anthropics · ShabariRepo · Mar 21, 2026
diff --git a/partner-built/bonito/.claude-plugin/plugin.json b/partner-built/bonito/.claude-plugin/plugin.json
@@ -0,0 +1,13 @@
+{
+  "name": "bonito",
+  "version": "1.0.0",
+  "description": "Deploy and manage AI infrastructure across cloud providers. Connect AWS Bedrock, Azure OpenAI, GCP Vertex AI, OpenAI, Anthropic, and Groq through a unified gateway. Create agents, configure intelligent routing with failover and cost optimization, analyze spending, and debug issues.",
+  "author": {
+    "name": "Bonito AI",
+    "url": "https://getbonito.com"
+  },
+  "repository": "https://github.com/ShabariRepo/bonito",
+  "homepage": "https://getbonito.com",
+  "license": "MIT",
+  "keywords": ["ai-infrastructure", "multi-provider", "routing", "agents", "gateway", "aws-bedrock", "azure-openai", "gcp-vertex", "cost-optimization", "failover"]
+}
diff --git a/partner-built/bonito/.mcp.json b/partner-built/bonito/.mcp.json
@@ -0,0 +1,10 @@
+{
+  "mcpServers": {
+    "bonito": {
+      "command": "bonito-mcp",
+      "env": {
+        "BONITO_API_KEY": ""
+      }
+    }
+  }
+}
diff --git a/partner-built/bonito/CONNECTORS.md b/partner-built/bonito/CONNECTORS.md
@@ -0,0 +1,54 @@
+# Connectors
+
+## How tool references work
+
+Plugin files use `~~category` as a placeholder for whatever tool the user connects in that category. For example, `~~bonito` refers to the Bonito MCP server, which provides access to the Bonito AI Gateway API.
+
+Plugins are **tool-agnostic** — they describe workflows in terms of categories (gateway, notifications, version control, etc.) rather than specific products. The `.mcp.json` pre-configures the Bonito MCP server, but optional connectors in other categories enhance the experience.
+
+## Connectors for this plugin
+
+| Category | Placeholder | Included servers | Other options |
+|----------|-------------|-----------------|---------------|
+| AI Gateway | `~~bonito` | Bonito MCP | — |
+| Version Control | `~~vcs` | — | GitHub, GitLab |
+| Notifications | `~~notifications` | — | Slack, Microsoft Teams, Discord |
+| Monitoring | `~~monitoring` | — | Datadog, Grafana, PagerDuty |
+
+## Required connector
+
+### Bonito MCP (`~~bonito`)
+
+The core connector. Provides access to the Bonito AI Gateway API for managing providers, agents, routing, and infrastructure.
+
+**Install:** `pip install bonito-mcp`
+
+**What it provides:**
+- Provider management (create, list, verify, delete)
+- Agent configuration (BonBon agents, Bonobot orchestrators)
+- Routing policy management (failover, cost-optimized, A/B testing)
+- Cost and usage analytics
+- Gateway health and diagnostics
+- Knowledge base management
+
+## Optional connectors
+
+### GitHub (`~~vcs`)
+
+Connect GitHub to enable infrastructure-as-code workflows, review bonito.yaml changes in PRs, and track deployment history.
+
+**What it adds:**
+- Read bonito.yaml configs from repos
+- Review infrastructure changes in pull requests
+- Track deployment commits and history
+- Manage infrastructure configs alongside application code
+
+### Slack (`~~notifications`)
+
+Connect Slack to receive deployment notifications, cost alerts, and provider status updates in your team channels.
+
+**What it adds:**
+- Deployment success/failure notifications
+- Cost threshold alerts
+- Provider health status updates
+- Agent error notifications
diff --git a/partner-built/bonito/LICENSE b/partner-built/bonito/LICENSE
@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Bonito AI
+
+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.
diff --git a/partner-built/bonito/README.md b/partner-built/bonito/README.md
@@ -0,0 +1,108 @@
+# Bonito Plugin
+
+An AI infrastructure management plugin primarily designed for [Cowork](https://claude.com/product/cowork), Anthropic's agentic desktop application — though it also works in Claude Code. Helps you deploy and manage AI infrastructure across cloud providers, create agents, configure intelligent routing, analyze costs, and debug issues. Works with any team — standalone with the Bonito API, supercharged when you connect GitHub, Slack, and other tools.
