Intelligence Marketplace — Architecture Document

Executive Summary

The Intelligence Marketplace is Uderia's collaborative ecosystem for sharing, discovering, and deploying enterprise AI assets. It transforms the platform from a single-user tool into a network where community-validated strategies, domain knowledge, agent configurations, behavioral skills, processing extensions, and entity-relationship models circulate freely — reducing token costs, accelerating onboarding, and enabling continuous improvement through fork-and-improve workflows.

Product Catalog

The marketplace serves six product types through a unified infrastructure:

Product Type	Description	Acquisition Models
Planner Repositories	Proven execution patterns and strategies for autonomous task completion	Subscribe, Fork
Knowledge Repositories	Reference documents and domain knowledge for planning context	Subscribe, Fork
Agent Packs	Bundled agent teams (coordinator, experts, knowledge collections) as portable `.agentpack` files	Subscribe, Fork
Skills	Pre-processing prompt injections that modify LLM behavior (Claude Code compatible `skill.json` + `.md`)	Install
Extensions	Reusable processing modules with tiered complexity (convention, simple, standard, LLM)	Install
Knowledge Graphs	Entity-relationship models for database topology, business concepts, and domain ontologies	Install, Fork

Design Principles

Unified Infrastructure — Shared tables (marketplace_sharing_grants) and consistent API patterns across all product types
Zero-Trust Security — JWT authentication on every endpoint, ownership validation on every mutation
Ownership Sovereignty — Publishers retain full control; consumers choose between reference subscriptions (live updates) and independent forks (full autonomy)
Admin Governance — Every product type has an administrative enable/disable toggle
Network Effects — Ratings, install counts, and subscriber metrics surface the highest-quality assets

System Architecture

Component Overview

┌──────────────────────────────────────────────────────────────────────┐
│                        Frontend (UI Layer)                           │
│  marketplaceHandler.js  │  skillHandler.js  │  agentPackHandler.js  │
│  ─────────────────────────────────────────────────────────────────── │
│  6 Product Type Tabs  │  Browse / My Assets  │  Search, Sort, Rate  │
└────────────────────────────┬─────────────────────────────────────────┘
                             │ REST API (JWT Auth)
                             ▼
┌──────────────────────────────────────────────────────────────────────┐
│                        Backend (API Layer)                           │
│  rest_routes.py          — Collections (Planner + Knowledge)        │
│  agent_pack_routes.py    — Agent Packs                              │
│  skills_routes.py        — Skills                                   │
│  extension_routes.py     — Extensions                               │
│  kg_marketplace_routes.py — Knowledge Graphs                        │
│  admin_routes.py         — Governance settings                      │
└────────────────────────────┬─────────────────────────────────────────┘
                             │
              ┌──────────────┼──────────────┐
              ▼              ▼              ▼
┌──────────────────┐ ┌──────────────┐ ┌──────────────────────────┐
│  SQLite Database  │ │   ChromaDB   │ │  File System             │
│  tda_auth.db      │ │ Vector Store │ │  marketplace_data/       │
│  ───────────────  │ │ ──────────── │ │  ──────────────────────  │
│  rag_collections  │ │  Embeddings  │ │  skills/{id}/            │
│  marketplace_*    │ │  per-coll.   │ │  agent-packs/{id}/       │
│  *_ratings        │ │              │ │  extensions/{id}/        │
│  sharing_grants   │ │              │ │  knowledge-graphs/{id}/  │
└──────────────────┘ └──────────────┘ └──────────────────────────┘

Data Flow Patterns

All product types follow three core lifecycle patterns:

Publish Flow

Publisher creates asset locally
    ↓
Publisher clicks "Publish"
    ↓
[Backend] JWT authentication + ownership validation
    ↓
[Backend] Validate: asset meets minimum requirements (e.g., ≥1 RAG case)
    ↓
[Database] Insert marketplace record (metadata, visibility, metrics)
    ↓
[Filesystem] Copy asset content to marketplace_data/{type}/{marketplace_id}/
    ↓
Asset appears in marketplace browse (if public) or targeted recipients

