qfai 1.8.2 → 1.8.3

package/assets/init/.qfai/assistant/skills/qfai-prototyping/references/l1-review-guide.md

@@ -2,10 +2,12 @@
 
 L1 checks implementation fidelity.
 
-## Inputs
+## Inputs (read from review-bundle.json)
 
-- screenshots
-- HTML snapshots
+- screenshots (round/candidate path, per declared screen)
+- HTML snapshots (round/candidate path, per declared screen)
+- accessibility snapshots (round/candidate path, per declared screen)
+- Playwright CLI command log (round/candidate path, per declared screen)
 - canonical UI contracts from `.qfai/contracts/ui/*.yaml`
 - latest code state
 
@@ -13,24 +15,23 @@ L1 checks implementation fidelity.
 
 For each declared screen:
 
-- the screen is reachable/rendered
-- screenshot exists
-- HTML snapshot exists
-- required elements are visibly present
-- required actions are wired or explicitly marked missing
+- the screen is reachable/rendered (confirm via goto log in command log)
+- screenshot, HTML, accessibility snapshot, and command log all exist at the round/candidate path
+- required elements are visibly present (cross-check screenshot + HTML + snapshot)
+- required actions are wired or explicitly marked missing (cross-check interaction commands in the command log vs `primaryTasks`)
 - blocking UI failures are identified
 
 ## Failure handling
 
-- Missing screenshot or HTML => score `0`, rerun required
+- Missing any of the 4 per-screen artifacts => score `0`, rerun required
 - Missing primary action wiring => blocking finding
 - Severe route/render failure => blocking finding
 
 ## Output
 
-Return:
+Write `evaluator-reviews/<candidate-id>.json` with:
 
 - per-screen findings
 - blocking/immediate-fix classification
-- a numeric score per axis in the range `0.0..1.0`
-- rationale tied to screenshot/HTML evidence
+- a numeric score per axis in the range `0..100`
+- rationale tied to screenshot / HTML / snapshot / command log refs (all entries in `evidenceRefs[]` MUST be concrete paths to existing artifacts)

package/assets/init/.qfai/assistant/skills/qfai-prototyping/references/l2-review-guide.md

@@ -2,14 +2,19 @@
 
 L2 checks product experience and design alignment.
 
-## Inputs
-
-- screenshots
-- HTML snapshots
-- `.qfai/contracts/design/evaluation-axes.yaml`
-- `.qfai/contracts/design/anchor-selection.yaml`
+## Inputs (read from review-bundle.json)
+
+- screenshots (round/candidate path, per declared screen)
+- HTML snapshots (round/candidate path, per declared screen)
+- accessibility snapshots (round/candidate path, per declared screen)
+- Playwright CLI command log (round/candidate path, per declared screen)
+- `.qfai/contracts/design/evaluation-rubric.yaml`
+- `.qfai/contracts/design/selected-direction.yaml`
 - `.qfai/contracts/design/design-system.yaml`
-- previous iteration score
+- legacy inputs, if present (skip if absent):
+  - `.qfai/contracts/design/evaluation-axes.yaml`
+  - `.qfai/contracts/design/anchor-selection.yaml`
+- previous round score
 
 ## 3-layer evaluation family
 
@@ -27,13 +32,14 @@ L2 must explicitly use all of:
 - product-specific differentiation is visible
 - selected anchor direction is reflected in the current UI
 - design system checklist is respected
+- interaction outcomes in the command log are consistent with the experience the designer intended
 - experience findings are recorded separately from blocking L1 findings
 
 ## Output
 
-Return:
+Write to `evaluator-reviews/<candidate-id>.json` with:
 
 - per-axis findings
 - revise/manual-review classification
-- a numeric score per axis in the range `0.0..1.0`
-- rationale tied to screenshot/HTML evidence and axis refs
+- a numeric score per axis in the range `0..100`
+- rationale tied to screenshot / HTML / snapshot / command log refs and axis refs (all entries in `evidenceRefs[]` MUST be concrete paths to existing artifacts)

package/assets/init/.qfai/assistant/skills/qfai-prototyping/references/reviewer-gate.md

@@ -1,15 +1,26 @@
 # Reviewer Gate
 
-The reviewer is an independent gate, not the implementation author.
+The reviewer is an independent gate, not the implementation author. The reviewer gate applies identically to all modes (spec-0012); modes differ only in `maxCycles`.
 
