, `feature_acceptance`, `feature_briefs` | Feature-level fields are additive and optional; a Stage 6 runtime reading a Stage 7 state file with `schema_version` downgraded to `4` will ignore these fields without data loss on the read path |
+| 6 | Stage 8 | `phase_summaries` | Additive; ignored by older runtimes (no data loss on downgrade) |
 
 **A7 forward-reject rule.** When `bin/buildanything-runtime.ts` reads `.build-state.json` at session start, if `schema_version > MAX_SUPPORTED_SCHEMA_VERSION`, the runtime refuses to proceed and emits a clear error pointing to the compat matrix (`docs/migration/sdk-host-compat.md`). This is the A7 defense against silent schema drift — an old runtime must never silently ignore fields a newer runtime persisted. See **Task 4.5.2** for the runtime implementation (out of scope for this prose-only update).
 
@@ -82,6 +83,7 @@
 | `verification` | object | yes | `{last_verify_result, last_verify_timestamp}`. `last_verify_result` is one of `"PRODUCTION_READY"`, `"NEEDS_WORK"`, `"BLOCKED"`, or `null`. |
 | `blockers` | array | no | Open blockers. Each: `{id, description, surfaced_at, type}`. Type is `"build"`, `"design"`, `"dep"`, or `"external"`. |
 | `decisions_pruned_at_phase0` | boolean | no | Default `false`. Set to `true` after Phase 0 archives stale decision rows. |
+| `phase_summaries` | array | no | Structured carry-forward summaries from completed phases. Each entry: `{phase, completed_at, artifacts[], decisions, status, carry_forward?}`. Max ~500 tokens per entry. Written at phase boundaries per the Context Budget Protocol. |
 
 ### Decision Log Pruning (Phase 0)
 

package/protocols/web-phase-branches.md

@@ -30,7 +30,7 @@ Before the Quality Gate 2 approval prompt in `commands/build.md` is rendered, di
 
 Call the Agent tool once:
 
-1. Description: "Visual DNA directional preview" — subagent_type: `design-brand-guardian` — prompt: "[CONTEXT header above — phase: 2. NOTE: `dna` is omitted — this step produces the preview, not the lock.] Read `docs/plans/design-doc.md` (#persona, #scope, #voice), `docs/plans/phase1-scratch/findings-digest.md` (reference signals), and `docs/plans/architecture.md` (stack constraints). Emit a 3-5 bullet DIRECTIONAL preview of the intended Visual DNA — brand read in one line, then proposed leanings on Scope, Character, Material/Motion, and Type. NO rationale paragraphs, NO reference citations, NO incompatibility-matrix work. This is a sanity-check for the user at Gate 2, not the locked card. Save to `docs/plans/visual-dna-preview.md` as a flat bullet list. Target 150 tokens of output, max 250."
+1. Description: "Visual DNA directional preview" — agent_type: `design-brand-guardian` — subagent_type: `design-brand-guardian` — prompt: "[CONTEXT header above — phase: 2. NOTE: `dna` is omitted — this step produces the preview, not the lock.] Read `docs/plans/design-doc.md` (#persona, #scope, #voice), `docs/plans/phase1-scratch/findings-digest.md` (reference signals), and `docs/plans/architecture.md` (stack constraints). Emit a 3-5 bullet DIRECTIONAL preview of the intended Visual DNA — brand read in one line, then proposed leanings on Scope, Character, Material/Motion, and Type. NO rationale paragraphs, NO reference citations, NO incompatibility-matrix work. This is a sanity-check for the user at Gate 2, not the locked card. Save to `docs/plans/visual-dna-preview.md` as a flat bullet list. Target 150 tokens of output, max 250."
 
 Output: `docs/plans/visual-dna-preview.md` — surfaced by the orchestrator in the Gate 2 prompt alongside Architecture + Sprint Task List. Phase 3.0 Brand Guardian re-invokes to produce the full locked 6-axis card; the preview is discarded after Gate 2 approval.
 
@@ -50,7 +50,7 @@ Dispatch a single agent to author Pass 1 of `DESIGN.md` (repo root). Pass 1 lock
 
 Call the Agent tool once:
 
-1. Description: "DESIGN.md Pass 1 — Brand DNA + Overview" — subagent_type: `design-brand-guardian` — prompt: "[CONTEXT header above — phase: 3. NOTE: Step 3.0 omits `dna` because this step PRODUCES it.] You are the Brand Guardian authoring Pass 1 of `DESIGN.md`. The format is specified by `protocols/design-md-spec.md` (vendored). The pipeline contract is in `protocols/design-md-authoring.md`. Read both before writing.
+1. Description: "DESIGN.md Pass 1 — Brand DNA + Overview" — agent_type: `design-brand-guardian` — subagent_type: `design-brand-guardian` — prompt: "[CONTEXT header above — phase: 3. NOTE: Step 3.0 omits `dna` because this step PRODUCES it.] You are the Brand Guardian authoring Pass 1 of `DESIGN.md`. The format is specified by `protocols/design-md-spec.md` (vendored). The pipeline contract is in `protocols/design-md-authoring.md`. Read both before writing.
 
 Inputs (Read tool): `docs/plans/product-spec.md` (## App Overview for product identity, ## Screen Inventory for what screens exist, ## Permissions & Roles for complexity level — a dense admin panel needs different DNA than a simple consumer app), `docs/plans/design-doc.md` (product concept, user, voice), `docs/plans/phase1-scratch/findings-digest.md` (reference sites the user mentioned, competitor aesthetic landscape), `docs/plans/architecture.md` (stack constraints — e.g. server-rendered Rails can't ship Three.js), `docs/plans/quality-targets.json` (perf budget constrains motion and material choices), `docs/plans/phase1-scratch/user-decisions.md`.
 