Subscribe / Install Flow

Consumer finds asset in marketplace
    ↓
Consumer clicks "Subscribe" / "Install"
    ↓
[Backend] JWT authentication
    ↓
[Backend] Validate: not owner (for subscribe), visibility access check
    ↓
[Database] INSERT subscription/sharing grant  OR  copy asset to user workspace
    ↓
[Database] INCREMENT subscriber_count / install_count
    ↓
Asset available in consumer's workspace

Fork Flow (Full Copy)

Consumer clicks "Fork"
    ↓
[Frontend] Display fork modal (custom name input)
    ↓
[Backend] JWT authentication
    ↓
[Database] Create new record (consumer as owner)
    ↓
[ChromaDB] Copy all embeddings to new collection (Collections only)
    ↓
[Filesystem] Copy all asset files to new directory
    ↓
Consumer owns independent copy — modify freely without affecting original

Product Type Architectures

1. Collections (Planner & Knowledge Repositories)

Collections are the foundational marketplace product. Planner Repositories contain proven execution patterns that guide the autonomous agent, while Knowledge Repositories provide reference documents and domain context for planning.

Database Tables:

-- Core collection table (pre-existing, extended for marketplace)
rag_collections (
    id, name, description, owner_user_id,
    is_marketplace_listed, visibility,      -- marketplace fields
    subscriber_count, average_rating,
    rag_case_count
)

-- Subscriptions (reference-based, no data duplication)
collection_subscriptions (
    id, user_id, collection_id, subscribed_at,
    UNIQUE(user_id, collection_id)
)

-- Ratings
collection_ratings (
    id, user_id, collection_id, rating CHECK(1-5), review,
    created_at, updated_at,
    UNIQUE(user_id, collection_id)
)

Access Control Matrix:

Action	Owner	Subscriber	Public User
View	Yes	Yes	Yes (if public)
Edit	Yes	No	No
Delete	Yes	No	No
Publish	Yes	No	No
Subscribe	No	N/A	Yes
Fork	Yes	Yes	Yes (if public)
Rate	No	Yes	Yes

API Endpoints:

Method	Endpoint	Description
GET	`/v1/marketplace/collections`	Browse with search, filters, pagination
POST	`/v1/marketplace/collections/:id/subscribe`	Subscribe (reference)
DELETE	`/v1/marketplace/subscriptions/:id`	Unsubscribe
POST	`/v1/marketplace/collections/:id/fork`	Fork (full copy incl. embeddings)
POST	`/v1/rag/collections/:id/publish`	Publish (owner only)
POST	`/v1/marketplace/collections/:id/rate`	Rate (1-5 stars + review)

Key Implementation Detail — Fork: Forking a collection is a deep copy operation that clones the database record, copies all ChromaDB vector embeddings to a new collection, and duplicates all RAG case files on disk. The consumer becomes the full owner of the fork.

2. Agent Packs

Agent Packs bundle complete agent teams — coordinator configurations, expert agents, and their associated knowledge collections — into portable .agentpack files for one-click distribution.

Pack Types: genie (full orchestrator), bundle (multi-agent), single (individual agent)

Database Tables:

-- Schema: 10_marketplace_agent_packs.sql
marketplace_agent_packs (
    id, source_installation_id, publisher_user_id,
    name, description, version, author, pack_type,
    visibility, manifest_summary_json, content_hash,
    download_count, install_count,
    created_at, updated_at
)

agent_pack_ratings (
    id, marketplace_pack_id, user_id,
    rating CHECK(1-5), comment,
    created_at, updated_at,
    UNIQUE(marketplace_pack_id, user_id)
)

API Endpoints:

