mango-brain 3.3.0__tar.gz → 3.3.2__tar.gz

mango_brain-3.3.2/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2025-2026 Federico Anastasi
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

{mango_brain-3.3.0 → mango_brain-3.3.2}/PKG-INFO

@@ -1,11 +1,12 @@
 Metadata-Version: 2.4
 Name: mango-brain
-Version: 3.3.0
+Version: 3.3.2
 Summary: The learning layer for Claude Code — persistent associative memory + development workflow. Claude Code gets smarter the longer you use it.
 Author-email: Federico Anastasi <federico.anastasi@outlook.com>
 License: MIT
 Requires-Python: >=3.11
 Description-Content-Type: text/markdown
+License-File: LICENSE
 Requires-Dist: fastapi>=0.110
 Requires-Dist: uvicorn>=0.27
 Requires-Dist: mcp>=1.0
@@ -24,6 +25,7 @@ Requires-Dist: pytest-asyncio>=0.23; extra == "dev"
 Requires-Dist: torch>=2.2; extra == "dev"
 Requires-Dist: sentence-transformers>=2.5; extra == "dev"
 Requires-Dist: scipy>=1.12; extra == "dev"
+Dynamic: license-file
 
 <p align="center">
   <img src="assets/logo.svg" alt="MangoBrain" width="80" />
@@ -41,15 +43,15 @@ Requires-Dist: scipy>=1.12; extra == "dev"
 </p>
 
 <p align="center">
-  <a href="https://pypi.org/project/mangobrain/"><img src="https://img.shields.io/pypi/v/mangobrain?style=flat-square&color=7c3aed&label=PyPI" /></a>
-  <img src="https://img.shields.io/badge/v3.3.0-7c3aed?style=flat-square" />
+  <a href="https://pypi.org/project/mango-brain/"><img src="https://img.shields.io/pypi/v/mangobrain?style=flat-square&color=7c3aed&label=PyPI" /></a>
+  <img src="https://img.shields.io/badge/v3.3.2-7c3aed?style=flat-square" />
   <img src="https://img.shields.io/badge/python-3.11+-blue?style=flat-square&logo=python&logoColor=white" />
   <img src="https://img.shields.io/badge/MCP-compatible-8A2BE2?style=flat-square" />
   <img src="https://img.shields.io/badge/license-MIT-green?style=flat-square" />
 </p>
 
 <p align="center">
-  <a href="https://mangobrain.dev">Website</a> · <a href="#getting-started">Install</a> · <a href="https://pypi.org/project/mangobrain/">PyPI</a>
+  <a href="https://mangobrain.mango-dev.space">Website</a> · <a href="#getting-started">Install</a> · <a href="https://pypi.org/project/mango-brain/">PyPI</a>
 </p>
 
 ---

{mango_brain-3.3.0 → mango_brain-3.3.2}/README.md

@@ -14,15 +14,15 @@
 </p>
 
 <p align="center">
-  <a href="https://pypi.org/project/mangobrain/"><img src="https://img.shields.io/pypi/v/mangobrain?style=flat-square&color=7c3aed&label=PyPI" /></a>
-  <img src="https://img.shields.io/badge/v3.3.0-7c3aed?style=flat-square" />
+  <a href="https://pypi.org/project/mango-brain/"><img src="https://img.shields.io/pypi/v/mangobrain?style=flat-square&color=7c3aed&label=PyPI" /></a>
+  <img src="https://img.shields.io/badge/v3.3.2-7c3aed?style=flat-square" />
   <img src="https://img.shields.io/badge/python-3.11+-blue?style=flat-square&logo=python&logoColor=white" />
   <img src="https://img.shields.io/badge/MCP-compatible-8A2BE2?style=flat-square" />
   <img src="https://img.shields.io/badge/license-MIT-green?style=flat-square" />
 </p>
 
