@bluefly/openstandardagents 0.4.6 → 0.4.9

package/examples/agentscope/react-assistant/README.md

@@ -0,0 +1,111 @@
+# AgentScope ReAct Assistant -- OSSA Example
+
+This example demonstrates an OSSA agent manifest that uses [AgentScope](https://github.com/modelscope/agentscope) as its runtime. It showcases how OSSA's declarative manifest format integrates with AgentScope's `ReActAgent` class, MCP tool servers, the A2A protocol, and Mem0 long-term memory.
+
+## What This Example Shows
+
+- **AgentScope runtime binding** via the `extensions.agentscope` block, specifying `ReActAgent` as the agent class, iteration limits, and memory backend.
+- **MCP tool integration** with two tool servers: a graph knowledge base (GKG) and a general web search/content retrieval endpoint.
+- **A2A protocol support** exposing the agent as a discoverable service with an agent card and skill declarations.
+- **Token efficiency controls** including model cascading (Haiku -> Sonnet -> Opus), adaptive budget allocation, and context compression.
+- **Security boundaries** with allow-listed domains, required capabilities, and data classification.
+
+## Manifest Structure
+
+| Section | Purpose |
+|---------|---------|
+| `metadata` | Identity, labels, publisher info |
+| `spec` | Role prompt, LLM config, taxonomy, tools, input/output schema |
+| `security` | Tier classification, network policy, data classification |
+| `protocols.mcp` | MCP client capabilities (tools, resources) |
+| `protocols.a2a` | A2A endpoint, agent card, skills |
+| `token_efficiency` | Budget, model cascade routing, serialization profile |
+| `extensions.agentscope` | AgentScope-specific: agent class, memory, orchestration, compression |
+
+## How AgentScope Maps to OSSA
+
+The `extensions.agentscope` block is the runtime-specific configuration:
+
+```yaml
+extensions:
+  agentscope:
+    version: "1.0.16"
+    agent_class: ReActAgent       # AgentScope agent type
+    capabilities: [rag, parallel_tool_calls]
+    memory_backend: mem0          # Long-term memory via Mem0
+    orchestration: msghub         # AgentScope MsgHub for multi-agent
+    max_iters: 10                 # ReAct loop iteration cap
+    formatter: anthropic          # Message format for Anthropic models
+    compression:
+      enable: true
+      trigger_threshold: 50000    # Compress context above 50k tokens
+      keep_recent: 5              # Always keep last 5 messages
+    skill_dirs: [./skills/]       # Local skill modules
+```
+
+An OSSA-to-AgentScope adapter would read this manifest and instantiate:
+
+```python
+import agentscope
+from agentscope.agents import ReActAgent
+
+agentscope.init(model_configs=[{
+    "config_name": "anthropic",
+    "model_type": "anthropic_chat",
+    "model_name": "claude-sonnet-4-20250514",
+    "temperature": 0.7,
+    "max_tokens": 4096,
+    "stream": True,
+}])
+
+agent = ReActAgent(
+    name="agentscope-react-assistant",
+    model_config_name="anthropic",
+    sys_prompt=manifest["spec"]["role"],
+    tools=loaded_mcp_tools,
+    max_iters=10,
+)
+```
+
+## MCP Tool Integration
+
+The manifest declares two MCP tool servers under `spec.tools`:
+
+1. **knowledge-base** -- Graph knowledge base at `gkg.blueflyagents.com` for codebase analysis and structured queries.
+2. **web-search** -- General MCP server at `mcp.blueflyagents.com` for web search and content retrieval.
+
+At runtime, the adapter connects to these SSE endpoints, discovers available tools, and registers them with the AgentScope `ReActAgent` tool list.
+
+## A2A Protocol
+
+The `protocols.a2a` block exposes this agent as an A2A-compatible service:
+
+- **Endpoint**: `https://agents.blueflyagents.com/a2a/react-assistant`
+- **Skills**: `question-answering`, `code-analysis`
+- **Streaming**: Enabled for real-time responses
+
+Other agents can discover and invoke this agent via the A2A protocol using the published agent card.
+
+## Token Efficiency
+
+The manifest uses a three-tier model cascade to optimize cost:
+
+1. Queries below complexity 0.3 route to `claude-haiku-4-5-20251001` (cheapest).
+2. Queries between 0.3 and 0.7 route to `claude-sonnet-4-20250514`.
+3. Complex queries above 0.7 route to `claude-opus-4-20250514`.
+
+Context compression triggers at 50k tokens, keeping the 5 most recent messages intact.
+
+## Prerequisites
+
+- AgentScope >= 1.0.16
+- Anthropic API key
+- Network access to `*.blueflyagents.com`
+- (Optional) Mem0 backend for persistent memory
+
+## Related
+
+- [OSSA Specification](https://openstandardagents.org)
+- [AgentScope Documentation](https://doc.agentscope.io/)
+- [MCP Protocol](https://modelcontextprotocol.io/)
+- [A2A Protocol](https://google.github.io/A2A/)