@@ -88,9 +88,9 @@ Research is now goal-directed — validate and enrich the locked DNA, not catalo
 
 Call the Agent tool 2 times in one message:
 
-1. Description: "Competitive visual audit" — subagent_type: `visual-research` — prompt: "[CONTEXT header above — phase: 3] Mode: `competitive-audit`. Read `DESIGN.md` (`## Overview > ### Brand DNA`) to understand the locked DNA. Find 5-8 rival UIs that exemplify the chosen DNA axes (NOT all competitors — only ones that nail the axes we chose). Use Playwright to screenshot each at desktop 1920x1080 and mobile 375x812. For each site, analyze which DNA axes it nails and which it doesn't. Save screenshots to `docs/plans/design-references/competitors/`. Append findings to `docs/plans/design-references.md` grouped by DNA axis (motion refs, material refs, typography refs, character refs, density refs). Optional caller-supplied competitor URLs: [list or 'none']."
+1. Description: "Competitive visual audit" — agent_type: `visual-research` — subagent_type: `visual-research` — prompt: "[CONTEXT header above — phase: 3] Mode: `competitive-audit`. Read `DESIGN.md` (`## Overview > ### Brand DNA`) to understand the locked DNA. Find 5-8 rival UIs that exemplify the chosen DNA axes (NOT all competitors — only ones that nail the axes we chose). Use Playwright to screenshot each at desktop 1920x1080 and mobile 375x812. For each site, analyze which DNA axes it nails and which it doesn't. Save screenshots to `docs/plans/design-references/competitors/`. Append findings to `docs/plans/design-references.md` grouped by DNA axis (motion refs, material refs, typography refs, character refs, density refs). Optional caller-supplied competitor URLs: [list or 'none']."
 
-2. Description: "Design inspiration mining" — subagent_type: `visual-research` — prompt: "[CONTEXT header above — phase: 3] Mode: `inspiration-mining`. Read `DESIGN.md` (`## Overview > ### Brand DNA`). Search Awwwards.com, Godly.website, and SiteInspire for award-winning sites that match the DNA axes. Use Playwright to screenshot the top 5-8 results at desktop 1920x1080 and mobile 375x812. Save to `docs/plans/design-references/inspiration/`. Append findings to `docs/plans/design-references.md` grouped by DNA axis. Tag every reference with the specific axis (or axes) it validates."
+2. Description: "Design inspiration mining" — agent_type: `visual-research` — subagent_type: `visual-research` — prompt: "[CONTEXT header above — phase: 3] Mode: `inspiration-mining`. Read `DESIGN.md` (`## Overview > ### Brand DNA`). Search Awwwards.com, Godly.website, and SiteInspire for award-winning sites that match the DNA axes. Use Playwright to screenshot the top 5-8 results at desktop 1920x1080 and mobile 375x812. Save to `docs/plans/design-references/inspiration/`. Append findings to `docs/plans/design-references.md` grouped by DNA axis. Tag every reference with the specific axis (or axes) it validates."
 
 Output: `docs/plans/design-references.md` — reference paths grouped by DNA axis, ready to feed Step 3.2 component mapping and Step 3.6 critic scoring.
 
@@ -110,7 +110,7 @@ This is the compositional step. The Visual Designer picks specific library compo
 
 Call the Agent tool once:
 
-1. Description: "Component library mapping" — subagent_type: `design-ui-designer` — prompt: "[CONTEXT header above — phase: 3] Read `DESIGN.md` (`## Overview > ### Brand DNA` for axis values; `### References` for reference paths), `docs/plans/design-references.md`, `docs/plans/product-spec.md` (## Screen Inventory for what screens exist, per-feature States and Empty/Loading/Error States sections for what component states are needed — e.g. a feature with 7 states needs more component variants than one with 3), and `docs/library-refs/component-library-catalog.md` (the static reference mapping DNA-axis combinations to library component variants). Pick specific component variants for each slot the product needs: hero, cards, cta, nav, marquee, chart, 3D, form elements, modals. The catalog is authoritative — when the DNA matches a row, use the variants that row specifies; do not reinvent. Write `docs/plans/component-manifest.md` with the locked component picks, one row per slot, naming the library and the variant. For any slot the catalog doesn't cover, emit a row tagged 'manifest gap' with a short fallback plan (stock shadcn primitive plus notes)."
+1. Description: "Component library mapping" — agent_type: `design-ui-designer` — subagent_type: `design-ui-designer` — prompt: "[CONTEXT header above — phase: 3] Read `DESIGN.md` (`## Overview > ### Brand DNA` for axis values; `### References` for reference paths), `docs/plans/design-references.md`, `docs/plans/product-spec.md` (## Screen Inventory for what screens exist, per-feature States and Empty/Loading/Error States sections for what component states are needed — e.g. a feature with 7 states needs more component variants than one with 3), and `docs/library-refs/component-library-catalog.md` (the static reference mapping DNA-axis combinations to library component variants). Pick specific component variants for each slot the product needs: hero, cards, cta, nav, marquee, chart, 3D, form elements, modals. The catalog is authoritative — when the DNA matches a row, use the variants that row specifies; do not reinvent. Write `docs/plans/component-manifest.md` with the locked component picks, one row per slot, naming the library and the variant. For any slot the catalog doesn't cover, emit a row tagged 'manifest gap' with a short fallback plan (stock shadcn primitive plus notes)."
 
 Output: `docs/plans/component-manifest.md` — locked component manifest.
 
