code-ai-installer 1.5.6 → 2.1.0

package/domains/content/locales/en/.agents/skills/gates/SKILL.md

@@ -0,0 +1,236 @@
+---
+name: gates
+description: Pipeline gates control — deliverables registry, criteria by gates and modes, severity, check log
+---
+# Gates — Pipeline Gates Control
+
+## When to Use
+- **Before every transition** between gates — check the deliverable of the current gate.
+- **When returning for rework** — re-check after corrections.
+- **In case of dispute about readiness** — formal arbiter: what is done and what is not.
+
+> **Distinction from `$board`:** `$gates` checks the **quality and completeness** of the deliverable. `$board` tracks **status and progress**. Gates answers "is it ready?", Board answers "where are we now?".
+
+## Input
+
+| Field | Required | Description |
+|------|:-----------:|----------|
+| Gate Name | ✅ | Which gate we are checking |
+| Pipeline Mode | ✅ | Full / Bugfix / Hotfix — determines the set of gates and criteria |
+| Gate Deliverables | ✅ | Artifacts that the agent hands over for review |
+| Iteration | ⬚ | If a re-check — iteration number + gaps from the previous one |
+
+## Gates and Deliverables Registry
+
+### Full Pipeline
+
+| # | Gate | Agent | Mandatory Deliverables | Skills |
+|---|------|-------|--------------------------|-------|
+| 1 | **Conductor** | Conductor | Mode determined, task decomposed, board created | `$board` |
+| 2 | **Strategist** | Strategist | Audience analysis, platform strategy, content calendar, competitor analysis | `$audience-analysis`, `$platform-strategy`, `$content-calendar`, `$competitor-content-analysis` |
+| 3 | **Researcher** | Researcher | Topic research, source verification, data storytelling (if data-driven) | `$topic-research`, `$source-verification`, `$data-storytelling` |
+| 4 | **Copywriter** | Copywriter | Final text by content brief, CTA, headline | `$content-brief`, `$cta-optimization`, `$headline-formulas`, `$storytelling-framework` |
+| 5 | **Visual Concept** | Visual Concept | Visual brief, image prompts, platform visual specs | `$visual-brief`, `$image-prompt-engineering`, `$platform-visual-specs` |
+| 6 | **Reviewer** | Reviewer | Content review checklist, brand compliance, fact checking, readability | `$content-review-checklist`, `$brand-compliance`, `$fact-checking`, `$readability-scoring` |
+| 7 | **Release Gate** | Conductor | Full package for publication, final checklist | `$content-release-gate` |
+
+### Bugfix Pipeline
+
+| # | Gate | Agent | Mandatory Deliverables |
+|---|------|-------|--------------------------|
+| 1 | **Conductor** | Conductor | Problem identified, Bugfix mode, board created |
+| 2 | **Copywriter / Researcher** | Copywriter / Researcher | Fixed content / fixed data |
+| 3 | **Reviewer** | Reviewer | Fix check: review checklist (focus on fixes) |
+| 4 | **Release Gate** | Conductor | Final fix check |
+
+### Hotfix Pipeline
+
+| # | Gate | Agent | Mandatory Deliverables |
+|---|------|-------|--------------------------|
+| 1 | **Conductor** | Conductor | Problem identified, Hotfix mode, blast radius ≈ 0 confirmed |
+| 2 | **Copywriter + Reviewer** | Copywriter | Fix + self-check by mini-checklist |
+
+## Readiness Criteria by Gates
+
+### Conductor
+- [ ] Pipeline mode determined and justified
+- [ ] Task decomposed (what we do, why, scope)
+- [ ] Board created via `$board`
+- [ ] First gate activated
+
+### Strategist
+- [ ] `$audience-analysis` — personas defined, validation passed