package/examples/agentscope/react-assistant/agent.ossa.yaml

@@ -0,0 +1,132 @@
+apiVersion: ossa/v0.4.6
+kind: Agent
+metadata:
+  name: agentscope-react-assistant
+  namespace: bluefly/examples
+  version: "1.0.0"
+  description: "Example ReAct agent powered by AgentScope runtime with MCP tool integration, A2A protocol support, and Mem0 long-term memory."
+  agentType: agentscope
+  agentKind: assistant
+  status: active
+  labels:
+    tier: experimental
+    category: example
+    framework: agentscope
+  identity:
+    agent_id: agentscope-react-assistant
+    namespace: bluefly/examples
+    version: "1.0.0"
+    publisher:
+      name: BlueFly Collective
+      website: https://bluefly.io
+
+spec:
+  role: >
+    You are a helpful assistant that uses tools to answer questions accurately.
+    Always verify information before responding. Use available MCP tools for
+    real-time data retrieval and knowledge base queries.
+  llm:
+    provider: anthropic
+    model: claude-sonnet-4-20250514
+    temperature: 0.7
+    maxTokens: 4096
+    streaming: true
+  taxonomy:
+    domain: agents
+    subdomain: assistant
+    capability: react-tool-use
+  tools:
+    - type: mcp
+      name: knowledge-base
+      endpoint: https://gkg.blueflyagents.com/mcp/sse
+      description: "Graph knowledge base for codebase analysis"
+    - type: mcp
+      name: web-search
+      endpoint: https://mcp.blueflyagents.com/api/mcp/sse
+      description: "Web search and content retrieval"
+  inputs:
+    type: object
+    properties:
+      query:
+        type: string
+        description: "User query or task"
+    required: [query]
+  outputs:
+    type: object
+    properties:
+      response:
+        type: string
+      tool_calls_made:
+        type: integer
+      confidence:
+        type: number
+
+security:
+  tier: tier_1_read
+  capabilities:
+    required: [network-outbound, tool-execution]
+    optional: [memory-extended]
+  network_access:
+    allowed_domains:
+      - "*.blueflyagents.com"
+      - "api.anthropic.com"
+    protocols: [https, wss]
+    egress_policy: allow-list
+  data_classification: internal
+
+protocols:
+  mcp:
+    version: "1.0.0"
+    role: client
+    capabilities:
+      tools: true
+      resources: true
+      prompts: false
+      sampling: false
+  a2a:
+    version: "0.2.1"
+    endpoint: https://agents.blueflyagents.com/a2a/react-assistant
+    agent_card:
+      name: AgentScope React Assistant
+      description: "ReAct agent with tool use and memory"
+      skills:
+        - id: question-answering
+          name: Question Answering
+          description: "Answer questions using tools and knowledge"
+        - id: code-analysis
+          name: Code Analysis
+          description: "Analyze code using knowledge graph"
+    capabilities:
+      streaming: true
+      pushNotifications: false
+      stateTransitionHistory: false
+
+token_efficiency:
+  serialization_profile: compact
+  observation_format: projected
+  budget:
+    max_input_tokens: 100000
+    max_output_tokens: 4096
+    allocation_strategy: adaptive
+  routing:
+    cascade: [claude-haiku-4-5-20251001, claude-sonnet-4-20250514, claude-opus-4-20250514]
+    complexity_threshold: [0.3, 0.7]
+    strategy: cost_optimized
+
+extensions:
+  agentscope:
+    version: "1.0.16"
+    agent_class: ReActAgent
+    capabilities:
+      - rag
+      - parallel_tool_calls
+    memory_backend: mem0
+    orchestration: msghub
+    max_iters: 10
+    formatter: anthropic
+    compression:
+      enable: true
+      trigger_threshold: 50000
+      keep_recent: 5
+    skill_dirs:
+      - ./skills/

package/examples/agentscope/react-assistant/skills/code-analysis/SKILL.md

@@ -0,0 +1,18 @@
+---
+name: code_analysis
+description: Analyze source code for quality issues, security vulnerabilities, and architectural patterns using the knowledge graph.
+---
+
+# Code Analysis Skill
+
+Use the knowledge graph MCP tools to analyze codebases:
+
+1. Query the graph for definitions and references
+2. Identify patterns and anti-patterns
+3. Check for security vulnerabilities
+4. Report findings with confidence scores
+
+## Available Tools
+- `search_codebase_definitions` — Find code definitions
+- `get_references` — Trace references
+- `repo_map` — Get repository structure