@@ -128,7 +128,7 @@ Run via the Bash tool:
 
 ### Step 3.2b — DNA Persona Check
 
-Call the Agent tool — description: "DNA persona check" — subagent_type: design-ux-researcher — prompt: "[CONTEXT header above — phase: 3] Read `DESIGN.md` (the full Pass 1 — `## Overview` including `### Brand DNA` is the locked 7-axis card and `### Rationale` explains why those axes were chosen) + docs/plans/design-doc.md (#persona and #jobs-to-be-done sections) + docs/plans/product-spec.md (## App Overview and per-feature Persona Constraints sections — these carry the specific behavioral patterns from research, e.g. 'user scans, doesn't read') + docs/plans/phase1-scratch/findings-digest.md. Validate: do the locked DNA axes actually serve this persona and these jobs-to-be-done? Cross-check each DNA axis against the persona's context (e.g., if persona is 'senior enterprise buyer on a tight schedule' but DNA chose Maximalist + Cinematic, that's wrong — Enterprise/Minimal/Subtle fits better). Report any DNA-persona mismatches. If mismatches found, the Brand Guardian may need to re-author DESIGN.md Pass 1 (backward edge to Step 3.0). Save findings to docs/plans/dna-persona-check.md."
+Call the Agent tool — description: "DNA persona check" — agent_type: design-ux-researcher — subagent_type: design-ux-researcher — prompt: "[CONTEXT header above — phase: 3] Read `DESIGN.md` (the full Pass 1 — `## Overview` including `### Brand DNA` is the locked 7-axis card and `### Rationale` explains why those axes were chosen) + docs/plans/design-doc.md (#persona and #jobs-to-be-done sections) + docs/plans/product-spec.md (## App Overview and per-feature Persona Constraints sections — these carry the specific behavioral patterns from research, e.g. 'user scans, doesn't read') + docs/plans/phase1-scratch/findings-digest.md. Validate: do the locked DNA axes actually serve this persona and these jobs-to-be-done? Cross-check each DNA axis against the persona's context (e.g., if persona is 'senior enterprise buyer on a tight schedule' but DNA chose Maximalist + Cinematic, that's wrong — Enterprise/Minimal/Subtle fits better). Report any DNA-persona mismatches. If mismatches found, the Brand Guardian may need to re-author DESIGN.md Pass 1 (backward edge to Step 3.0). Save findings to docs/plans/dna-persona-check.md."
 
 ### Step 3.3 — UX Architecture + Page Layouts (single agent)
 
@@ -136,7 +136,7 @@ Structural design must align to the locked DNA — a Dense layout behaves differ
 
 Call the Agent tool once:
 
-1. Description: "UX architecture + page layouts" — subagent_type: `design-ux-architect` — mode: "bypassPermissions" — prompt: "[CONTEXT header above — phase: 3] Read the page spec schema at `protocols/page-spec-schema.md` before writing. Then read these inputs via your Read tool:
+1. Description: "UX architecture + page layouts" — agent_type: `design-ux-architect` — subagent_type: `design-ux-architect` — mode: "bypassPermissions" — prompt: "[CONTEXT header above — phase: 3] Read the page spec schema at `protocols/page-spec-schema.md` before writing. Then read these inputs via your Read tool:
   - Product spec: `docs/plans/product-spec.md` (FULL document — this is your source of truth. Screen Inventory is your screen list. Per-feature sections define what each screen does, what data it shows, what states exist, what errors look like, persona constraints, business rules)
   - Visual DNA: `DESIGN.md` `## Overview > ### Brand DNA` (Density axis drives layout — Airy = generous whitespace, Dense = compact data. Character and Motion axes shape navigation transitions and interaction patterns)
   - Components: `docs/plans/component-manifest.md` (which library components for which slots — use these in your wireframes)
@@ -173,7 +173,7 @@ Validate the UX architecture against the target persona's actual goals and jobs-
 
 Call the Agent tool once:
 
-1. Description: "UX flow validation" — subagent_type: `design-ux-researcher` — prompt: "[CONTEXT header above — phase: 3] Read `docs/plans/ux-architecture.md`, `docs/plans/page-specs/` (the ASCII wireframes — validate that layouts serve the persona), `docs/plans/product-spec.md` (per-feature Happy Path and Persona Constraints — these are the behavioral source of truth the flows must implement), `docs/plans/design-doc.md` (#persona, #jobs-to-be-done, #scope sections), and `DESIGN.md`. For each user flow in the UX architecture, walk through it as the target persona: narrate the steps, flag friction points, check if the flow serves the persona's jobs-to-be-done efficiently. Specifically check: (1) Are there screens or sections the persona doesn't need? (2) Are critical tasks reachable in the minimum number of steps? (3) Does the information hierarchy match what the persona cares about most? (4) Does the navigation pattern fit the persona's context (mobile-first for on-the-go users, sidebar for desktop power users, etc.)? (5) Does the responsive strategy degrade gracefully for the persona's primary device? Report findings to `docs/plans/ux-flow-validation.md` with pass/flag per flow. If critical flow issues are found, the UX Architect should revise `ux-architecture.md` before proceeding (backward edge to Step 3.3)."
+1. Description: "UX flow validation" — agent_type: `design-ux-researcher` — subagent_type: `design-ux-researcher` — prompt: "[CONTEXT header above — phase: 3] Read `docs/plans/ux-architecture.md`, `docs/plans/page-specs/` (the ASCII wireframes — validate that layouts serve the persona), `docs/plans/product-spec.md` (per-feature Happy Path and Persona Constraints — these are the behavioral source of truth the flows must implement), `docs/plans/design-doc.md` (#persona, #jobs-to-be-done, #scope sections), and `DESIGN.md`. For each user flow in the UX architecture, walk through it as the target persona: narrate the steps, flag friction points, check if the flow serves the persona's jobs-to-be-done efficiently. Specifically check: (1) Are there screens or sections the persona doesn't need? (2) Are critical tasks reachable in the minimum number of steps? (3) Does the information hierarchy match what the persona cares about most? (4) Does the navigation pattern fit the persona's context (mobile-first for on-the-go users, sidebar for desktop power users, etc.)? (5) Does the responsive strategy degrade gracefully for the persona's primary device? Report findings to `docs/plans/ux-flow-validation.md` with pass/flag per flow. If critical flow issues are found, the UX Architect should revise `ux-architecture.md` before proceeding (backward edge to Step 3.3)."
 