 <p align="center">
-  <a href="https://mangobrain.dev">Website</a> · <a href="#getting-started">Install</a> · <a href="https://pypi.org/project/mangobrain/">PyPI</a>
+  <a href="https://mangobrain.mango-dev.space">Website</a> · <a href="#getting-started">Install</a> · <a href="https://pypi.org/project/mango-brain/">PyPI</a>
 </p>
 
 ---

{mango_brain-3.3.0 → mango_brain-3.3.2}/mango_brain.egg-info/PKG-INFO

@@ -1,11 +1,12 @@
 Metadata-Version: 2.4
 Name: mango-brain
-Version: 3.3.0
+Version: 3.3.2
 Summary: The learning layer for Claude Code — persistent associative memory + development workflow. Claude Code gets smarter the longer you use it.
 Author-email: Federico Anastasi <federico.anastasi@outlook.com>
 License: MIT
 Requires-Python: >=3.11
 Description-Content-Type: text/markdown
+License-File: LICENSE
 Requires-Dist: fastapi>=0.110
 Requires-Dist: uvicorn>=0.27
 Requires-Dist: mcp>=1.0
@@ -24,6 +25,7 @@ Requires-Dist: pytest-asyncio>=0.23; extra == "dev"
 Requires-Dist: torch>=2.2; extra == "dev"
 Requires-Dist: sentence-transformers>=2.5; extra == "dev"
 Requires-Dist: scipy>=1.12; extra == "dev"
+Dynamic: license-file
 
 <p align="center">
   <img src="assets/logo.svg" alt="MangoBrain" width="80" />
@@ -41,15 +43,15 @@ Requires-Dist: scipy>=1.12; extra == "dev"
 </p>
 
 <p align="center">
-  <a href="https://pypi.org/project/mangobrain/"><img src="https://img.shields.io/pypi/v/mangobrain?style=flat-square&color=7c3aed&label=PyPI" /></a>
-  <img src="https://img.shields.io/badge/v3.3.0-7c3aed?style=flat-square" />
+  <a href="https://pypi.org/project/mango-brain/"><img src="https://img.shields.io/pypi/v/mangobrain?style=flat-square&color=7c3aed&label=PyPI" /></a>
+  <img src="https://img.shields.io/badge/v3.3.2-7c3aed?style=flat-square" />
   <img src="https://img.shields.io/badge/python-3.11+-blue?style=flat-square&logo=python&logoColor=white" />
   <img src="https://img.shields.io/badge/MCP-compatible-8A2BE2?style=flat-square" />
   <img src="https://img.shields.io/badge/license-MIT-green?style=flat-square" />
 </p>
 
 <p align="center">
-  <a href="https://mangobrain.dev">Website</a> · <a href="#getting-started">Install</a> · <a href="https://pypi.org/project/mangobrain/">PyPI</a>
+  <a href="https://mangobrain.mango-dev.space">Website</a> · <a href="#getting-started">Install</a> · <a href="https://pypi.org/project/mango-brain/">PyPI</a>
 </p>
 
 ---

{mango_brain-3.3.0 → mango_brain-3.3.2}/mango_brain.egg-info/SOURCES.txt

@@ -1,3 +1,4 @@
+LICENSE
 README.md
 pyproject.toml
 mango_brain.egg-info/PKG-INFO
@@ -46,6 +47,28 @@ server/skills/health-check/SKILL.md
 server/skills/memorize/SKILL.md
 server/skills/smoke-test/SKILL.md
 server/skills/task/SKILL.md
+server/work/agents/creator.md
+server/work/agents/mem-manager.md
+server/work/agents/researcher.md
+server/work/agents/reviewer.md
+server/work/prompts/init-work.md
+server/work/prompts/memory-definition.md
+server/work/rules/mangobrain-remember-work.md
+server/work/rules/mangobrain-workflow-work.md
+server/work/skills/brain-init-work/SKILL.md
+server/work/skills/brief/SKILL.md
+server/work/skills/create/SKILL.md
+server/work/skills/elaborate-work/SKILL.md
+server/work/skills/health-check-work/SKILL.md
+server/work/skills/memorize-work/SKILL.md
+server/work/skills/smoke-test-work/SKILL.md
+server/work/templates/CLAUDE.md
+server/work/templates/rules/audience.md
+server/work/templates/rules/brand.md
+server/work/templates/rules/channels.md
+server/work/templates/rules/product.md
+server/work/templates/rules/strategy.md
+server/work/templates/rules/tone.md
 tests/test_core.py
 tests/test_init.py
 tests/test_phase2.py