Method	Endpoint	Description
POST	`/v1/agent-packs/:installation_id/publish`	Publish to marketplace
GET	`/v1/marketplace/agent-packs`	Browse published packs
GET	`/v1/marketplace/agent-packs/:id`	Detail view
POST	`/v1/marketplace/agent-packs/:id/install`	Subscribe (sharing grant)
DELETE	`/v1/marketplace/agent-packs/:id/subscribe`	Unsubscribe
POST	`/v1/marketplace/agent-packs/:id/fork`	Fork (physical copy)
POST	`/v1/marketplace/agent-packs/:id/rate`	Rate/review
DELETE	`/v1/marketplace/agent-packs/:id`	Unpublish

Key Implementation Detail — Source Tracking: Installed agent packs maintain a source_installation_id reference back to the publisher's original, enabling provenance tracking and future update notifications.

3. Skills

Skills are pre-processing prompt injections that modify LLM behavior before query execution — enforcing SQL best practices, requiring concise responses, enabling chain-of-thought reasoning, etc. Skills use the Claude Code compatible format (skill.json + <name>.md), ensuring portability between Uderia and Claude Code environments.

Skill Format:

skill-name/
├── skill.json      ← Manifest (name, version, description, author, tags, main_file)
└── skill-name.md   ← Pure markdown content (the actual prompt injection)

With optional Uderia enhancement:

{
  "name": "sql-expert",
  "version": "1.0.0",
  "description": "SQL optimization guidance",
  "author": "admin",
  "tags": ["sql", "database"],
  "main_file": "sql-expert.md",
  "uderia": {
    "allowed_params": ["strict", "lenient"],
    "param_descriptions": { "strict": "Enforce strict SQL standards" },
    "injection_target": "system_prompt",
    "icon": "database"
  }
}

Database Tables:

-- Schema: 17_marketplace_skills.sql
marketplace_skills (
    id, skill_id, publisher_user_id,
    name, version, description, author, tags_json,
    injection_target, visibility,
    install_count, download_count,
    created_at, updated_at
)

skill_ratings (
    id, marketplace_skill_id, user_id,
    rating CHECK(1-5), comment,
    created_at, updated_at,
    UNIQUE(marketplace_skill_id, user_id)
)

API Endpoints:

Method	Endpoint	Description
POST	`/v1/skills/:skill_id/publish`	Publish user skill
GET	`/v1/marketplace/skills`	Browse with search/sort/paginate
GET	`/v1/marketplace/skills/:id`	Detail + ratings
POST	`/v1/marketplace/skills/:id/install`	Install (copies to `~/.tda/skills/`)
POST	`/v1/marketplace/skills/:id/rate`	Rate (1-5 + comment)
DELETE	`/v1/marketplace/skills/:id`	Unpublish (publisher only)

Key Implementation Detail — Hot Reload: After installation, the skill manager hot-reloads so the new skill is immediately available for activation via #skill-name syntax without restart.

Export Format: Skills export as .skill files (ZIP containing skill.json + <name>.md). Import accepts both .skill and .zip.

4. Extensions

Extensions are reusable processing modules that extend agent capabilities. They follow a tiered complexity model that determines their execution environment and resource requirements.

Extension Tiers:

Tier	Description	Execution Model
`convention`	Naming/format conventions	Static rules
`simple`	Basic transformations	Lightweight processing
`standard`	Full processing pipelines	Standard execution
`llm`	LLM-powered extensions	Requires model inference

Database Tables:

marketplace_extensions (
    id, extension_id, publisher_user_id,
    name, version, description, author,
    extension_tier, tags_json, visibility,
    install_count, download_count,
    content_hash, manifest_json,
    created_at, updated_at
)

extension_ratings (
    id, marketplace_extension_id, user_id,
    rating CHECK(1-5), comment,
    created_at, updated_at,
    UNIQUE(marketplace_extension_id, user_id)
)

API Endpoints:

Method	Endpoint	Description
POST	`/v1/extensions/:id/publish`	Publish extension
GET	`/v1/marketplace/extensions`	Browse with search/sort/paginate
GET	`/v1/marketplace/extensions/:id`	Detail + ratings
POST	`/v1/marketplace/extensions/:id/install`	Install into system
POST	`/v1/marketplace/extensions/:id/rate`	Rate (1-5 + comment)
DELETE	`/v1/marketplace/extensions/:id`	Unpublish

5. Knowledge Graphs

Knowledge Graphs represent entity-relationship models — database schemas, business concepts, domain ontologies — built through the Intelligence Performance page. Published KGs can be installed into specific target profiles with merge semantics.

Database Tables:

-- Schema: 23_marketplace_knowledge_graphs.sql
marketplace_knowledge_graphs (
    id, source_profile_id, publisher_user_id,
    name, description, version, author, domain,
    entity_count, relationship_count,
    entity_types_json, relationship_types_json, tags_json,
    visibility, download_count, install_count,
    created_at, updated_at
)

knowledge_graph_ratings (
    id, marketplace_kg_id, user_id,
    rating CHECK(1-5), comment,
    created_at, updated_at,
    UNIQUE(marketplace_kg_id, user_id)
)

kg_marketplace_settings (
    id, kg_marketplace_enabled BOOLEAN DEFAULT 1,
    updated_at
)

API Endpoints:

Method	Endpoint	Description
POST	`/v1/knowledge-graph/:profile_id/publish`	Publish KG
GET	`/v1/marketplace/knowledge-graphs`	Browse with search, sort, domain filter
GET	`/v1/marketplace/knowledge-graphs/:id`	Detail + user's rating
POST	`/v1/marketplace/knowledge-graphs/:id/install`	Install into target profile
POST	`/v1/marketplace/knowledge-graphs/:id/fork`	Fork (alias for install, increments download_count)
POST	`/v1/marketplace/knowledge-graphs/:id/rate`	Rate (1-5 + comment)
DELETE	`/v1/marketplace/knowledge-graphs/:id`	Unpublish (publisher only)
GET/POST	`/v1/admin/kg-marketplace-settings`	Admin governance

Key Implementation Details:

Profile-scoped install: Unlike Skills (global), KGs install into a specific target profile via GraphStore.import_bulk(). The install modal includes a profile picker.
Merge semantics: Entities are matched by (name, entity_type) — existing entities are upserted, not replaced.
Source provenance: Installed entities receive source = 'marketplace' for tracking.
Export compatibility: Published KGs use the same format as /v1/knowledge-graph/export, with relationships containing both ID-based and name-based references.

Shared Infrastructure

Targeted Sharing (marketplace_sharing_grants)

A unified table enables fine-grained visibility control across all product types:

marketplace_sharing_grants (
    id,
    resource_type TEXT,    -- 'collection', 'agent_pack', 'skill', 'extension', 'knowledge_graph'
    resource_id INTEGER,
    granted_to_user_id TEXT,
    granted_by_user_id TEXT,
    created_at DATETIME
)

When a publisher sets visibility to targeted, they can grant access to specific users. The marketplace browse query joins against this table to filter results per-user.

Visibility Model

All product types share a two-tier visibility model:

Visibility	Behavior
Public	Discoverable by all authenticated users in marketplace browse
Targeted	Visible only to users with an explicit sharing grant

Private assets (not published) are invisible to the marketplace entirely.