+- [ ] `$platform-strategy` — platforms selected, frequency fixed
+- [ ] `$content-calendar` — calendar filled, balance check passed
+- [ ] `$competitor-content-analysis` — gaps identified, recommendations formulated
+- [ ] All artifacts aligned with each other (personas ↔ platforms ↔ calendar)
+
+### Researcher
+- [ ] Topic researched, key facts gathered (`$topic-research`)
+- [ ] Sources verified (`$source-verification`)
+- [ ] Data storytelling prepared (if data-driven content)
+- [ ] Facts ready for handoff to Copywriter
+
+### Copywriter
+- [ ] Text written according to `$content-brief`
+- [ ] Topic, thesis, persona match the brief
+- [ ] CTA optimized (`$cta-optimization`)
+- [ ] Volume matches the platform
+- [ ] ToV aligns with `$brand-guidelines`
+
+### Visual Concept
+- [ ] Visual brief created (`$visual-brief`)
+- [ ] AI prompts prepared (if AI generation)
+- [ ] Dimensions align with platform (`$platform-visual-specs`)
+- [ ] Style aligns with `$brand-guidelines`
+
+### Reviewer
+- [ ] `$content-review-checklist` — fully passed, decision APPROVED
+- [ ] `$brand-compliance` — status COMPLIANT
+- [ ] `$fact-checking` — status PASS / VERIFIED
+- [ ] `$readability-scoring` — within acceptable range (if applicable)
+- [ ] All comments addressed (if it was REQUIRES CHANGES)
+
+### Release Gate
+- [ ] All previous gates passed (status `[✓]` on `$board`)
+- [ ] `$content-release-gate` checklist passed — 0 blockers
+- [ ] Publication package formed
+- [ ] User sign-off obtained
+
+## Criteria Severity
+
+| Severity | Definition | Impact on Decision |
+|----------|-------------|-------------------|
+| 🔴 **Blocker** | Mandatory deliverable is missing or critically incomplete | FAIL — return for rework |
+| 🟡 **Gap** | Deliverable is present, but incomplete or has remarks | CONDITIONAL PASS — can be handed off, but gaps are logged |
+| 🔵 **Note** | Minor remark, does not affect transition | PASS — remark is passed to the next gate as context |
+
+## Decisions
+
+| Decision | Condition | Action |
+|---------|---------|----------|
+| **PASS** | 0 Blocker + 0 Gap | → Request user Approved → `$handoff` → next gate |
+| **CONDITIONAL PASS** | 0 Blocker + 1-2 Gap | → Log gaps → Request user Approved with note of gaps → `$handoff` with note |
+| **FAIL** | ≥ 1 Blocker | → Return to agent with specific list of gaps → Do not request Approved |
+
+> **Absolute rule:** Never skip gates, even if the task seems simple. FAIL → rework → re-check. No exceptions.
+
+## Protocol
+
+### Step 1 — Context Definition
+1. Determine which gate is being checked.
+2. Determine pipeline mode (Full / Bugfix / Hotfix).
+3. Load readiness criteria for this gate from the registry.
+4. If a re-check — load gaps from the previous iteration.
+
+### Step 2 — Deliverables Check
+1. For each readiness criterion — check: done ✅ / not done ❌ / N/A.
+2. For each ❌ — assign severity (🔴 Blocker / 🟡 Gap / 🔵 Note).
+3. For each ❌ — describe specific gap: what exactly is missing / what to fix.
+
+### Step 3 — Decision Making
+1. Count severity.
+2. Apply decision rules (PASS / CONDITIONAL PASS / FAIL).
+3. Add entry to check log.
+
+### Step 4 — Action by Result
+- **PASS:** show to user → request Approved → `$handoff` → update `$board`.
+- **CONDITIONAL PASS:** show gaps → request Approved confirming gaps → `$handoff` with note → update `$board`.
+- **FAIL:** show gaps → return to agent → update `$board` (status `[↩] Rework`).