 Output: `docs/plans/ux-flow-validation.md`.
 
@@ -183,7 +183,7 @@ The Visual Designer re-invokes as writer this time, producing the much richer Vi
 
 Call the Agent tool once:
 
-1. Description: "Visual design spec" — subagent_type: `design-ui-designer` — prompt: "[CONTEXT header above — phase: 3] Second invocation as writer. Read `DESIGN.md`, `docs/plans/component-manifest.md`, `docs/plans/ux-architecture.md`, `docs/plans/design-references.md`, `docs/plans/product-spec.md` (per-feature States and Empty/Loading/Error States — the state matrix must cover every state the product spec defines, not just generic defaults), and `docs/plans/page-specs/` (the ASCII wireframes — the typography ramp and spacing scale must work for the actual page layouts, not just in isolation). Write `DESIGN.md` with ALL the following layers:
+1. Description: "Visual design spec" — agent_type: `design-ui-designer` — subagent_type: `design-ui-designer` — prompt: "[CONTEXT header above — phase: 3] Second invocation as writer. Read `DESIGN.md`, `docs/plans/component-manifest.md`, `docs/plans/ux-architecture.md`, `docs/plans/design-references.md`, `docs/plans/product-spec.md` (per-feature States and Empty/Loading/Error States — the state matrix must cover every state the product spec defines, not just generic defaults), and `docs/plans/page-specs/` (the ASCII wireframes — the typography ramp and spacing scale must work for the actual page layouts, not just in isolation). Write `DESIGN.md` with ALL the following layers:
 
 **TOKENS** (existing): color system (hex, light + dark), typography scale, spacing (8px base), shadows, radius.
 
@@ -213,7 +213,7 @@ Run via the Bash tool:
 
 Call the Agent tool once:
 
-1. Description: "Inclusive visuals check" — subagent_type: `design-inclusive-visuals-specialist` — prompt: "[CONTEXT header above — phase: 3] Read `DESIGN.md`, `docs/plans/component-manifest.md`, and `DESIGN.md`. Audit for representation gaps, imagery bias, color choices that exclude colorblind users, contrast failures, and culturally-specific iconography that doesn't translate. Write findings to `docs/plans/inclusive-visuals-audit.md`."
+1. Description: "Inclusive visuals check" — agent_type: `design-inclusive-visuals-specialist` — subagent_type: `design-inclusive-visuals-specialist` — prompt: "[CONTEXT header above — phase: 3] Read `DESIGN.md`, `docs/plans/component-manifest.md`, and `DESIGN.md`. Audit for representation gaps, imagery bias, color choices that exclude colorblind users, contrast failures, and culturally-specific iconography that doesn't translate. Write findings to `docs/plans/inclusive-visuals-audit.md`."
 
 Output: `docs/plans/inclusive-visuals-audit.md`.
 
@@ -225,28 +225,18 @@ This is the only Phase 3 step that writes code. Wrapped in a generator/critic me
 
 Call the Agent tool once:
 
-1. Description: "Build living style guide" — subagent_type: `engineering-frontend-developer` — mode: "bypassPermissions" — prompt: "[CONTEXT header above — phase: 3] [COMPLEXITY: L] Read `docs/plans/component-manifest.md` and `DESIGN.md`. Build a `/design-system` route with rendered, interactive examples of every chosen variant from the manifest. **HARD-GATE: Import from the installed libraries. Do NOT write components from scratch when the manifest names one.** Every component must be interactive (hover, focus, transitions all work). Mobile-responsive. This ships with the product. Commit: 'feat: living style guide'."
+1. Description: "Build living style guide" — agent_type: `engineering-frontend-developer` — subagent_type: `engineering-frontend-developer` — mode: "bypassPermissions" — prompt: "[CONTEXT header above — phase: 3] Read `docs/plans/component-manifest.md` and `DESIGN.md`. Build a `/design-system` route with rendered, interactive examples of every chosen variant from the manifest. **HARD-GATE: Import from the installed libraries. Do NOT write components from scratch when the manifest names one.** Every component must be interactive (hover, focus, transitions all work). Mobile-responsive. This ships with the product. Commit: 'feat: living style guide'."
 
 **Metric loop wrapper** (per `protocols/metric-loop.md`):
 
-- **Critic** — Call the Agent tool — description: "Design critic scoring pass" — subagent_type: `design-critic` — prompt: "[CONTEXT header above — phase: 3] SCORING CRITERIA CHECKLIST: [paste the checklist from `active_metric_loop.scoring_criteria_checklist` in `.build-state.json` — NOT the raw reference docs]. Capture the rendered `/design-system` route via Playwright screenshot (desktop 1920x1080 + mobile 375x812). Also read `docs/plans/page-specs/` to understand what page compositions these components will be used in — score components in the context of their actual usage, not just in isolation. Score the gap on **7 DNA axes** (Scope fit, Density, Character, Material, Motion, Type, Copy — 20 points each) plus **5 craft dimensions** (whitespace rhythm, visual hierarchy, motion coherence, color harmony, typographic refinement — 20 points each). Total 240. Target 195. <!-- Scoring scale: see agents/design-critic.md for authoritative thresholds --> Every finding must cite a specific element with file:line reference AND reference the checklist criteria — score a gap, not an opinion. Suggest concrete improvements ('the card padding is 16px but the checklist says Density: Airy — 32px — bump to 32px'). Iteration 1 MAY Read `docs/plans/design-references.md` for visual comparison; iteration 2+ MUST NOT unless diagnosis explicitly flags a visual-reference gap. Default verdict: NEEDS WORK. Never edit code. Max 5 iterations before exit."
+- **Critic** — Call the Agent tool — description: "Design critic scoring pass" — agent_type: `design-critic` — subagent_type: `design-critic` — prompt: "[CONTEXT header above — phase: 3] SCORING CRITERIA CHECKLIST: [paste the checklist from `active_metric_loop.scoring_criteria_checklist` in `.build-state.json` — NOT the raw reference docs]. Capture the rendered `/design-system` route via Playwright screenshot (desktop 1920x1080 + mobile 375x812). Also read `docs/plans/page-specs/` to understand what page compositions these components will be used in — score components in the context of their actual usage, not just in isolation. Score the gap on **7 DNA axes** (Scope fit, Density, Character, Material, Motion, Type, Copy — 20 points each) plus **5 craft dimensions** (whitespace rhythm, visual hierarchy, motion coherence, color harmony, typographic refinement — 20 points each). Total 240. Target 195. <!-- Scoring scale: see agents/design-critic.md for authoritative thresholds --> Every finding must cite a specific element with file:line reference AND reference the checklist criteria — score a gap, not an opinion. Suggest concrete improvements ('the card padding is 16px but the checklist says Density: Airy — 32px — bump to 32px'). Iteration 1 MAY Read `docs/plans/design-references.md` for visual comparison; iteration 2+ MUST NOT unless diagnosis explicitly flags a visual-reference gap. Default verdict: NEEDS WORK. Never edit code. Max 5 iterations before exit."
 