Rating Architecture

Each product type has its own rating table following a consistent pattern:

1-5 star scale with optional text comment
One rating per user per asset (UNIQUE constraint, upsert on re-rate)
Self-rating prevention — publishers cannot rate their own assets
Average rating computed and cached on the marketplace record for efficient browse queries

Metrics Tracking

All product types track engagement through consistent counter fields:

Metric	Incremented On
`subscriber_count`	Subscribe action
`download_count`	Fork / download action
`install_count`	Install action
`average_rating`	Rating submission (recomputed)

Security Architecture

Authentication

All marketplace endpoints require JWT authentication:

Authorization: Bearer <JWT_TOKEN>

Token issued on login, includes user_id claim
Configurable expiration (default: 24 hours)
Middleware validates and extracts user_id before handler execution

Authorization Rules

Action	Constraint
Publish	Must be asset owner
Unpublish	Must be asset publisher
Subscribe	Cannot subscribe to own assets
Fork	Anyone with visibility access
Install	Anyone with visibility access
Rate	Cannot rate own assets
Manage sharing grants	Must be asset publisher
Admin settings	Must have admin role

Input Validation & Threat Mitigation

Threat	Mitigation
SQL Injection	Parameterized queries throughout
XSS	Output escaping (`escapeHtml()`), Content Security Policy
CSRF	JWT in Authorization header (not cookies)
Unauthorized Access	JWT validation + ownership checks on every mutation
Data Exfiltration	Visibility-based access control per asset
Content Integrity	`content_hash` on stored marketplace files

Frontend Architecture

UI Structure

The marketplace frontend is organized around two main views and six product type tabs:

Main Tabs:

Browse Marketplace — discover and acquire community assets
My Assets — manage your published and installed assets

Product Type Tabs:

Tab	Handler	Icon
Planner Repositories	`marketplaceHandler.js`	Clipboard
Knowledge Repositories	`marketplaceHandler.js`	Chart
Agent Packs	`marketplaceHandler.js`	Cube
Extensions	`marketplaceHandler.js`	Puzzle
Skills	`marketplaceHandler.js`	Wand
Knowledge Graphs	`marketplaceHandler.js`	Graph/Link

Shared UI Components:

Search bar with type-ahead
Sort dropdown (subscribers, downloads, installs, rating, recent)
Visibility filter (public / targeted)
Paginated card grid with glass-morphism design
Rating modal (interactive 5-star selection + comment)
Fork modal (custom name input + explainer)
Install modal (target selector where applicable)
Publish modal (name, description, version, visibility)

Card Design

Each product type renders branded cards with consistent structure:

Asset name and description
Publisher username
Type badge with product-specific accent color
Star rating (visual + numeric)
Engagement metrics (subscribers, installs, downloads)
Context-aware action buttons (Subscribe / Fork / Install / Rate / Publish / Unpublish)

Technology Stack

Layer	Technology
Frontend	HTML5, Tailwind CSS, Vanilla JavaScript (ES6 modules)
Backend	Python 3.10+, Quart (async Flask)
Database	SQLite (`tda_auth.db`)
Vector Store	ChromaDB (embeddings for Collections)
Authentication	JWT (PyJWT)
File Storage	Local filesystem (`marketplace_data/`)

File Structure

uderia/
├── src/trusted_data_agent/
│   ├── api/
│   │   ├── rest_routes.py              # Collection marketplace endpoints
│   │   ├── agent_pack_routes.py        # Agent Pack marketplace endpoints
│   │   ├── skills_routes.py            # Skills marketplace endpoints
│   │   ├── extension_routes.py         # Extension marketplace endpoints
│   │   ├── kg_marketplace_routes.py    # Knowledge Graph marketplace endpoints
│   │   └── admin_routes.py             # Governance settings
│   ├── agent/
│   │   └── rag_retriever.py            # Collection access control, fork logic
│   ├── skills/
│   │   └── settings.py                 # Skills marketplace governance
│   ├── kg/
│   │   └── settings.py                 # KG marketplace governance
│   └── auth/
│       ├── database.py                 # Schema bootstrap
│       └── models.py                   # Subscription, Rating models
├── static/js/handlers/
│   ├── marketplaceHandler.js           # Collections, Agent Packs, Extensions, KGs UI
│   └── skillHandler.js                 # Skills marketplace UI
├── schema/
│   ├── 10_marketplace_agent_packs.sql
│   ├── 11_marketplace_sharing_grants.sql
│   ├── 14_marketplace_extensions.sql
│   ├── 16_skill_settings.sql
│   ├── 17_marketplace_skills.sql
│   └── 23_marketplace_knowledge_graphs.sql
├── marketplace_data/                   # Published asset files
│   ├── skills/{marketplace_id}/
│   ├── agent-packs/{marketplace_id}/
│   ├── extensions/{marketplace_id}/
│   └── knowledge-graphs/{marketplace_id}/
└── templates/
    └── index.html                      # Marketplace UI