 ## Reviewer must verify
 
-- all declared screens have screenshot evidence
-- all declared screens have HTML snapshot evidence
+- all declared screens have all 4 per-screen artifacts for every active candidate in every round (screenshot, HTML, accessibility snapshot, command log)
+- canonical latest paths mirror the newest accepted winner/polish state
+- every round has `command-plans.json`, `review-bundle.json`, and per-candidate evaluator reviews
+- `review-bundle.json` contains all required fields (candidates, axisDefs, designSystemChecklist, commandPlanRef)
+- evaluator review `evidenceRefs[]` entries are concrete artifact refs (no placeholders)
 - L1 and L2 evaluators used the required inputs
 - the 3-layer evaluation family was referenced
 - missing evidence triggered rerun rather than waiver
-- `qfai validate --fail-on error` passed
+- `qfai validate --profile prototyping --fail-on error` passed
+- `prototyping.json` `maxCycles` matches the mode (no mode invariant violations)
+- winner_selected is true
+- post_selection_polish_completed is true
+- breakthrough_checked is true
+- best_of_history_present is true
+- all_reviewer_axes_perfect_100 is true
+- completion_eligible is true only after the completion certificate is valid
+- no completion claim is based on a 95-point threshold
 
 ## Reviewer output
 
@@ -21,4 +32,14 @@ Required fixes:
 - ...
 Evidence checked:
 - ...
+Gate fields:
+- mode: low-cost|standard|full-harness
+- maxCycles: <number matching mode>
+- winner_selected: true|false
+- post_selection_polish_completed: true|false
+- breakthrough_checked: true|false
+- best_of_history_present: true|false
+- all_reviewer_axes_perfect_100: true|false
+- completion_eligible: true|false
+- completion_certificate_valid: true|false
 ```

package/assets/init/.qfai/assistant/skills/qfai-sdd/SKILL.md

@@ -200,7 +200,7 @@ Follow `.qfai/assistant/instructions/shared-skill-operating-baseline.md#delta-re
 - `05_Examples.md` must include `EX-ID` and `BR-Ref` mappings.
 - `06_Test-Cases.md` must include `TC-ID`, `EX-Ref`, `AC-Refs`, and `Type`.
 - `06_Test-Cases.md` quality depth must include normal-path plus error or boundary coverage.
-- Do not complete the stage until `qfai validate --fail-on error --format github | tee .qfai/report/validate.log` exits with `error=0`.
+- Do not complete the stage until `qfai validate --profile sdd --fail-on error --format github | tee .qfai/report/validate.log` exits with `error=0`.
 - Reference direction rules from `.qfai/specs/README.md` must be enforced.
 - Keep `specs/` definition-only and operational status under `.qfai/report/run-*`.
 - Traceability depth and density-smell review rules live in:
@@ -246,7 +246,7 @@ The canonical file set is defined by skill templates under `.qfai/assistant/skil
 8. Execute Phase 2 (Slice) and pass slice gate for each target spec.
 9. Execute Phase 3 (Plan finalize) after at least one slice gate passes.
 10. Execute Phase 4 (Delta update).
-11. Run `qfai validate --fail-on error --format github | tee .qfai/report/validate.log`.
+11. Run `qfai validate --profile sdd --fail-on error --format github | tee .qfai/report/validate.log`.
 12. Review `.qfai/report/specs-coverage/spec-*.md` and triage density-smell warnings.
 13. If validate fails, fix source-layer artifacts and repeat until `error=0`.
 
@@ -314,7 +314,7 @@ When declaring DONE, include:
 - [ ] `10_Plan.md` is finalized as How-only.
 - [ ] `specs/plan.md` was not created.
 - [ ] `09_delta.md` (or `*_delta.md`) contains adoption/rejection rationale.
-- [ ] `qfai validate --fail-on error --format github` ran and produced `error=0`.
+- [ ] `qfai validate --profile sdd --fail-on error --format github` ran and produced `error=0`.
 - [ ] `.qfai/report/specs-coverage/spec-*.md` was reviewed.
 - [ ] Quality gate checks are recorded in evidence.
 - [ ] Evidence file exists and is complete.