-- **Generator (re-invocation, iteration 2+)** — Call the Agent tool — description: "Apply critic's top issue" — subagent_type: `engineering-frontend-developer` — mode: "bypassPermissions" — prompt: "TARGETED FIX from metric loop diagnosis: [paste top issue from Step 3 diagnosis]. Files: [paste file paths]. Relevant criteria from checklist: [paste the specific checklist values that relate to the top issue — e.g., 'Density: Airy — 32px card padding']. Apply ONLY the top issue. Do not re-critique. Do not refactor other parts. Re-render the `/design-system` route. Return the commit SHA." NOTE: Do NOT include `[CONTEXT header above]` on iteration 2+ — the generator already has the codebase context from iteration 1. Per `protocols/metric-loop.md` Step 4 iteration-aware context rule.
+- **Generator (re-invocation, iteration 2+)** — Call the Agent tool — description: "Apply critic's top issue" — agent_type: `engineering-frontend-developer` — subagent_type: `engineering-frontend-developer` — mode: "bypassPermissions" — prompt: "TARGETED FIX from metric loop diagnosis: [paste top issue from Step 3 diagnosis]. Files: [paste file paths]. Relevant criteria from checklist: [paste the specific checklist values that relate to the top issue — e.g., 'Density: Airy — 32px card padding']. Apply ONLY the top issue. Do not re-critique. Do not refactor other parts. Re-render the `/design-system` route. Return the commit SHA." NOTE: Do NOT include `[CONTEXT header above]` on iteration 2+ — the generator already has the codebase context from iteration 1. Per `protocols/metric-loop.md` Step 4 iteration-aware context rule.
 
 - **Exit conditions:** quality target hit (score ≥ 195), stall (no score improvement for 2 consecutive rounds), or max iterations (5 total).
 
 Record the score history to `docs/plans/build-log.md` under `## Design Critic Loop`.
 
-### Step 3.7 — A11y Design Review (single agent)
-
-WCAG 2.2 AA runtime check on the rendered style guide plus any key product pages that exist at this point.
-
-Call the Agent tool once:
-
-1. Description: "A11y design review" — subagent_type: `a11y-architect` — prompt: "[CONTEXT header above — phase: 3] WCAG 2.2 AA runtime check on the rendered `/design-system` route and any key product pages. Check contrast, focus order, keyboard navigation, screen reader labels, reduced-motion variants, and touch targets (>= 44px). Use Playwright and axe-core. Save findings to `docs/plans/a11y-design-review.md` with severity tags (Critical / Serious / Moderate / Minor)."
-
-Output: `docs/plans/a11y-design-review.md`.
-
 ### Step 3.8 — Autonomous Quality Gate
 
 Log to `docs/plans/build-log.md`: final screenshot paths, Design Critic score history (per-round totals plus per-axis subscores), a11y findings count by severity, a DNA compliance score derived from the critic's 7 DNA-axis subscores, and the DESIGN.md lint result (broken-refs count, warning count, hash). No user pause.
@@ -273,15 +263,15 @@ Step 4.0 is three sequential dispatches: project scaffolding, design system setu
 
 #### 4.0.a — Project scaffolding
 
-Call the Agent tool — description: "Project scaffolding" — subagent_type: `engineering-rapid-prototyper` — mode: "bypassPermissions" — prompt: "[CONTEXT header above — phase: 4] [COMPLEXITY: M] Set up the project from the architecture. Read `docs/plans/architecture.md` via your Read tool before starting. Create directory structure, dependencies, build tooling, linting config, test framework with one passing test, .gitignore, .env.example. Read `DESIGN.md` Scope axis and only install the component libraries the DNA needs — never ship Three.js for an internal admin panel. Commit: 'feat: initial scaffolding'."
+Call the Agent tool — description: "Project scaffolding" — agent_type: `engineering-rapid-prototyper` — subagent_type: `engineering-rapid-prototyper` — mode: "bypassPermissions" — prompt: "[CONTEXT header above — phase: 4] Set up the project from the architecture. Read `docs/plans/architecture.md` via your Read tool before starting. Create directory structure, dependencies, build tooling, linting config, test framework with one passing test, .gitignore, .env.example. Read `DESIGN.md` Scope axis and only install the component libraries the DNA needs — never ship Three.js for an internal admin panel. Commit: 'feat: initial scaffolding'."
 