+
+## Validation (Quality Gate)
+
+Gate check is considered complete if:
+
+- [ ] All readiness criteria passed (✅ / ❌ / N/A — none empty)
+- [ ] For each ❌ specified: severity, gap description, what to fix
+- [ ] Decision aligns with rules (not subjective)
+- [ ] Check log updated
+- [ ] If PASS/CONDITIONAL PASS — user Approved obtained
+- [ ] `$board` updated
+
+## Integration with other skills
+
+| Skill | Interaction |
+|------|----------------|
+| `$board` | **Update:** gate status by check result |
+| `$handoff` | **Consumer:** on PASS — handoff with attached gate check |
+| `$content-release-gate` | **Final gate:** Release Gate = the last `$gates` check |
+| All gate skills | **Criteria source:** skill deliverables = check criteria |
+
+## Anti-patterns
+
+| Mistake | Why it's bad | How to do it right |
+|--------|-------------|---------------|
+| Skipping a gate "because it's simple" | Errors go unnoticed | Never skip. No exceptions |
+| PASS without checking all criteria | Unnoticed gaps | Each criterion = ✅ / ❌ / N/A |
+| "Generally ok" instead of specifics | Agent doesn't know what to fix | Specific gap: "missing CTA in text" |
+| FAIL without gap descriptions | Agent returns and doesn't know what to do | Every FAIL = list of specific reworks |
+| Approved without user sign-off | Pipeline protocol violation | Always an explicit Approved from the user |
+| Re-check without comparing to previous gaps | Gap might not have been fixed | Load gaps from previous iteration, check each |
+| All criteria of equal importance | Trivial blocks, critical passes | Severity: Blocker / Gap / Note |
+
+## Output Template
+
+```
+### Gate Check — [Gate Name]
+
+**Mode:** [Full / Bugfix / Hotfix]
+**Iteration:** [1 / 2 / N]
+**Previous gaps:** [list or "First check"]
+
+---
+
+#### Readiness Criteria
+
+| # | Criterion | Sev | Status | Gap / Comment |
+|---|----------|-----|--------|-------------------|
+| 1 | [Criterion from registry] | 🔴 | ✅/❌ | |
+| 2 | [Criterion from registry] | 🔴 | ✅/❌ | |
+| 3 | [Criterion from registry] | 🟡 | ✅/❌ | |
+| 4 | [Criterion from registry] | 🔵 | ✅/❌/N/A | |
+
+---
+
+#### Summary
+
+| Severity | Amount | Items |
+|----------|-----------|--------|
+| 🔴 Blocker | [N] | #[list] |
+| 🟡 Gap | [N] | #[list] |
+| 🔵 Note | [N] | #[list] |
+
+### Decision: [ PASS / CONDITIONAL PASS / FAIL ]
+
+**Justification:** [by rules: 0 blocker + 0 gap = PASS]
+
+---
+
+#### On PASS / CONDITIONAL PASS:
+**Deliverables for handoff:**
+- [Artifact 1]
+- [Artifact 2]
+
+**Gaps for next gate (if CONDITIONAL):**
+- [Gap 1 — what to consider]
+
+→ Ready for handoff to **[Next Agent]**. Waiting for **"Approved"**.
+
+#### On FAIL:
+**Gaps for rework:**
+
+| # | Gap | Sev | What to fix | Responsible |
+|---|-----|-----|---------------|---------------|
+| [N] | [description] | 🔴 | [specific action] | [agent] |
+
+→ Return to **[Current Agent]**. After fix — re-check (iteration [N+1]).
+
+---
+
+#### Check Log
+
+| Iteration | Date | Decision | Gaps | Fixed |
+|----------|------|---------|-------|------------|
+| 1 | [date] | [PASS/FAIL] | [list] | — |
+| 2 | [date] | [PASS/FAIL] | [list] | [what was fixed] |
+```

package/domains/content/locales/en/.agents/skills/google-stitch-content/SKILL.md

@@ -0,0 +1,334 @@
+---
+name: google-stitch-content
+description: Visual content concepts generation via Google Stitch (AI) with MCP — carousels, covers, banners, social media visuals
+metadata:
+  version: "1.0"
+  last_verified: "2026-04-04"
+  mcp_server: "StitchMCP"
+  adapted_from: ".agents/skills/google-stitch-skill (UX/UI domain, v3.0)"
+---
+
+# Skill: Google Stitch for Content Visuals (MCP-Integrated)
+
+## Purpose
+
+Describes **how and when** Visual Concept uses Google Stitch to generate visual concepts for content:
+- Fast generation of visual layouts for social media, banners, carousel slides.
+- Rapid ideation: 2-3 visual variants in minutes.
+- HTML/CSS code generation for email templates and landing pages.
+
+> [!IMPORTANT]
+> Stitch is accessible **directly via MCP** — the agent generates, iterates, and manages screens without manual copying.
+> Stitch is a tool for **rapid prototyping of visual concepts**, not final design.
+> The output requires brand compliance checking and adaptation to `$brand-guidelines`.