package/assets/init/.qfai/assistant/skills/qfai-sdd/references/rcp_footer.md

@@ -31,7 +31,7 @@
 
 ## Validate Hard Gate（必須）
 
-- 各 review cycle で `qfai validate --fail-on error --format github` を実行していること
+- 各 review cycle で `qfai validate --profile sdd --fail-on error --format github` を実行していること
 - `.qfai/report/validate.log` が存在し、最新の成果物に対応していること
 
 ---

package/assets/init/.qfai/assistant/skills/qfai-sdd/references/sdd-quality-gate.md

@@ -20,7 +20,7 @@ Use this file for the full quality gate checklist behind `/qfai-sdd`.
 
 ## Validation Checks
 
-- `qfai validate --fail-on error --format github | tee .qfai/report/validate.log`
+- `qfai validate --profile sdd --fail-on error --format github | tee .qfai/report/validate.log`
 - `error=0`
 - `.qfai/report/specs-coverage/spec-*.md` reviewed
 - Density-smell warnings triaged

package/assets/init/.qfai/assistant/skills/qfai-sdd/templates/contracts/absorption-policy.sample.yaml

@@ -0,0 +1,7 @@
+# Downstream absorption policy generated by /qfai-sdd
+minAbsorptionsPerSurvivor: 2
+require_rejected_reason: true
+allow_adapt_required: true
+coherence_review:
+  block_on_regression_alert: true
+  block_on_blocking_findings: true

package/assets/init/.qfai/assistant/skills/qfai-sdd/templates/contracts/evaluation-rubric.sample.yaml

@@ -8,9 +8,26 @@ axes:
     weight: 1
   - id: functionality
     weight: 1
-hard_floors:
-  - functionality
-  - accessibility-risk
+  - id: accessibility-risk
+    weight: 1
+  - id: implementation-plausibility
+    weight: 1
 weighted_axes:
   - design-quality
   - originality
+hard_floors:
+  - id: functionality
+    min_score: 80
+  - id: accessibility-risk
+    min_score: 80
+  - id: conceptFit
+    min_score: 85
+absorbable_categories:
+  - layout
+  - interaction
+  - content-hierarchy
+  - visual-language
+  - navigation
+  - motion
+coherence:
+  regression_threshold: 5

package/assets/init/.qfai/assistant/skills/qfai-sdd/templates/contracts/evaluator-calibration.sample.yaml

@@ -1,9 +1,15 @@
 # Downstream evaluator calibration generated by /qfai-sdd
 good_critique_examples:
   - Skeptical, specific, and actionable feedback
+  - Names which candidate strengths should be harvested and why
 too_lenient_examples:
   - Praise that ignores blandness
 blandness_fail_examples:
   - Technically correct but generic
 originality_fail_examples:
   - Library defaults with no product-specific decisions
+concept_fit_fail_examples:
+  - A refinement that improves novelty but breaks the exploration brief anchors
+coherence_regression_red_flags:
+  - Newly absorbed ideas overpower the selected concept
+  - Navigation or hierarchy changes make the surface feel like a different product

package/assets/init/.qfai/assistant/skills/qfai-sdd/templates/contracts/exploration-brief.sample.yaml

@@ -8,3 +8,12 @@ brand_signals:
   - Confident
 differentiation_targets:
   - Avoid generic dashboard defaults
+parallel_candidate_routing: required
+concept_anchors:
+  - id: CONCEPT-0001
+    statement: Calm operational confidence
+  - id: CONCEPT-0002
+    statement: Guided depth without clutter
+non_goals:
+  - Loud novelty-first visuals
+  - Route switching that depends on hidden runtime state

package/assets/init/.qfai/assistant/skills/qfai-verify/SKILL.md

@@ -95,7 +95,7 @@ Use the shared schema.
 - Follow `.qfai/assistant/instructions/shared-skill-delegation-baseline.md#reviewer-gate-baseline`.
 - Reviewer checks:
   - required roles were delegated;
-  - validate evidence exists: `qfai validate --fail-on error` completed with `error=0`;
+  - validate evidence exists: `qfai validate --profile verify --fail-on error` completed with `error=0`;
   - declared screens have mandatory screenshot and HTML evidence under `.qfai/evidence/prototyping/`;
   - Drift Protocol enforced;
   - test-layer policy enforced against `test-layers.md`.