 #### 4.0.b — Design system setup
 
-Call the Agent tool — description: "Design system setup" — subagent_type: `engineering-frontend-developer` — mode: "bypassPermissions" — prompt: "[CONTEXT header above — phase: 4] Implement the design system from the Visual Design Spec. Read `DESIGN.md` via your Read tool before starting. Create CSS tokens matching the spec's color system, typography scale, spacing system, shadow/elevation tokens, and base layout components. The living style guide from Phase 3 is the reference implementation — components must match. Commit: 'feat: design system'."
+Call the Agent tool — description: "Design system setup" — agent_type: `engineering-frontend-developer` — subagent_type: `engineering-frontend-developer` — mode: "bypassPermissions" — prompt: "[CONTEXT header above — phase: 4] Implement the design system from the Visual Design Spec. Read `DESIGN.md` via your Read tool before starting. Create CSS tokens matching the spec's color system, typography scale, spacing system, shadow/elevation tokens, and base layout components. The living style guide from Phase 3 is the reference implementation — components must match. Commit: 'feat: design system'."
 
 #### 4.0.c — Acceptance test scaffolding
 
-Call the Agent tool — description: "Scaffold acceptance tests" — subagent_type: `engineering-frontend-developer` — mode: "bypassPermissions" — prompt: "[CONTEXT header above — phase: 4] Read docs/plans/sprint-tasks.md. For every task with a Behavioral Test field, create a Playwright test stub in tests/e2e/acceptance/. Use Page Object Model. Each test should: navigate to the page, perform the interaction, assert the expected outcome. Tests should FAIL right now (features aren't built yet) — that's correct. Also ensure agent-browser is available (run `which agent-browser`). Commit: 'test: scaffold acceptance tests from sprint tasks'."
+Call the Agent tool — description: "Scaffold acceptance tests" — agent_type: `engineering-frontend-developer` — subagent_type: `engineering-frontend-developer` — mode: "bypassPermissions" — prompt: "[CONTEXT header above — phase: 4] Read docs/plans/sprint-tasks.md. For every task with a Behavioral Test field, create a Playwright test stub in tests/e2e/acceptance/. Use Page Object Model. Each test should: navigate to the page, perform the interaction, assert the expected outcome. Tests should FAIL right now (features aren't built yet) — that's correct. Also ensure agent-browser is available (run `which agent-browser`). Commit: 'test: scaffold acceptance tests from sprint tasks'."
 
 ## Phase 4 — Build per-task flow (web branch)
 
@@ -299,7 +289,7 @@ No magic parallelism cap — the dependency graph is the limit within a feature.
 
 The Briefing Officer's feature brief specifies the agent type (`subagent_type`) for each task — the orchestrator reads it from the brief rather than deciding itself.
 
-Call the Agent tool — description: "[task name]" — subagent_type: `[from BO brief]` — mode: "bypassPermissions" — prompt: "[CONTEXT header above — phase: 4] [COMPLEXITY: S/M/L from sprint-tasks.md].
+Call the Agent tool — description: "[task name]" — agent_type: `[from BO brief]` — subagent_type: `[from BO brief]` — mode: "bypassPermissions" — prompt: "[CONTEXT header above — phase: 4].
 
 TASK: [task description from BO brief]
 
@@ -355,9 +345,9 @@ Read the NFRs from `docs/plans/quality-targets.json` (and `docs/plans/sprint-tas
 
 Call the Agent tool 5 times in one message:
 
-1. Description: "API testing" — subagent_type: `testing-api-tester` — Prompt: "[CONTEXT header above — phase: 5] Comprehensive API validation: all endpoints, edge cases, error responses, auth flows. NFR targets: Read `docs/plans/quality-targets.json` via your Read tool for performance and reliability thresholds. Report findings with counts."
+1. Description: "API testing" — agent_type: `testing-api-tester` — subagent_type: `testing-api-tester` — Prompt: "[CONTEXT header above — phase: 5] Comprehensive API validation: all endpoints, edge cases, error responses, auth flows. NFR targets: Read `docs/plans/quality-targets.json` via your Read tool for performance and reliability thresholds. Report findings with counts."
 
-2. Description: "Performance audit" — subagent_type: `testing-performance-benchmarker` — Prompt: "[CONTEXT header above — phase: 5] Measure response times, identify bottlenecks, flag performance issues. NFR targets: Read `docs/plans/quality-targets.json` via your Read tool for performance thresholds. Report benchmarks AGAINST these targets.
+2. Description: "Performance audit" — agent_type: `testing-performance-benchmarker` — subagent_type: `testing-performance-benchmarker` — Prompt: "[CONTEXT header above — phase: 5] Measure response times, identify bottlenecks, flag performance issues. NFR targets: Read `docs/plans/quality-targets.json` via your Read tool for performance thresholds. Report benchmarks AGAINST these targets.
 
 **Bundle budget per Scope axis** (read `DESIGN.md` Scope field):
 - Marketing:     500KB gzipped (excluding images), LCP <= 2.5s
@@ -367,11 +357,11 @@ Call the Agent tool 5 times in one message:
 
 Exceeding the budget by >25% auto-blocks the Phase 6 LRR SRE chapter. Budget violations route back to Phase 3.2 (component mapping — swap a heavy variant for a lighter one) OR Phase 4 (code-splitting, lazy-loading, dynamic imports). Report budget-compliance per Scope axis, with the exact gzipped bundle size and LCP measurement."
 
-3. Description: "Accessibility audit" — subagent_type: `a11y-architect` — Prompt: "[CONTEXT header above — phase: 5] WCAG 2.2 AA runtime compliance audit on all interfaces. NFR target: Read `docs/plans/quality-targets.json` via your Read tool for accessibility thresholds. Check screen reader, keyboard nav, contrast, focus order, reduced-motion variants, touch targets >= 44px. Report issues with severity tags (Critical/Serious/Moderate/Minor). This is the same agent that sets constraints at Phase 2 and judges at Phase 6 LRR — keep the standards consistent across all three invocations."
+3. Description: "Accessibility audit" — agent_type: `a11y-architect` — subagent_type: `a11y-architect` — Prompt: "[CONTEXT header above — phase: 5] Light-touch accessibility sweep — flag only Critical and Serious WCAG 2.2 AA violations. Skip Moderate/Minor. WCAG 2.2 AA runtime compliance audit on all interfaces. NFR target: Read `docs/plans/quality-targets.json` via your Read tool for accessibility thresholds. Check screen reader, keyboard nav, contrast, focus order, reduced-motion variants, touch targets >= 44px. Report issues with severity tags (Critical/Serious/Moderate/Minor). This is the same agent that sets constraints at Phase 2 and judges at Phase 6 LRR — keep the standards consistent across all three invocations."
 