+
+> Adapted from `.agents/skills/google-stitch-skill` (UX/UI domain) for the content domain.
+
+---
+
+## What is Google Stitch
+
+Google Stitch is an AI tool from Google Labs. It turns text prompts into UI layouts with export to Figma and HTML/CSS/React code.
+
+**Engine:** Gemini 3 Pro / Gemini 3 Flash.
+**URL:** https://stitch.withgoogle.com
+**Access:** Google account + StitchMCP server connected to the agent.
+
+---
+
+## When to Use in Content Pipeline
+
+### ✅ Use for
+- Generating **visual layouts** for carousel slides (Instagram, LinkedIn PDF).
+- Generating **covers** for posts, videos, articles.
+- Generating **banners** and visual blocks for email newsletters.
+- Fast creation of **layout variants** for A/B testing visuals.
+- **Email templates** — HTML/CSS scaffold for `$email-copywriting`.
+- **Landing page** sections — visual blocks for landing pages.
+
+### ❌ Do NOT use for
+- **Photo content** — Stitch generates UI, not photographs. For photos → `$image-prompt-engineering` (Midjourney/DALL-E).
+- **Illustrations and art visuals** — Stitch ≠ Midjourney. For art → `$image-prompt-engineering`.
+- Content requiring a **pixel-perfect** result — Stitch = concept, not final.
+- **Video** — Stitch does not support animations and video.
+
+### Place in the Pipeline
+
+```
+$content-brief (topic, visual requirements)
+       ↓
+$moodboard (style, palette, mood)
+       ↓
+$visual-brief (detailed spec)
+       ↓
+[STITCH MCP: Rapid Visual Ideation]    ← this skill
+  generate_screen_from_text → edit_screens → generate_variants
+       ↓
+$brand-compliance (visual check)
+       ↓
+$content-review-checklist → $content-release-gate
+```
+
+---
+
+## StitchMCP — API Reference (brief)
+
+| Tool | Purpose | When |
+|------|-----------|-------|
+| `create_project` | Create a project | New campaign / series of visuals |
+| `list_projects` / `get_project` | Find existing | Iteration of an existing campaign |
+| `list_screens` / `get_screen` | Get screens / code | Viewing results |
+| `generate_screen_from_text` | 🎯 **Generate visual from prompt** | Main tool |
+| `edit_screens` | Editing | Refining after initial generation |
+| `generate_variants` | Generate variants | A/B visuals, layout exploration |
+
+### Models
+
+| ID | When |
+|----|-------|
+| `GEMINI_3_FLASH` | First iterations, exploring variants (faster) |
+| `GEMINI_3_PRO` | Final variants, complex layouts (higher quality) |
+
+### Device Types for Content
+
+| DeviceType | Content Usage |
+|------------|----------------------|
+| `MOBILE` | Instagram Stories/Reels (9:16), mobile banners |
+| `DESKTOP` | LinkedIn carousel (16:9), email templates, landing sections |
+| `AGNOSTIC` | Square visuals (1:1), OG images, universal |
+
+---
+
+## Input
+
+| Field | Required | Description |
+|------|:-----------:|----------|
+| Visual brief | ✅ | From `$visual-brief` — what to depict, composition, mood |
+| Brand guidelines | ✅ | From `$brand-guidelines` — palette (HEX), fonts, style |
+| Platform + dimensions | ✅ | From `$platform-visual-specs` — to select deviceType |
+| Text for visual | ⬚ | Headline, CTA — if there is text on the visual |
+| Moodboard | ⬚ | From `$moodboard` — style, references |
+
+---
+
+## Operating Protocol
+
+### Step 1 — Project Preparation
+
+```
+IF campaign/series → create_project(title: "[Brand] — [Campaign]")
+IF single visual → use existing project or create a new one
+→ remember projectId
+```
+
+### Step 2 — Prompt Formulation
+
+Prompt structure for content visuals:
+
+```
+[Visual Type] + [Platform/Dimensions] + [Content Elements] + [Brand Style] + [Constraints]
+```
+
+#### Content Prompt Checklist
+
+- [ ] **Type:** Carousel slide / Banner / Email block / Post cover / OG image / Landing section
+- [ ] **Dimensions:** Exact px or ratio (1080×1080, 1080×1350, 1200×627)
+- [ ] **Content Elements:** Headline (text), subheadline, CTA button, icons, placeholder image
+- [ ] **Brand:** Colors (HEX), font, logo (position), style (minimal / bold / editorial)
+- [ ] **Mood:** From `$moodboard` or `$visual-brief`
+- [ ] **Constraints:** "No stock photos", "No illustrations", "Text only", safe zones
+
+#### Prompt Examples for Content
+
+**Instagram carousel (slide):**
+```
+Social media carousel slide, 1080×1080px.