@@ -132,7 +132,7 @@ Follow `.qfai/assistant/instructions/shared-skill-operating-baseline.md#delta-re
   - `.qfai/evidence/` is intentionally NOT tracked by Git (it ships with a local `.gitignore`).
   - Do NOT commit evidence files; summarize key outcomes in the PR description instead.
 - You MUST run the mandatory checks listed below and record outcomes.
-- In CI, you MUST keep QFAI validation on default/full mode (`qfai validate --fail-on error`). Do NOT use `--phase refinement`.
+- In CI, you MUST keep QFAI validation on full-scan mode (`qfai validate --profile verify --fail-on error` or default `qfai validate --fail-on error`). Do NOT use partial profiles.
 - Waivers are only for `warning` / `info` findings. If a waiver attempts to suppress an `error`, treat it as a failure and fix the root cause.
 - You MUST stop and escalate if any gate fails without an actionable fix list.
 - Completion must be approved by a reviewer who did not run the gates.
@@ -148,7 +148,7 @@ Run quality gates and produce evidence that the change is correct and safe.
 ## Success Criteria (Definition of Done)
 
 - Repo quality gates PASS (format/lint/type/test/build/etc).
-- QFAI checks PASS (at minimum: `qfai validate`, and optionally `qfai report`).
+- QFAI checks PASS (at minimum: `qfai validate --profile verify`, and optionally `qfai report`).
 - Declared screens have mandatory screenshot and HTML evidence.
 - A concise evidence summary exists (copy‑paste for PR).
 - The PR-ready summary includes **Change Classification (Primary/Tags)** per `.qfai/assistant/instructions/change-classification.md`.
@@ -341,12 +341,12 @@ If unknown, propose defaults and mark assumptions.
 
 Run (adjust as needed):
 
-- `qfai validate --fail-on error`
+- `qfai validate --profile verify --fail-on error`
 - `qfai report` (if used in this repo)
 
 Notes:
 
-- CI must run default/full validation only. `--phase refinement` is local-only.
+- CI must run default/full validation only. Partial profiles are local skill checks only.
 - If `QFAI-WAIVER-002` appears, remove the invalid waiver and resolve the underlying `error` finding.
 
 Capture:
@@ -454,7 +454,7 @@ Evidence must include:
 1. QFAI validation:
 
    ```bash
-   qfai validate --fail-on error
+   qfai validate --profile verify --fail-on error
    ```
 
 2. Repository standard gates (discover from package.json/CI/docs):

package/assets/init/.qfai/contracts/design/README.md

@@ -6,6 +6,8 @@ Provide the downstream execution truth for exploration-first prototyping and fin
 
 These files are version-managed and may be read directly by `/qfai-prototyping`, `/qfai-implement`, `/qfai-atdd`, and `qfai validate`.
 
+> **Prototyping harness (spec-0012)**: `evaluation-rubric.yaml` is the source of evaluator axes, absorbable categories, and concept-fit hard floors. `absorption-policy.yaml` defines minimum absorption and curation expectations between rounds. `design-system.yaml` remains the downstream checklist for winner extraction and polish.
+
 ## Status After Init
 
 After `qfai init`, this directory contains only this README. This is the normal initial state. `/qfai-sdd` creates design files when a UI-bearing capability is normalized for downstream execution.
@@ -17,8 +19,9 @@ The absence of design files is not a defect for non-UI capabilities. For UI-bear
 Typical files:
 
 - `exploration-brief.yaml` — machine-readable exploration brief generated from discussion
-- `evaluation-rubric.yaml` — machine-readable evaluator rubric with weighted originality/design criteria
+- `evaluation-rubric.yaml` — machine-readable evaluator rubric with weighted axes, hard floors, and absorbable categories
 - `evaluator-calibration.yaml` — evaluator alignment examples and anti-leniency guidance
+- `absorption-policy.yaml` — round-to-round absorption thresholds and curation rules
 - `selected-direction.yaml` — current winning direction, rationale, and carry-forward rules