-4. Description: "Security audit" — subagent_type: `engineering-security-engineer` — Prompt: "[CONTEXT header above — phase: 5] Security review: auth, input validation, data exposure, dependency vulnerabilities. NFR targets: Read `docs/plans/quality-targets.json` via your Read tool for security thresholds. Report findings with severity."
+4. Description: "Security audit" — agent_type: `engineering-security-engineer` — subagent_type: `engineering-security-engineer` — Prompt: "[CONTEXT header above — phase: 5] Security review: auth, input validation, data exposure, dependency vulnerabilities. NFR targets: Read `docs/plans/quality-targets.json` via your Read tool for security thresholds. Report findings with severity."
 
-5. Description: "Brand Guardian drift check" — subagent_type: `design-brand-guardian` — Prompt: "[CONTEXT header above — phase: 5] You are the Phase 5 drift check (proposed state §5 re-invite). Read `DESIGN.md` (the DNA card locked at Phase 3.0) + the actually-built pages via Playwright screenshots under `docs/plans/evidence/brand-drift/` (write production screenshots there as PNG/JPG files, one per page audited, named `<screen-id>.png`). Score whether Phase 4 implementers stayed true to the DNA or drifted away from it. Specifically check each of the 6 DNA axes (Scope / Density / Character / Material / Motion / Type) against what the built product actually renders. Report drift count and specific elements (file:line references). Save findings to `docs/plans/evidence/brand-drift.md`. This is a drift check only — the Phase 6 LRR Brand Guardian chapter does the verdict. You do NOT issue a pass/fail here, only surface findings for the LRR chapter to read."
+5. Description: "Brand Guardian drift check" — agent_type: `design-brand-guardian` — subagent_type: `design-brand-guardian` — Prompt: "[CONTEXT header above — phase: 5] You are the Phase 5 drift check (proposed state §5 re-invite). Read `DESIGN.md` (the DNA card locked at Phase 3.0) + the actually-built pages via Playwright screenshots under `docs/plans/evidence/brand-drift/` (write production screenshots there as PNG/JPG files, one per page audited, named `<screen-id>.png`). Score whether Phase 4 implementers stayed true to the DNA or drifted away from it. Specifically check each of the 6 DNA axes (Scope / Density / Character / Material / Motion / Type) against what the built product actually renders. Report drift count and specific elements (file:line references). Save findings to `docs/plans/evidence/brand-drift.md`. This is a drift check only — the Phase 6 LRR Brand Guardian chapter does the verdict. You do NOT issue a pass/fail here, only surface findings for the LRR chapter to read."
 
 #### Step 5.1.idx — Brand drift screenshots graph index
 
@@ -422,9 +412,9 @@ HARD-GATE: ALL 3 ITERATIONS ARE MANDATORY. Do NOT stop after iteration 1 even if
 
 **Iteration 1 — Generate & Run:**
 
-Call the Agent tool — description: "E2E test generation" — subagent_type: `engineering-frontend-developer` — mode: "bypassPermissions" — prompt:
+Call the Agent tool — description: "E2E test generation" — agent_type: `engineering-frontend-developer` — subagent_type: `engineering-frontend-developer` — mode: "bypassPermissions" — prompt:
 
-"[CONTEXT header above — phase: 5] [COMPLEXITY: L] Generate and run end-to-end Playwright tests for cross-feature User Journeys ONLY (single-feature happy paths are covered by Track B at Step 5.2 — do NOT duplicate them here).
+"[CONTEXT header above — phase: 5] Generate and run end-to-end Playwright tests for cross-feature User Journeys ONLY (single-feature happy paths are covered by Track B at Step 5.2 — do NOT duplicate them here).
 
 INPUTS:
 Read these files via your Read tool before starting — do NOT expect pasted content:
@@ -459,13 +449,13 @@ Record results: total tests, pass count, fail count, failure details. Log to `do
 
 **Iteration 2 — Fix & Re-run:**
 
-Call the Agent tool — description: "E2E fix iteration 2" — subagent_type: `engineering-frontend-developer` — mode: "bypassPermissions" — prompt: "[CONTEXT header above — phase: 5] [COMPLEXITY: M] Fix E2E test failures from iteration 1: [paste failure details — test names, error messages, screenshot paths]. Diagnose each as real bug, flaky test, or missing selector. Fix accordingly — do NOT delete or skip tests. Re-run ALL tests. Commit: 'fix: e2e test failures iteration 2'."
+Call the Agent tool — description: "E2E fix iteration 2" — agent_type: `engineering-frontend-developer` — subagent_type: `engineering-frontend-developer` — mode: "bypassPermissions" — prompt: "[CONTEXT header above — phase: 5] Fix E2E test failures from iteration 1: [paste failure details — test names, error messages, screenshot paths]. Diagnose each as real bug, flaky test, or missing selector. Fix accordingly — do NOT delete or skip tests. Re-run ALL tests. Commit: 'fix: e2e test failures iteration 2'."
 
 Record results in the E2E table. Identify flaky candidates (passed iter 1, failed iter 2 or vice versa).
 