+Bold headline "5 Email Marketing Mistakes" in white, 48px, Inter Bold.
+Background: solid dark navy (#1A1A2E). Accent: teal (#00C9A7).
+Subtext: "Mistake #1: No segmentation" in 20px, light gray.
+Bottom: page indicator dots (5 dots, first active).
+Minimal style, no illustrations, no stock photos.
+Logo small bottom-right corner.
+```
+
+**Email banner:**
+```
+Email header banner, 600×200px, desktop.
+Left: headline "Your Weekly Digest" in 24px Inter Bold, dark text (#1A1A2E).
+Right: abstract geometric shape, teal (#00C9A7) accent.
+Background: light gray (#F5F5F5). Bottom border: 2px teal line.
+Clean, minimal, SaaS-style. No photos.
+```
+
+**LinkedIn post image:**
+```
+Social media post image, 1200×627px, desktop.
+Center: large stat "340%" in bold 72px, accent color (#4361EE).
+Below: "Growth in AI-generated content in 2026" in 18px, dark gray.
+Background: white with subtle grid pattern.
+Bottom-left: small logo. Bottom-right: source attribution "McKinsey, 2026" in 12px gray.
+Data visualization style, minimal, professional.
+```
+
+### Step 3 — Generation
+
+```
+generate_screen_from_text(
+  projectId: "<id>",
+  prompt: "<prompt>",
+  deviceType: "MOBILE" | "DESKTOP" | "AGNOSTIC",
+  modelId: "GEMINI_3_FLASH"  // Flash for first iterations
+)
+→ remember screenId
+```
+
+> ⚠️ Generation can take up to a few minutes. **DO NOT repeat the call** — wait for the response.
+
+### Step 4 — Iteration (edit_screens)
+
+2-3 iterations for refinement:
+
+```
+edit_screens(
+  projectId: "<id>",
+  selectedScreenIds: ["<screenId>"],
+  prompt: "Change headline color to #00C9A7, move CTA button to bottom-center",
+  modelId: "GEMINI_3_FLASH"
+)
+```
+
+**Types of edits for content visuals:**
+- **Color:** `"Change [element] color to [HEX]"`
+- **Text:** `"Replace headline with '[new text]'"`
+- **Position:** `"Move [element] to [position]"`
+- **Size:** `"Make headline larger, 56px"`
+- **Remove:** `"Remove [element]"`
+- **Add:** `"Add CTA button '[text]' at bottom-center, teal background"`
+
+### Step 5 — Variants for A/B
+
+```
+generate_variants(
+  projectId: "<id>",
+  selectedScreenIds: ["<screenId>"],
+  prompt: "Generate layout variations with different text placement",
+  variantOptions: { numberOfVariants: 3 }
+)
+```
+
+### Step 6 — Result Retrieval
+
+```
+get_screen(projectId, screenId, name)
+→ code / metadata
+→ for carousel: repeat for each slide
+```
+
+---
+
+## Carousels — Special Workflow
+
+For Instagram / LinkedIn carousels (multiple slides):
+
+```
+1. create_project(title: "[Brand] — Carousel: [topic]")
+2. For each slide:
+   generate_screen_from_text(prompt: "Slide [N] of [total]: [slide content]")
+3. Important: specify a consistent style in every prompt:
+   "Same style as previous: [colors, font, size, style]"
+4. edit_screens for each slide → brand consistency
+5. get_screen for each → save
+```
+
+> ⚠️ Stitch struggles with consistency > 3 screens. For long carousels (10 slides) — generate in batches of 2-3, explicitly repeating the style parameters.