 - `design-system.yaml` — extracted final design system produced after direction convergence
 - `design-tokens*.yaml` — optional token definitions
@@ -28,6 +31,7 @@ Typical files:
 - `exploration-brief.yaml`
 - `evaluation-rubric.yaml`
 - `evaluator-calibration.yaml`
+- `absorption-policy.yaml`
 - `selected-direction.yaml`
 - `design-system.yaml`
 - `design-tokens.yaml`
@@ -38,3 +42,4 @@ Typical files:
 - **Not** a replacement for specs or UI contracts
 - **Not** an excuse for downstream skills to read discussion-side artifacts directly
 - **Not** a place to finalize a winner before prototyping convergence
+- **Not** a place to store round evidence; that belongs under `.qfai/evidence/prototyping/`

package/assets/init/.qfai/contracts/ui/README.md

@@ -7,6 +7,8 @@ The contract must describe screen structure, action coverage targets, and stable
 
 > **Note:** UI contracts are the downstream execution truth for screen obligations. `/qfai-sdd` may derive them from discussion-side exploration, but `/qfai-prototyping`, `/qfai-implement`, and `/qfai-atdd` must read `contracts/ui/*.yaml` instead of reading `discussion-*/uiux/40_screen_contracts.md` directly.
 
+> **Prototyping harness (spec-0012)**: `screens[].id`, `screens[].route`, and `screens[].primary_tasks` (snake_case in YAML; surfaced as `primaryTasks` in the parsed `CanonicalScreenContract`) feed the round-based command-plan builders consumed by the AI evaluator sub-agent. Changes to screen IDs or routes must propagate to `.qfai/evidence/prototyping/rounds/<rN>/candidates/<candidate-id>/<screen-id>.*` evidence.
+
 ## File rules
 
 - File name: `ui-XXXX-<slug>.yaml`

package/assets/init/.qfai/discussion/README.md

@@ -123,11 +123,15 @@ UI-bearing discussion packs (`ui_bearing: true`) may include a `prototyping.yaml
 
 ### Canonical namespaced schema (when present)
 
+The three modes (`low-cost`, `standard`, `full-harness`) share the same evidence obligations and validator gates; the only mode-dependent value is `maxCycles` (1 / 3 / 20). Choose `recommended_mode` based on the project's iteration budget, and list every mode the project allows under `allowed_modes`.
+
 ```yaml
 prototyping:
-  recommended_mode: full-harness
-  rationale: Exploration-first prototyping requires the full-harness runtime loop in packages/qfai.
+  recommended_mode: standard
+  rationale: Standard mode (maxCycles=3) matches our review cadence.
   allowed_modes:
+    - low-cost
+    - standard
     - full-harness
   surface: web
 ```
@@ -136,18 +140,19 @@ prototyping:
 
 If you create this artifact, populate all 4 fields.
 
-| Field              | Required | Description                                    |
-| ------------------ | -------- | ---------------------------------------------- |
-| `recommended_mode` | yes      | `full-harness`                                 |
-| `rationale`        | yes      | Non-empty string explaining the recommendation |
-| `allowed_modes`    | yes      | Unique array; must contain only `full-harness` |
-| `surface`          | yes      | `web`, `mobile`, `desktop`, or `mixed`         |
+| Field              | Required | Description                                                                |
+| ------------------ | -------- | -------------------------------------------------------------------------- |
+| `recommended_mode` | yes      | One of `low-cost`, `standard`, `full-harness`                              |
+| `rationale`        | yes      | Non-empty string explaining the recommendation                             |
+| `allowed_modes`    | yes      | Unique non-empty array drawn from `low-cost` / `standard` / `full-harness` |
+| `surface`          | yes      | `web`, `mobile`, `desktop`, or `mixed`                                     |
 
 ### Current behavior
 
 - Current discussion-pack readiness does not block on missing `prototyping.yaml`.