API Reference — Complete Endpoint Catalog

Collection Endpoints

Method	Endpoint	Auth	Description
GET	`/v1/marketplace/collections`	Yes	Browse collections
POST	`/v1/marketplace/collections/:id/subscribe`	Yes	Subscribe
DELETE	`/v1/marketplace/subscriptions/:id`	Yes	Unsubscribe
POST	`/v1/marketplace/collections/:id/fork`	Yes	Fork (deep copy)
POST	`/v1/rag/collections/:id/publish`	Owner	Publish
POST	`/v1/marketplace/collections/:id/rate`	Yes	Rate

Agent Pack Endpoints

Method	Endpoint	Auth	Description
POST	`/v1/agent-packs/:installation_id/publish`	Owner	Publish
GET	`/v1/marketplace/agent-packs`	Yes	Browse
GET	`/v1/marketplace/agent-packs/:id`	Yes	Detail
POST	`/v1/marketplace/agent-packs/:id/install`	Yes	Subscribe
DELETE	`/v1/marketplace/agent-packs/:id/subscribe`	Yes	Unsubscribe
POST	`/v1/marketplace/agent-packs/:id/fork`	Yes	Fork
POST	`/v1/marketplace/agent-packs/:id/rate`	Yes	Rate
DELETE	`/v1/marketplace/agent-packs/:id`	Publisher	Unpublish

Skill Endpoints

Method	Endpoint	Auth	Description
POST	`/v1/skills/:skill_id/publish`	Owner	Publish
GET	`/v1/marketplace/skills`	Yes	Browse
GET	`/v1/marketplace/skills/:id`	Yes	Detail
POST	`/v1/marketplace/skills/:id/install`	Yes	Install
POST	`/v1/marketplace/skills/:id/rate`	Yes	Rate
DELETE	`/v1/marketplace/skills/:id`	Publisher	Unpublish

Extension Endpoints

Method	Endpoint	Auth	Description
POST	`/v1/extensions/:id/publish`	Owner	Publish
GET	`/v1/marketplace/extensions`	Yes	Browse
GET	`/v1/marketplace/extensions/:id`	Yes	Detail
POST	`/v1/marketplace/extensions/:id/install`	Yes	Install
POST	`/v1/marketplace/extensions/:id/rate`	Yes	Rate
DELETE	`/v1/marketplace/extensions/:id`	Publisher	Unpublish

Knowledge Graph Endpoints

Method	Endpoint	Auth	Description
POST	`/v1/knowledge-graph/:profile_id/publish`	Owner	Publish
GET	`/v1/marketplace/knowledge-graphs`	Yes	Browse
GET	`/v1/marketplace/knowledge-graphs/:id`	Yes	Detail
POST	`/v1/marketplace/knowledge-graphs/:id/install`	Yes	Install into profile
POST	`/v1/marketplace/knowledge-graphs/:id/fork`	Yes	Fork
POST	`/v1/marketplace/knowledge-graphs/:id/rate`	Yes	Rate
DELETE	`/v1/marketplace/knowledge-graphs/:id`	Publisher	Unpublish

Admin Endpoints

Method	Endpoint	Auth	Description
GET/POST	`/v1/admin/kg-marketplace-settings`	Admin	KG marketplace toggle
GET/POST	`/v1/admin/skill-settings`	Admin	Skills marketplace toggle

Common API Patterns

Pagination (all browse endpoints):