{mango_brain-3.3.0 → mango_brain-3.3.2}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "mango-brain"
-version = "3.3.0"
+version = "3.3.2"
 description = "The learning layer for Claude Code — persistent associative memory + development workflow. Claude Code gets smarter the longer you use it."
 readme = "README.md"
 license = {text = "MIT"}
@@ -50,6 +50,7 @@ server = [
     "agents/*.md",
     "rules/*.md",
     "prompts/**/*.md",
+    "work/**/*.md",
     "mangobrain.default.toml",
     "dashboard_dist/**/*",
 ]

{mango_brain-3.3.0 → mango_brain-3.3.2}/server/cli.py

@@ -23,6 +23,14 @@ AGENTS_SRC = PACKAGE_DIR / "agents"
 RULES_SRC = PACKAGE_DIR / "rules"
 PROMPTS_SRC = PACKAGE_DIR / "prompts"
 
+# Work paths
+WORK_DIR = PACKAGE_DIR / "work"
+WORK_SKILLS_SRC = WORK_DIR / "skills"
+WORK_AGENTS_SRC = WORK_DIR / "agents"
+WORK_RULES_SRC = WORK_DIR / "rules"
+WORK_PROMPTS_SRC = WORK_DIR / "prompts"
+WORK_TEMPLATES_SRC = WORK_DIR / "templates"
+
 
 # ── Commands ──────────────────────────────────────────────────────────────────
 
@@ -259,6 +267,62 @@ def cmd_install(args: argparse.Namespace) -> None:
     print()
 
 
+def cmd_install_work(args: argparse.Namespace) -> None:
+    """Install MangoBrain Work into a project — marketing, content, strategy workflow."""
+    project_path = Path(args.path).resolve() if args.path else Path.cwd()
+    project_name = args.project or project_path.name
+
+    if not project_path.exists():
+        print(f"Error: path does not exist: {project_path}")
+        sys.exit(1)
+
+    print("=" * 60)
+    print("  MangoBrain Work — Installation")
+    print("=" * 60)
+
+    # ── Step 1: Install Work files ──
+    print("\n[1/4] Installing Work skills, agents, rules, and prompts...")
+    installed = _install_work_files(project_path)
+    print(f"  {len(installed)} files installed into {project_path / '.claude'}")
+
+    # ── Step 2: Install templates ──
+    print("\n[2/4] Installing templates...")
+    templates_installed = _install_work_templates(project_path)
+    print(f"  {len(templates_installed)} template files installed")
+
+    # ── Step 3: Configure MCP ──
+    print("\n[3/4] Configuring MCP server...")
+    _setup_mcp_json(project_path)
+
+    # ── Step 4: Create project structure ──
+    print("\n[4/4] Setting up project structure...")
+    briefs_dir = project_path / "briefs"
+    briefs_dir.mkdir(exist_ok=True)
+    print(f"  Created briefs/")
+
+    # Patch CLAUDE.md with Work section
+    _patch_claude_md_work(project_path, project_name)
+
+    # ── Done ──
+    print("\n" + "=" * 60)
+    print("  MangoBrain Work installed!")
+    print("=" * 60)
+    print(f"\n  Installed into: {project_path}")
+    print(f"\n  Document skills (optional, recommended):")
+    print(f"  Install via Claude Code slash commands or copy from:")
+    print(f"  https://github.com/anthropics/claude-code/tree/main/skills")
+    print(f"    - pptx  (presentations)")
+    print(f"    - xlsx  (spreadsheets)")
+    print(f"    - docx  (documents)")
+    print(f"    - pdf   (PDF manipulation)")
+    print(f"    - brand-guidelines")
+    print(f"\n  Next steps:")
+    print(f"  1. Start the server:  mangobrain serve --api")
+    print(f"  2. Restart Claude Code/Desktop")
+    print(f"  3. Run /brain-init-work to initialize brand memory")
+    print()
+
+
 def cmd_doctor(args: argparse.Namespace) -> None:
     """Check MangoBrain health."""
     print("MangoBrain Doctor")
@@ -413,6 +477,120 @@ def _install_files(project_path: Path) -> list[str]:
     return installed
 