 - When `prototyping.yaml` is present, prefer the canonical namespaced schema under the `prototyping:` key.
-- `recommended_mode` should be included in `allowed_modes`. In packages/qfai, this means `recommended_mode` should be `full-harness` and `allowed_modes` should contain only `full-harness`.
+- `recommended_mode` MUST be included in `allowed_modes`.
+- Mode invariant (spec-0017 REQ-0001): the three modes share obligations except for `maxCycles` (1 for `low-cost`, 3 for `standard`, 20 for `full-harness`). Picking a different mode does not relax any other gate.
 
 ## Suggested naming
 

package/assets/init/.qfai/evidence/README.md

@@ -3,75 +3,95 @@
 ## Purpose
 
 Evidence files record what was actually executed and observed.
-`packages/qfai` v1.7.15 treats prototyping as `full-harness` only and UI-only.
+`packages/qfai` treats prototyping as a Playwright CLI + AI evaluator harness with unified obligations across all modes (spec-0012).
 
 ## Prototyping artifacts
 
-Canonical files:
+Round-scoped artifacts (for each round `<rN>`):
 
-- `.qfai/evidence/prototyping.md`
-- `.qfai/evidence/prototyping.json`
-- `.qfai/evidence/render.json`
-- `.qfai/evidence/browser-qa.json`
-- `.qfai/evidence/fullHarness.exit.json`
-- `.qfai/evidence/fullHarness.handoff.json`
-- `.qfai/evidence/fullHarness.fakeUiDetection.json`
+- `.qfai/evidence/prototyping/rounds/<rN>/command-plans.json` — candidate-aware Playwright CLI command plans
+- `.qfai/evidence/prototyping/rounds/<rN>/review-bundle.json` — evaluator input bundle
+- `.qfai/evidence/prototyping/rounds/<rN>/evaluator-reviews/<candidate-id>.json` — evaluator output per candidate
+- `.qfai/evidence/prototyping/rounds/<rN>/harvest.json` — harvest template for `r5|r3|r2`
+- `.qfai/evidence/prototyping/rounds/<rN>/narrow-decision.json` — survivor decision for `r5|r3|r2`
+- `.qfai/evidence/prototyping/rounds/<rN>/absorption-plan.json` — absorption plan for `r3|r2|r1`
+- `.qfai/evidence/prototyping/rounds/<rN>/reimplementation.json` — reimplementation record for `r3|r2|r1`
+- `.qfai/evidence/prototyping/rounds/<rN>/candidates/<candidate-id>/<screen-id>.png` — screenshot per declared screen
+- `.qfai/evidence/prototyping/rounds/<rN>/candidates/<candidate-id>/<screen-id>.html` — HTML snapshot per declared screen
+- `.qfai/evidence/prototyping/rounds/<rN>/candidates/<candidate-id>/<screen-id>.snapshot.txt` — accessibility snapshot per declared screen
+- `.qfai/evidence/prototyping/rounds/<rN>/candidates/<candidate-id>/<screen-id>.commands.json` — executed command log per declared screen
+
+Cross-round rollups:
+
+- `.qfai/evidence/prototyping.json` — `rounds[]` / `polishCycles[]` rollup with best-of-history / breakthrough / reviewer gate sections
+- `.qfai/evidence/prototyping.md` — reviewer-readable summary
+- `.qfai/evidence/breakthrough.json` — breakthrough decisions
+
+Canonical latest paths (mirror the newest accepted winner/polish state):
+
+- `.qfai/evidence/prototyping/screenshots/<screen-id>.png`
+- `.qfai/evidence/prototyping/html/<screen-id>.html`
 
 ## Execution contract
 
 Supported prototyping surfaces are `web`, `mobile`, `desktop`, and `mixed`.
 `cli`, API-only, backend-only, and `ui_bearing: false` classifications are not prototyping execution targets.
 
-## Obligation matrix
+Browser tool: `playwright-cli` (the only supported value per spec-0012).
 
-| surface / mode         | specs    | runtimeGate | uiFidelity | render evidence | browser QA | fullHarness |
-| ---------------------- | -------- | ----------- | ---------- | --------------- | ---------- | ----------- |
-| web / full-harness     | required | required    | required   | required        | required   | required    |
-| mobile / full-harness  | required | required    | required   | required        | required   | required    |