 **Iteration 3 — Final Stability Run:**
 
-Call the Agent tool — description: "E2E stability run" — subagent_type: `engineering-frontend-developer` — mode: "bypassPermissions" — prompt: "[CONTEXT header above — phase: 5] [COMPLEXITY: M] Final E2E stability run (3 of 3). Previous results — Iter 1: [pass/fail counts], Iter 2: [pass/fail counts], Flaky candidates: [list]. Run ALL tests with --repeat-each=3. Quarantine inconsistent tests with test.fixme(). Fix remaining consistent failures. PASS CRITERIA: 95%+ pass rate (quarantined flaky tests excluded but logged). Commit: 'test: e2e stability fixes iteration 3'."
+Call the Agent tool — description: "E2E stability run" — agent_type: `engineering-frontend-developer` — subagent_type: `engineering-frontend-developer` — mode: "bypassPermissions" — prompt: "[CONTEXT header above — phase: 5] Final E2E stability run (3 of 3). Previous results — Iter 1: [pass/fail counts], Iter 2: [pass/fail counts], Flaky candidates: [list]. Run ALL tests with --repeat-each=3. Quarantine inconsistent tests with test.fixme(). Fix remaining consistent failures. PASS CRITERIA: 95%+ pass rate (quarantined flaky tests excluded but logged). Commit: 'test: e2e stability fixes iteration 3'."
 
 Record final results. Include in the Phase 6.0 Reality Check evidence sweep (see `commands/build.md` Phase 6 Step 6.0).
 
@@ -475,7 +465,7 @@ Run the agent-browser dogfood skill against the running app. Unlike Track B (whi
 
 Start the dev server if not running. Then invoke the dogfood skill:
 
-Call the Agent tool — description: "Dogfood the app" — subagent_type: `testing-evidence-collector` — mode: "bypassPermissions" — prompt: "[CONTEXT header above — phase: 5] Run the agent-browser dogfood skill against the running app at http://localhost:[port]. Explore every reachable page. Click every button. Fill every form. Check console for errors. Report a structured list of issues with severity ratings (critical/high/medium/low), screenshots, and repro steps. Save screenshots under `docs/plans/evidence/dogfood/` (one PNG/JPG per finding, named after the finding_id), and emit `docs/plans/evidence/dogfood/findings.json` (machine-readable mirror of findings.md — schema: `[{finding_id, severity, description, screenshot_path, affected_screen_id}, ...]` per agents/testing-evidence-collector.md \"Dogfood Evidence Outputs\") so the Slice 5 indexer can wire `screenshot_evidences_finding` edges.
+Call the Agent tool — description: "Dogfood the app" — agent_type: `testing-evidence-collector` — subagent_type: `testing-evidence-collector` — mode: "bypassPermissions" — prompt: "[CONTEXT header above — phase: 5] Run the agent-browser dogfood skill against the running app at http://localhost:[port]. Explore every reachable page. Click every button. Fill every form. Check console for errors. Report a structured list of issues with severity ratings (critical/high/medium/low), screenshots, and repro steps. Save screenshots under `docs/plans/evidence/dogfood/` (one PNG/JPG per finding, named after the finding_id), and emit `docs/plans/evidence/dogfood/findings.json` (machine-readable mirror of findings.md — schema: `[{finding_id, severity, description, screenshot_path, affected_screen_id}, ...]` per agents/testing-evidence-collector.md \"Dogfood Evidence Outputs\") so the Slice 5 indexer can wire `screenshot_evidences_finding` edges.
 
 If dogfood skill is not available, use agent-browser manually: snapshot each page, click all interactive elements, check errors and network requests.
 
@@ -495,7 +485,7 @@ Run via the Bash tool:
 
 #### Step 5.3c — Fake Data Detector
 
-Call the Agent tool — description: "Fake data audit" — subagent_type: `silent-failure-hunter` — mode: "bypassPermissions" — prompt: "[CONTEXT header above — phase: 5] Run the Fake Data Detector Protocol (protocols/fake-data-detector.md). Check for mock/hardcoded data in production paths. Static analysis: grep for Math.random() business data, hardcoded API responses, setTimeout faking async, placeholder text. Dynamic analysis: inspect HAR files from docs/plans/evidence/ for missing real API calls, static responses, absent WebSocket traffic. Report findings with file:line references and severity."
+Call the Agent tool — description: "Fake data audit" — agent_type: `silent-failure-hunter` — subagent_type: `silent-failure-hunter` — mode: "bypassPermissions" — prompt: "[CONTEXT header above — phase: 5] Run the Fake Data Detector Protocol (protocols/fake-data-detector.md). Check for mock/hardcoded data in production paths. Static analysis: grep for Math.random() business data, hardcoded API responses, setTimeout faking async, placeholder text. Dynamic analysis: inspect HAR files from docs/plans/evidence/ for missing real API calls, static responses, absent WebSocket traffic. Report findings with file:line references and severity."
 
 **Fix loop:** For each CRITICAL finding:
 1. Spawn a fix agent with: the finding (file:line, what's fake, what it should be), and the relevant source files.
@@ -516,6 +506,6 @@ The orchestrator-side fix-loop dispatch lives in `commands/build.md` Step 5.5. M
 
 ### Step 7.1 — Documentation (web)
 
-Call the Agent tool — description: "Documentation" — subagent_type: `engineering-technical-writer` — mode: "bypassPermissions" — prompt: "[CONTEXT header above — phase: 7] Write project docs: README with setup/architecture/usage, API docs if applicable, deployment notes. Commit: 'docs: project documentation'."
+Call the Agent tool — description: "Documentation" — agent_type: `engineering-technical-writer` — subagent_type: `engineering-technical-writer` — mode: "bypassPermissions" — prompt: "[CONTEXT header above — phase: 7] Write project docs: README with setup/architecture/usage, API docs if applicable, deployment notes. Commit: 'docs: project documentation'."
 
 Deployment target per the design doc (Vercel/Netlify/Railway/Fly.io/etc.) — include the deploy flow specific to that target in the README.