 
+def _install_work_files(project_path: Path) -> list[str]:
+    """Copy Work skills/agents/rules/prompts into a project's .claude/ directory."""
+    installed = []
+    claude_dir = project_path / ".claude"
+
+    # Work Skills
+    if WORK_SKILLS_SRC.exists():
+        for skill_dir in WORK_SKILLS_SRC.iterdir():
+            if skill_dir.is_dir() and (skill_dir / "SKILL.md").exists():
+                dest = claude_dir / "skills" / skill_dir.name / "SKILL.md"
+                dest.parent.mkdir(parents=True, exist_ok=True)
+                shutil.copy2(skill_dir / "SKILL.md", dest)
+                installed.append(f"skills/{skill_dir.name}/SKILL.md")
+
+    # Work Agents
+    if WORK_AGENTS_SRC.exists():
+        for agent_file in WORK_AGENTS_SRC.glob("*.md"):
+            dest = claude_dir / "agents" / agent_file.name
+            dest.parent.mkdir(parents=True, exist_ok=True)
+            shutil.copy2(agent_file, dest)
+            installed.append(f"agents/{agent_file.name}")
+
+    # Work Rules
+    if WORK_RULES_SRC.exists():
+        for rule_file in WORK_RULES_SRC.glob("*.md"):
+            dest = claude_dir / "rules" / rule_file.name
+            dest.parent.mkdir(parents=True, exist_ok=True)
+            shutil.copy2(rule_file, dest)
+            installed.append(f"rules/{rule_file.name}")
+
+    # Work Prompts
+    if WORK_PROMPTS_SRC.exists():
+        prompts_dest = claude_dir / "prompts" / "mangobrain"
+        for md_file in WORK_PROMPTS_SRC.rglob("*.md"):
+            rel = md_file.relative_to(WORK_PROMPTS_SRC)
+            dest = prompts_dest / rel
+            dest.parent.mkdir(parents=True, exist_ok=True)
+            shutil.copy2(md_file, dest)
+            installed.append(f"prompts/mangobrain/{rel}")
+
+    return installed
+
+
+def _install_work_templates(project_path: Path) -> list[str]:
+    """Copy Work templates (CLAUDE.md template + rule templates) into the project."""
+    installed = []
+    templates_dest = project_path / ".claude" / "templates"
+
+    if WORK_TEMPLATES_SRC.exists():
+        for md_file in WORK_TEMPLATES_SRC.rglob("*.md"):
+            rel = md_file.relative_to(WORK_TEMPLATES_SRC)
+            dest = templates_dest / rel
+            dest.parent.mkdir(parents=True, exist_ok=True)
+            shutil.copy2(md_file, dest)
+            installed.append(f"templates/{rel}")
+
+    return installed
+
+
+def _patch_claude_md_work(project_path: Path, project_name: str) -> None:
+    """Add MangoBrain Work section to project's CLAUDE.md."""
+    claude_md_inner = project_path / ".claude" / "CLAUDE.md"
+    claude_md_root = project_path / "CLAUDE.md"
+    marker = "## MangoBrain"
+
+    if claude_md_inner.exists():
+        claude_md = claude_md_inner
+    elif claude_md_root.exists():
+        claude_md = claude_md_root
+    else:
+        claude_md = claude_md_root
+
+    if claude_md.exists():
+        content = claude_md.read_text(encoding="utf-8")
+        if marker in content:
+            return
+    else:
+        content = f"# {project_name}\n\n"
+
+    section = f"""
+{marker} Work — Memory-Powered Workflow
+
+MangoBrain provides persistent memory and a complete workflow for marketing, content, and strategy work.
+
+### Workflow Skills
+- `/brief` — Intake with memory context, produces structured brief in briefs/
+- `/create` — Full pipeline: Researcher -> Creator -> Reviewer -> Mem-manager
+- `/brain-init-work` — Initialize brand memory (guided setup)
+- `/memorize-work` — End-of-session memory sync
+- `/elaborate-work` — Periodic memory consolidation
+- `/health-check-work` — Memory health diagnosis
+- `/smoke-test-work` — Query verification
+
+### Agents (spawned by /create, not invoked directly)
+- **researcher** — Gathers context from memory, docs, web
+- **creator** — Produces deliverables (copy, presentations, documents)
+- **reviewer** — QA against brief and brand guidelines
+- **mem-manager** — Persists session knowledge
+
+### Output Structure
+- `briefs/` — Brief files ({{date}}-{{HHmm}}-{{slug}}.md)
+- `docs/` — Strategy, research, brand documents
+- `content/` — Social, copy, editorial
+- `media/` — Graphics, video
+- `deliverables/` — Presentations, reports, exports
+
+### Rules (auto-loaded)
+- `mangobrain-remember-work.md` — Query strategy for Work projects
+- `mangobrain-workflow-work.md` — Brief -> Create workflow integration
+"""
+
+    claude_md.write_text(content + section, encoding="utf-8")
+
+
 def _setup_mcp_json(project_path: Path) -> None:
     """Create or update .mcp.json with MangoBrain server entry."""
     mcp_json = project_path / ".mcp.json"
@@ -529,6 +707,12 @@ def main() -> None:
     p_install.add_argument("--project", "-p", help="Project name (default: folder name)")
     p_install.set_defaults(func=cmd_install)
 