+
+---
+
+## Brand Compliance After Stitch
+
+Stitch **does not know** your brand. After generation — mandatory check:
+
+| Element | Check | Action if it does not match |
+|---------|----------|---------------------------|
+| Colors | Do HEX match `$brand-guidelines`? | `edit_screens` → fix colors |
+| Font | Correct font? | Stitch often uses system fonts. Mark for replacement |
+| Logo | Correct variant, position? | `edit_screens` → add/move |
+| Style | Matches `$moodboard`? | Iterate or re-generate |
+| Text | Free of errors? Correct CTA? | `edit_screens` → fix |
+| Safe zone | Is key content out of the safe zone? | `edit_screens` → move |
+
+---
+
+## Validation (Quality Gate)
+
+A Stitch visual is considered ready for handoff if:
+
+- [ ] Passed ≥ 2 iterations via `edit_screens`
+- [ ] Colors conform to `$brand-guidelines` (HEX verified)
+- [ ] Text on visual is error-free
+- [ ] Dimensions conform to `$platform-visual-specs`
+- [ ] Safe zones are respected
+- [ ] For carousel: style is consistent across slides
+- [ ] Screen ID documented
+- [ ] Prompt saved (for reproducibility)
+- [ ] A/B variants generated (if needed)
+
+---
+
+## Stitch Limitations for Content
+
+| Limitation | Impact | Workaround |
+|-------------|---------|------------|
+| Does not generate photos | Only UI/layout, not photo-realistic images | For photos → `$image-prompt-engineering` (Midjourney/DALL-E) |
+| Consistency > 3 screens is low | 10-slide carousels — varying styles | Explicitly repeat style in each prompt |
+| No animations | Cannot make animated Stories/Reels | Only static visuals |
+| Fonts are system default | Doesn't know your brand fonts | Mark for replacement, specify font in prompt |
+| Doesn't know the brand | No auto-apply brand guidelines | Explicitly state HEX, font, style in each prompt |
+| MCP generation is slow | Up to a few minutes per screen | Do not repeat call, wait |
+| Copy to Figma — UI only | MCP doesn't support Figma export | Manual step via browser |
+
+---
+
+## Anti-patterns
+
+| Mistake | Why it's bad | How to do it right |
+|--------|-------------|---------------|
+| Prompt "beautiful banner for Instagram" | Stitch won't guess the style, colors, content | Specific prompt: size + text + HEX colors + style |
+| Accepting first generation without iterations | First result is rarely perfect | Minimum 2 iterations via `edit_screens` |
+| Using Stitch for photos | Stitch generates UI, not photos | `$image-prompt-engineering` for photos |
+| Failing to check brand compliance | Colors, fonts, logos may flagrantly fail | Check after every generation |
+| Repeating call when "stuck" | Duplication of generations, budget loss | Wait. Check via `get_screen` |
+| 10 edits in one `edit_screens` | Stitch loses context | Maximum 3-5 edits at a time |
+| Stitch-output as final visual | Not brand-compliant, system fonts | Stitch = concept. Final = after brand check |
+| Carousel without unified style | Every slide looks different | Repeat style parameters in every prompt |
+
+---
+
+## Integration with other skills
+
+| Skill | Interaction |
+|------|----------------|
+| `$visual-brief` | **Source:** Spec for generation → Stitch prompt |
+| `$brand-guidelines` | **Source:** colors, fonts, style → prompt parameters |
+| `$moodboard` | **Source:** mood, style → style parameters of the prompt |
+| `$platform-visual-specs` | **Source:** dimensions, ratio → deviceType and size |
+| `$image-prompt-engineering` | **Complement:** Stitch = UI/layout, image_prompt = photo/art |
+| `$brand-compliance` | **Consumer:** checking Stitch-output for brand alignment |
+| `$email-copywriting` | **Consumer:** HTML-scaffold from Stitch for email templates |
+
+---
+
+## Output (Deliverables)
+
+```
+### Stitch Output — [Content / Campaign]
+
+**Project ID:** [id]
+**Screen IDs:** [list]
+**Prompt (saved):** [for reproducibility]
+**Iterations:** [list of edit_screens]
+**Brand Check:** [✅ colors / ✅ style / ⚠️ font — mark for replacement]
+**Variants:** [A/B — which ones, for what]
+
+→ Handoff to `$brand-compliance` → `$content-review-checklist` → `$content-release-gate`
+```