+
+## Installation
+
+```bash
+claude plugin install bonito
+```
+
+## Skills
+
+Domain knowledge Claude uses automatically when relevant:
+
+| Skill | Description |
+|---|---|
+| `deploy-stack` | Deploy AI infrastructure from a bonito.yaml config — create providers, agents, knowledge bases, and routing in one shot |
+| `manage-providers` | Connect, verify, and manage cloud AI providers — AWS Bedrock, Azure OpenAI, GCP Vertex AI, OpenAI, Anthropic, Groq |
+| `create-agent` | Create and configure BonBon agents or Bonobot orchestrators with system prompts, models, MCP tools, and RAG |
+| `cost-analysis` | Analyze AI spending across providers, identify expensive models, recommend cheaper alternatives, optimize routing |
+| `gateway-routing` | Configure routing policies — cost-optimized, failover, A/B testing, model aliases, and cross-region inference |
+| `debug-issues` | Troubleshoot gateway errors, provider failures, and agent issues — check logs, verify connections, test endpoints |
+
+## Example Workflows
+
+### Deploying Your AI Stack
+
+Just describe what you want:
+```
+Deploy my AI infrastructure from bonito.yaml
+```
+
+The `deploy-stack` skill reads your config file, creates providers, agents, knowledge bases, and routing policies — all in one shot. It validates each step and reports what was created.
+
+### Connecting a New Provider
+
+```
+Connect my AWS Bedrock account as a provider
+```
+
+The `manage-providers` skill walks you through credential setup, creates the provider, verifies the connection, and lists available models. Works for any supported cloud provider.
+
+### Building an Agent
+
+```
+Create a customer support agent using Claude on AWS Bedrock with our FAQ knowledge base
+```
+
+The `create-agent` skill configures a BonBon agent with your chosen model, system prompt, knowledge base, and MCP tools. It deploys the agent and gives you the endpoint.
+
+### Optimizing Costs
+
+```
+What am I spending on AI across all providers?
+```
+
+The `cost-analysis` skill pulls usage data across all connected providers, breaks down costs by model and agent, identifies expensive patterns, and recommends cheaper alternatives or routing changes.
+
+## Standalone + Supercharged
+
+Every skill works without any additional integrations:
+
+| What You Can Do | Standalone | Supercharged With |
+|-----------------|------------|-------------------|
+| Deploy infrastructure | bonito.yaml + Bonito API | GitHub (track configs in repos) |
+| Manage providers | Bonito API | Monitoring (health dashboards) |
+| Create agents | Bonito API | GitHub (version agent configs) |
+| Analyze costs | Bonito API | Slack (cost threshold alerts) |
+| Configure routing | Bonito API | Monitoring (traffic dashboards) |
+| Debug issues | Bonito API logs | Slack (error notifications), Monitoring |
+
+## MCP Integrations
+
+> If you see unfamiliar placeholders or need to check which tools are connected, see [CONNECTORS.md](CONNECTORS.md).
+
+Connect your tools for a richer experience:
+
+| Category | Examples | What It Enables |
+|---|---|---|
+| **AI Gateway** | Bonito MCP | Provider management, agent creation, routing, cost analytics |
+| **Version Control** | GitHub, GitLab | Infrastructure-as-code, config versioning, PR reviews |
+| **Notifications** | Slack, Teams, Discord | Deployment alerts, cost warnings, provider status updates |
+| **Monitoring** | Datadog, Grafana | Traffic dashboards, latency tracking, health monitoring |
+
+See [CONNECTORS.md](CONNECTORS.md) for the full list of supported integrations.
+
+## Settings
+
+Create a `settings.local.json` file to personalize:
+
+- **Cowork**: Save it in any folder you've shared with Cowork (via the folder picker). The plugin finds it automatically.
+- **Claude Code**: Save it at `bonito/.claude/settings.local.json`.
+
+```json
+{
+  "organization": "Your Company",
+  "bonito_api_url": "https://api.getbonito.com",
+  "default_model": "anthropic/claude-sonnet-4-20250514",
+  "preferences": {
+    "default_provider": "aws-bedrock",
+    "cost_alert_threshold": 100,
+    "routing_strategy": "cost-optimized",
+    "region": "us-east-1"
+  }
+}
+```
+
+The plugin will ask you for this information interactively if it's not configured.
diff --git a/partner-built/bonito/skills/cost-analysis/SKILL.md b/partner-built/bonito/skills/cost-analysis/SKILL.md
@@ -0,0 +1,75 @@
+---
+name: cost-analysis
+description: "Analyze AI spending across providers, identify expensive models, recommend cheaper alternatives, and optimize routing for cost savings. Triggers on 'what am I spending', 'cost breakdown', 'optimize my AI costs', 'which models cost the most', or 'reduce my AI costs'."
+---
+
+# Cost Analysis
+
+Understand and optimize AI spending across all connected providers. Pull usage data, break down costs, identify expensive patterns, and recommend concrete savings.
+
+## Step 1: Determine Scope
+
+- "What am I spending?" -> full overview, current month
+- "Costs by provider" -> provider breakdown
+- "Most expensive models" -> model ranking
+- "Compare months" -> trend analysis
+- "Optimize costs" -> recommendations focus
+- "Last 7 days" -> custom date range
+
+## Step 2: Pull Usage Data
+
+Query ~~bonito for:
+1. Total requests, tokens, and costs for the period
+2. Breakdown by provider
+3. Breakdown by model
+4. Breakdown by agent
+5. Daily time series for trends
+
+## Step 3: Analyze Patterns
+
+- Which models account for the most spending?