+    # install-work
+    p_install_work = sub.add_parser("install-work", help="Install MangoBrain Work (marketing/content/strategy)")
+    p_install_work.add_argument("--path", help="Path to the project directory (default: current dir)")
+    p_install_work.add_argument("--project", "-p", help="Project name (default: folder name)")
+    p_install_work.set_defaults(func=cmd_install_work)
+
     # doctor
     p_doctor = sub.add_parser("doctor", help="Check MangoBrain system health")
     p_doctor.set_defaults(func=cmd_doctor)

{mango_brain-3.3.0 → mango_brain-3.3.2}/server/database.py

@@ -625,7 +625,8 @@ class Database:
             clauses.append("m.project = ?")
             params.append(project)
         where = f"WHERE {' AND '.join(clauses)}"
-        # Priority: fewest edges first (sparse graph nodes need connections),
+        # Priority: never-elaborated memories first (regardless of edge count),
+        # then fewest edges (sparse nodes need connections),
         # then lowest elaboration_count, then oldest elaboration_date.
         cur = await self.conn.execute(
             f"""SELECT m.*
@@ -639,6 +640,7 @@ class Database:
                 ) ec ON m.id = ec.mid
                 {where}
                 ORDER BY
+                    CASE WHEN m.elaboration_count = 0 THEN 0 ELSE 1 END ASC,
                     COALESCE(ec.edge_count, 0) ASC,
                     m.elaboration_count ASC,
                     m.elaboration_date IS NOT NULL,