?page=1&per_page=10

Search (all browse endpoints):

?search=keyword

Sort (all browse endpoints):

?sort_by=rating|recent|installs|downloads|subscribers

Error Response Format:

{
  "error": "Human-readable error message",
  "details": "Optional additional context"
}

HTTP Status Codes:

200 — Success
400 — Invalid input
401 — Missing/invalid JWT
403 — Insufficient permissions (or marketplace disabled)
404 — Resource not found
409 — Conflict (already subscribed, already rated, etc.)
500 — Server error

Governance & Administration

Per-Product-Type Toggles

Administrators control marketplace availability through settings tables:

Product Type	Setting	Location
Skills	`user_skills_marketplace_enabled`	`skill_settings` table
Knowledge Graphs	`kg_marketplace_enabled`	`kg_marketplace_settings` table

When disabled:

Publish returns 403 Forbidden
Marketplace browse button hidden in UI
Install returns 403 Forbidden
Existing installations are not affected

Managed via: Administration > App Config > Settings

Performance & Scalability

Current Performance Characteristics

Operation	Typical Latency	Notes
Browse (paginated)	< 200ms	JOIN-optimized queries, 10 items/page
Subscribe / Unsubscribe	< 50ms	Simple INSERT/DELETE
Rate	< 50ms	Upsert + average recomputation
Fork (Collection, 100 embeddings)	1-2s	ChromaDB copy + file copy
Install (Skill)	< 200ms	File copy + hot-reload
Install (KG, 500 entities)	1-3s	Bulk import with merge

Database Indexes

CREATE INDEX idx_subscriptions_user ON collection_subscriptions(user_id);
CREATE INDEX idx_subscriptions_collection ON collection_subscriptions(collection_id);
CREATE INDEX idx_ratings_collection ON collection_ratings(collection_id);
-- Similar indexes exist for all product-type rating and subscription tables

Scalability Considerations

Component	Current	Production Scale
Database	SQLite (single-instance)	PostgreSQL for multi-user
Vector Store	ChromaDB (local)	Distributed (Milvus, Weaviate)
File Storage	Local filesystem	S3 / blob storage
Search	SQL LIKE queries	Full-text search / Elasticsearch
Caching	None	Redis for browse result caching

Document History

Version	Date	Changes
1.0	2024	Initial guide (Phases 1-4: Collections)
1.1	Feb 2026	Added Skills (Phase 5) and Knowledge Graphs (Phase 6)
1.2	Mar 2026	Added Extensions and Agent Packs
2.0	Mar 2026	Elevated to Architecture Document — unified all 6 product types, removed phase-by-phase build log, restructured around architectural concerns

FilesExpand file tree

MARKETPLACE_ARCHITECTURE.md

Latest commit

History

MARKETPLACE_ARCHITECTURE.md

File metadata and controls

Intelligence Marketplace — Architecture Document

Executive Summary

Product Catalog

Design Principles

System Architecture

Component Overview

Data Flow Patterns

Publish Flow

Subscribe / Install Flow

Fork Flow (Full Copy)

Product Type Architectures

1. Collections (Planner & Knowledge Repositories)

2. Agent Packs

3. Skills

4. Extensions

5. Knowledge Graphs

Shared Infrastructure

Targeted Sharing (marketplace_sharing_grants)

Visibility Model

Rating Architecture

Metrics Tracking

Security Architecture

Authentication

Authorization Rules

Input Validation & Threat Mitigation

Frontend Architecture

UI Structure

Card Design

Technology Stack

File Structure

API Reference — Complete Endpoint Catalog

Collection Endpoints

Agent Pack Endpoints

Skill Endpoints

Extension Endpoints

Knowledge Graph Endpoints

Admin Endpoints

Common API Patterns

Governance & Administration

Per-Product-Type Toggles

Performance & Scalability

Current Performance Characteristics

Database Indexes

Scalability Considerations

Document History