-| desktop / full-harness | required | required    | required   | required        | required   | required    |
-| mixed / full-harness   | required | required    | required   | required        | required   | required    |
+## Obligation matrix (spec-0012)
 
-`low-cost` and `standard` are unsupported in `packages/qfai` v1.7.15.
+Mode invariant: every row below is identical except for the `maxCycles` column. `maxCycles` is the only mode-dependent field.
+
+| surface / mode         | specs    | runtimeGate | uiFidelity | playwright evidence | reviewBundle | evaluatorReview | bestOfHistory | breakthrough | reviewerGate | maxCycles |
+| ---------------------- | -------- | ----------- | ---------- | ------------------- | ------------ | --------------- | ------------- | ------------ | ------------ | --------- |
+| web / low-cost         | required | required    | required   | required            | required     | required        | required      | required     | required     | 1         |
+| web / standard         | required | required    | required   | required            | required     | required        | required      | required     | required     | 3         |
+| web / full-harness     | required | required    | required   | required            | required     | required        | required      | required     | required     | 20        |
+| mobile / low-cost      | required | required    | required   | required            | required     | required        | required      | required     | required     | 1         |
+| mobile / standard      | required | required    | required   | required            | required     | required        | required      | required     | required     | 3         |
+| mobile / full-harness  | required | required    | required   | required            | required     | required        | required      | required     | required     | 20        |
+| desktop / low-cost     | required | required    | required   | required            | required     | required        | required      | required     | required     | 1         |
+| desktop / standard     | required | required    | required   | required            | required     | required        | required      | required     | required     | 3         |
+| desktop / full-harness | required | required    | required   | required            | required     | required        | required      | required     | required     | 20        |
+| mixed / low-cost       | required | required    | required   | required            | required     | required        | required      | required     | required     | 1         |
+| mixed / standard       | required | required    | required   | required            | required     | required        | required      | required     | required     | 3         |
+| mixed / full-harness   | required | required    | required   | required            | required     | required        | required      | required     | required     | 20        |
+
+Choosing a lower mode buys fewer cycles, not a weaker gate.
 
 ## Truthfulness rules
 
-- `mode.effective` must be `full-harness`.
-- `uiFidelity.mode` must be `interactive`.
-- Canonical screen contracts in `discussion-*/uiux/40_screen_contracts.md` are mandatory.
-- Browser QA is mandatory per screen.
-- Calibration is resolved from `fullHarness.calibrationRef.packPath`; scalar caller overrides are invalid.
-- `runtimeGate.evidenceRefs` must contain concrete render/browser QA/spec refs only.
-- `specCoverage` refs must use concrete declared refs plus concrete observed refs. Self-reference and synthetic strings are invalid.
+- `mode.effective` must be one of `low-cost`, `standard`, `full-harness`.
+- `maxCycles` must match `PROTOTYPING_MAX_CYCLES[mode]` or `QFAI-PROT-MODE-001` is raised.
+- Browser tool must be `playwright-cli`.
+- `uiFidelity.mode` must be `interactive` (captured via the Playwright CLI command plans).