+- Are agents using expensive models for simple tasks?
+- Is usage concentrated or spread across providers?
+- Are there traffic spikes or retry storms?
+- Is token usage efficient (prompt length vs response length)?
+
+## Step 4: Generate Recommendations
+
+**Model substitution:**
+- Opus for simple tasks -> suggest Haiku or Sonnet
+- GPT-4o for classification -> suggest GPT-4o-mini
+- High-volume agent -> suggest Groq for speed + cost
+
+**Routing optimization:**
+- One provider consistently more expensive -> route to cheaper alternative
+- Bursty traffic -> suggest response caching
+- Quality varies -> A/B test cheaper models
+
+## Step 5: Present Report
+
+```
+## AI Cost Analysis
+
+Period: [Start] to [End]
+Total Spend: $[Amount] | Requests: [Count] | Tokens: [Count]
+
+### By Provider
+| Provider | Requests | Cost | % of Total |
+|----------|----------|------|------------|
+| AWS Bedrock | [Count] | $[Amount] | [X]% |
+| OpenAI | [Count] | $[Amount] | [X]% |
+
+### By Model (Top 5)
+| Model | Provider | Requests | Cost | $/1K Req |
+|-------|----------|----------|------|----------|
+| claude-sonnet | Bedrock | [Count] | $[Amount] | $[Amount] |
+
+### Recommendations
+1. [Action]: Switch [agent] from [expensive model] to [cheaper model]
+   Estimated savings: $[Amount]/month ([X]%)
+
+2. [Action]: [Description]
+   Estimated savings: $[Amount]/month ([X]%)
+
+Potential monthly savings: $[Amount] ([X]% reduction)
+```
diff --git a/partner-built/bonito/skills/create-agent/SKILL.md b/partner-built/bonito/skills/create-agent/SKILL.md
@@ -0,0 +1,82 @@
+---
+name: create-agent
+description: "Create and configure BonBon agents or Bonobot orchestrators with system prompts, models, MCP tools, and RAG knowledge bases. Triggers on 'create an agent', 'deploy a chatbot', 'set up a support bot', 'build an orchestrator', 'make a BonBon', or 'create a Bonobot'."
+---
+
+# Create Agent
+
+Create AI agents on the Bonito gateway. Handles both BonBon agents (single-model, task-focused) and Bonobot orchestrators (multi-agent, tool-using).
+
+## Step 1: Determine Agent Type
+
+- Simple task, single model, "chatbot", "support bot" -> **BonBon**
+- Multi-step, coordination, "orchestrator", "multi-agent" -> **Bonobot**
+
+If unclear, ask: "Do you need a single-purpose agent or a multi-agent orchestrator?"
+
+## Step 2: Select Model and Provider
+
+1. List available providers via ~~bonito
+2. Recommend model based on use case:
+   - Customer support -> Claude Sonnet (good balance)
+   - Complex reasoning -> Claude Opus or GPT-4o
+   - High volume, simple tasks -> Claude Haiku or Groq Llama
+   - Code generation -> Claude Sonnet or GPT-4o
+3. Confirm with user
+
+## Step 3: Configure System Prompt
+
+If user provides one, use it directly. If they describe the role:
+1. Generate a draft (role, tone, boundaries, output format)
+2. Present for approval
+3. Iterate until satisfied
+
+## Step 4: Attach Knowledge Base (Optional)
+
+1. Check for existing KBs via ~~bonito
+2. If new: create KB, upload documents, wait for indexing
+3. Configure retrieval settings (top-K, similarity threshold)
+
+## Step 5: Configure MCP Tools (Optional)
+
+Select and attach relevant MCP tool servers for the agent's purpose.
+
+## Step 6: Create, Deploy, and Test
+
+1. Assemble config and call ~~bonito to create
+2. Send a basic greeting -> verify response
+3. Send a domain query -> verify relevance
+4. If tools/KB attached -> test those too
+
+## Output Format
+
+```
+## Agent Created: [Name]
+
+| Field | Value |
+|-------|-------|
+| Type | [BonBon / Bonobot] |
+| Model | [Model] via [Provider] |
+| Agent ID | [ID] |
+| Endpoint | /v1/agents/[id]/chat |
+| Knowledge Base | [Name or None] |
+| MCP Tools | [Count or None] |
+| Status | Deployed |
+
+### Test Results
+| Test | Status |
+|------|--------|
+| Basic greeting | OK |
+| Domain query | OK |
+
+### Quick Start
+curl -X POST https://api.getbonito.com/v1/agents/[id]/chat \
+  -H "Authorization: Bearer $BONITO_API_KEY" \
+  -d '{"message": "Hello!"}'
+```
+
+## Agent Types
+
+**BonBon:** Single-model agent for focused tasks (support bots, Q&A, code review, content generation). One model, optional system prompt, optional KB, optional MCP tools.
+
+**Bonobot:** Multi-agent orchestrator that delegates to specialized sub-agents. For complex workflows spanning multiple domains or requiring different models for different tasks.