mango_brain-3.3.2/server/work/agents/creator.md

@@ -0,0 +1,190 @@
+---
+name: creator
+description: Content Creator agent. Produces deliverables (copy, presentations, spreadsheets, documents, strategies) using document skills and connectors. No memory access — receives all context from Researcher.
+tools: Read, Write, Bash, Glob
+model: sonnet
+---
+
+# Creator Agent
+
+You are a Content Creator agent. You produce deliverables — copy, documents, strategies, plans — based on a brief and research context. You are creative, precise, and brand-consistent.
+
+---
+
+## TOOLS AVAILABLE
+
+- `Read` — Read reference files
+- `Write` — Create local deliverable files (.md, .txt)
+
+### Connectors (use when available)
+
+| Connector | When to use |
+|-----------|------------|
+| **Canva** | Social post graphics, carousels, stories, brand visuals |
+| **Google Drive** | Shared documents, read reference docs |
+| **Remotion** | Video content, motion graphics, animated Reels/intros |
+
+### Tool routing — which tool for which deliverable
+
+```
+Deliverable                Skill/Tool                  Save to
+─────────────────────────────────────────────────────────────────
+Social post (visual)       Canva / image description   content/social/{channel}/
+Social post (copy only)    text output                 content/social/{channel}/
+Instagram carousel         Canva (multi-page)          content/social/instagram/
+Story / Reel cover         Canva                       content/social/instagram/
+Video / Reel / Intro       Remotion                    media/video/
+Presentation / pitch deck  pptx skill (pptxgenjs)      deliverables/presentations/
+Budget / spreadsheet       xlsx skill (xlsx-populate)   deliverables/reports/
+Editorial calendar         xlsx skill or text table     content/editorial/
+Strategy document          docx skill or .md            docs/strategy/
+Brand guidelines           brand-guidelines skill       docs/brand/
+Competitor analysis        .md                          docs/research/
+Copy (landing, email, ad)  text / .md                   content/copy/
+PDF export                 pdf skill                    deliverables/
+Graphics / exports         Canva export                 media/graphics/
+Any other document         .md                          docs/
+```
+
+### Document skills (installed in .claude/skills/)
+
+These skills teach you how to produce high-quality office documents. **Always use the skill workflow** — read the skill's SKILL.md before creating the file.
+
+| Skill | Library | Use for |
+|-------|---------|---------|
+| **pptx** | pptxgenjs (JS) | Presentations, pitch decks, slide reports |
+| **xlsx** | xlsx-populate (JS) | Budgets, calendars, tracking, KPI dashboards |
+| **docx** | docx (JS/Python) | Formal documents, contracts, briefs |
+| **pdf** | various | PDF export and manipulation |
+| **brand-guidelines** | — | Structured brand book generation |
+
+**Rules:**
+- **Always produce the actual deliverable.** Don't describe what a slide should look like — create the .pptx. Don't describe a budget — create the .xlsx with formulas.
+- **Use the installed skills.** Read `.claude/skills/{skill}/SKILL.md` for the correct workflow. The skills produce much better output than ad-hoc code. The code stays on disk and can be reused as a template.
+- **Always save to the right folder.** Create the folder if it doesn't exist. Follow the project structure from CLAUDE.md.
+- **Name files clearly.** Use descriptive names with dates when relevant: `pitch-deck-investors-v1.pptx`, `instagram-post-launch-01.md`, `budget-marketing-q2-2026.xlsx`.
+- **Save the generation code.** Keep the .js/.py scripts alongside the output — they serve as reusable templates for future deliverables with the same style.
+
+## TOOLS NOT AVAILABLE
+
+- `remember` / `memorize` — You have NO access to MangoBrain. All memory context comes from the Researcher's output, pre-loaded in your prompt.
+- This is by design: you focus 100% on creation, zero retrieval overhead.
+
+---
+
+## WORKFLOW
+
+### Step 1 — Understand the assignment
+
+You receive:
+1. **Brief** — what to produce, for whom, which channel, constraints
+2. **Research context** — brand voice, audience, past content, gotchas
+
+Read both carefully. Identify:
+- Exact deliverable format (number of pieces, length, channel specs)
+- Non-negotiable constraints (brand do/don't, must-include elements)
+- Creative space (where you have freedom to propose)
+
+### Step 2 — Plan the approach
+
+Before creating, outline your approach (internally):
+- What angle/hook will you use?
+- What structure fits the format?
+- How will you differentiate alternatives?
+
+### Step 3 — Create
+
+Produce the deliverable(s) following these principles:
+
+**Copy & content:**
+- Match the tone from research context exactly
+- Lead with value/hook — no filler intros
+- CTA must be clear and actionable
+- Adapt language to the audience persona
+- If multiple pieces requested, ensure variety (different angles, not just rewording)
+
+**Strategy & planning documents:**
+- Structure with clear sections and actionable items
+- Include timelines when relevant
+- Connect recommendations to brief objectives
+- Use data/references from research context to support claims
+
+**Visual deliverables (Canva):**
+- Follow brand visual identity from research context
+- Use specified colors, fonts, logo placement
+- Format to channel specs (dimensions, safe zones)
+- For carousels: create multi-page design, consistent style across slides
+- Export in the right format for the channel (PNG for feed, MP4 for Reels)
+
+**Presentations (Canva):**
+- Use Canva presentation template — not slides as separate images
+- Professional layout, consistent with brand visual identity
+- One key message per slide — no walls of text
+- Use the brand colors and fonts from research context
+
+**Tables & calendars (text output):**
+- Structured markdown tables for editorial calendars, plans, tracking
+- Clear columns: date, channel, content type, topic, status, notes
+- Group by week or month for readability
+
+### Step 4 — Alternatives
+
+When the brief allows, produce **2-3 alternatives**:
+- **Option A**: safest, most brand-aligned
+- **Option B**: creative stretch, higher impact potential
+- **Option C** (if applicable): experimental, different angle entirely
+
+For each, include a 1-line rationale explaining the creative choice.
+
+### Step 5 — Report
+
+---
+
+## OUTPUT FORMAT
+
+```yaml
+deliverable:
+  type: "{copy | strategy | editorial_plan | presentation | social_posts | document}"
+  channel: "{from brief}"
+  tool_used: "{canva | google_drive | local_file | text_output}"
+  pieces_produced: {number}
+
+output:
+  option_a:
+    content: |
+      {the actual content}
+    rationale: "{why this approach — linked to brief/brand}"
+
+  option_b:
+    content: |
+      {the actual content}
+    rationale: "{why this approach}"
+
+  option_c:  # if applicable
+    content: |
+      {the actual content}
+    rationale: "{why this approach}"
+
+creative_choices:
+  - "{decision}: {why — linked to research context}"
+  - "{decision}: {why}"
+
+brief_compliance:
+  format: "yes | partial | no"
+  tone: "yes | partial | no"
+  constraints_met: ["{list of constraints and status}"]
+  missing: ["{anything from brief not addressed, with reason}"]
+```
+
+---
+
+## RULES
+
+- **You are not the strategist.** Follow the brief. If you disagree with the brief's direction, note it in `creative_choices` but still deliver what was asked
+- **Brand consistency is non-negotiable.** The tone, voice, and visual identity from research context are hard constraints, not suggestions
+- **No placeholder content.** Every piece must be production-ready. No "[insert X here]" or "customize this"
+- **Match the user's language** for content (Italian content → write in Italian). Research context is in English but the deliverable language follows the brief/project
+- **Don't explain the creative process at length.** The output speaks for itself. Rationale is 1 line per choice
+- **Always use the native tool.** Canva for visuals, PowerPoint for presentations, Excel for spreadsheets. Don't describe what something should look like — create it with the tool
+- **File creation**: if the deliverable is a document (strategy, plan), create a real file, don't just output text
+- **If a connector is unavailable**, fall back gracefully: describe the visual/slide/sheet in detail so the user can create it manually, and note which tool would have been used