+- Evidence capture is performed by the AI evaluator sub-agent via the Playwright CLI command plans generated by QFAI.
+- Canonical screen contracts in `.qfai/contracts/ui/*.yaml` are mandatory.
+- evaluator review `evidenceRefs[]` entries must contain concrete artifact refs that point to existing files; placeholders (`""`, `"tbd"`, `"TBD"`) are rejected.
+- Canonical latest paths must mirror the newest accepted winner/polish artifacts.
 - `mockPaths` is a negative-only ledger. Allowed values are `fail|finding` only.
 
-## fullHarness semantics
-
-Required fields:
+## Prototyping completion gate (spec-0012)
 
-- `enabled = true`
-- `runId`
-- `calibrationRef.configPath`
-- `calibrationRef.packPath`
-- `calibrationRef.packVersion`
-- `iterationCount`
-- `bestIteration`
-- `status`
-- `reviewerSignoff`
-- `reviewerLogs`
-- `iterations`
-- `scoringTrace`
-- `limitations`
+Completion requires all of the following for every mode:
 
-Review semantics:
+- all 4 per-screen artifacts present for every declared screen in the completion round / polish cycle
+- at least one `polish` cycle completed after winner selection
+- `bestOfHistory` present
+- `breakthrough` present
+- independent reviewer gate returned `PASS`
+- every reviewer sub-agent scored every evaluation axis at `100/100`
+- `qfai validate --profile prototyping --fail-on error` passes
 
-- `finalDecision = accepted` -> `reviewerSignoff.status = approved`
-- `finalDecision = rejected` -> `reviewerSignoff.status = rejected`
-- `finalDecision = abandoned` -> `reviewerSignoff.status = abandoned`
-- `reviewerLogs[last].verdict` must align with the final decision and termination semantics.
+If the polish-cycle budget is exhausted before the gate is satisfied, the run does not complete and returns `REVISE`.
 
 ## Prohibited patterns
 
-- `low-cost` or `standard` prototyping metadata
+- `browserProvider` or `renderProvider` config keys (rejected per spec-0012)
+- `playwright-mcp` as standard browser tool
+- Node Playwright direct invocation for evidence capture
+- Mode differences other than `maxCycles`
 - `cli` prototyping execution
 - self-reference such as `prototyping.json#/runtimeGate`
 - synthetic refs such as `specs: ...`

package/assets/init/root/.github/workflows/qfai-validate.yml

@@ -0,0 +1,39 @@
+# QFAI validate CI workflow
+#
+# Generated by `qfai init` (spec-0017 REQ-0009). Runs `qfai validate --profile
+# full --fail-on error` on every push to main/master and every pull request.
+#
+# The `full` profile includes the QFAI-TEST-001 test-todo stub gate, so any
+# `it.todo` / `test.todo` / `describe.todo` in your test suite will fail CI
+# (you can opt out via `validation.testStrategy.forbidTestTodoStubs: false`
+# in qfai.config.yaml).
+#
+# If your project uses pnpm / yarn instead of npm, replace the `npm ci` step
+# with your package manager's install command (e.g. `pnpm install
+# --frozen-lockfile` + `pnpm/action-setup@v4`). `npx qfai` works with any
+# Node package manager.
+
+name: qfai validate
+
+on:
+  push:
+    branches: [main, master]
+  pull_request:
+
+jobs:
+  validate:
+    name: qfai validate (full profile, fail on error)
+    runs-on: ubuntu-latest
+    timeout-minutes: 10
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-node@v4
+        with:
+          # Match the QFAI repo's CI baseline (Node 20 LTS) and the package
+          # `engines: ">=18.0.0"`. Bump deliberately when QFAI raises its
+          # supported floor.
+          node-version: "20"
+          cache: npm
+      - run: npm ci
+      - name: qfai validate
+        run: npx qfai validate --profile full --fail-on error

package/assets/init/root/qfai.config.yaml

@@ -40,5 +40,4 @@ prototyping:
     packPath: .qfai/evidence/calibration.yaml
   execution:
     targetUrl: null
-    browserProvider: playwright
-    renderProvider: playwright
+    browserTool: playwright-cli