package/examples/agentscope/react-assistant/skills/question-answering/SKILL.md

@@ -0,0 +1,18 @@
+---
+name: question_answering
+description: Answer questions accurately using web search and knowledge base tools, with source citations.
+---
+
+# Question Answering Skill
+
+Answer user questions by:
+
+1. Search the knowledge base first
+2. Fall back to web search if needed
+3. Synthesize information from multiple sources
+4. Always cite sources with URLs
+
+## Strategy
+- Prefer authoritative sources
+- Cross-reference multiple results
+- State confidence level in answers

package/examples/autonomous-evolution/self-evolving-agent.ossa.yaml

@@ -1,4 +1,4 @@
-apiVersion: ossa/v0.4.6
+apiVersion: ossa/v0.4.7
 kind: Agent
 metadata:
   name: self-evolving-agent

package/examples/bridges/.gitlab-ci.yml

@@ -68,9 +68,9 @@ validate:ossa:
 
 build:docker:
   stage: build
-  image: docker:24-dind
+  image: docker:27-dind
   services:
-    - docker:24-dind
+    - docker:27-dind
   before_script:
     - docker login -u $CI_REGISTRY_USER -p $CI_REGISTRY_PASSWORD $DOCKER_REGISTRY
   script:

package/examples/build-once-use-everywhere/agent.ossa.yaml

@@ -1,4 +1,4 @@
-apiVersion: ossa/v0.4.6
+apiVersion: ossa/v0.4.7
 kind: Agent
 metadata:
   name: universal-agent

package/examples/claude-code/code-reviewer.ossa.yaml

@@ -1,4 +1,4 @@
-apiVersion: ossa/v0.4.6
+apiVersion: ossa/v0.4.7
 kind: Agent
 metadata:
   name: claude-code-reviewer

package/examples/claude-code/ossa-validator.ossa.yaml

@@ -1,4 +1,4 @@
-apiVersion: ossa/v0.4.6
+apiVersion: ossa/v0.4.7
 kind: Agent
 metadata:
   name: ossa-validator

package/examples/common_npm/agent-router.ossa.yaml

@@ -1,4 +1,4 @@
-apiVersion: ossa/v0.4.6
+apiVersion: ossa/v0.4.7
 kind: Agent
 metadata:
   name: agent-router

package/examples/contracts/data-consumer.ossa.yaml

@@ -8,7 +8,7 @@
 # 3. Contract validation between consumer and provider
 # 4. Schema compatibility testing
 
-apiVersion: ossa/v0.4.6
+apiVersion: ossa/v0.4.7
 kind: Agent
 metadata:
   name: data-consumer

package/examples/contracts/data-producer-v2.ossa.yaml

@@ -11,7 +11,7 @@
 # Use this to test:
 #   ossa contract breaking-changes examples/contracts/data-producer.ossa.yaml examples/contracts/data-producer-v2.ossa.yaml
 
-apiVersion: ossa/v0.4.6
+apiVersion: ossa/v0.4.7
 kind: Agent
 metadata:
   name: data-producer

package/examples/contracts/data-producer.ossa.yaml

@@ -8,7 +8,7 @@
 # 3. Test schema compatibility with consumer agents
 # 4. Detect breaking changes between versions
 
-apiVersion: ossa/v0.4.6
+apiVersion: ossa/v0.4.7
 kind: Agent
 metadata:
   name: data-producer

package/examples/drupal/QUICKSTART.md

@@ -422,13 +422,27 @@ Response → Entity Storage → Views
 - **Issues**: https://gitlab.com/blueflyio/ossa/openstandardagents/-/issues
 - **OSSA Spec**: https://openstandardagents.org/
 
+## Example: AgentScope ReAct Assistant
+
+A complete Drupal module integrating [AgentScope](https://github.com/modelscope/agentscope) ReAct agents:
+
+```bash
+# Copy to Drupal
+cp -r examples/drupal/ai_agents_agentscope /path/to/drupal/web/modules/custom/
+drush en ai_agents_agentscope
+drush cr
+```
+
+See [`ai_agents_agentscope/README.md`](./ai_agents_agentscope/README.md) for details.
+
 ## Next Steps
 
 1. Read the [full documentation](../../docs/adapters/drupal-module-generator.md)
 2. Try the [content moderator example](./content-moderator.ossa.yaml)
-3. Create your own agent manifests
-4. Generate and deploy modules
-5. Share your agents with the community
+3. Try the [AgentScope example](./ai_agents_agentscope/)
+4. Create your own agent manifests
+5. Generate and deploy modules
+6. Share your agents with the community
 
 ---
 

package/examples/drupal/ai_agents_agentscope/README.md

@@ -0,0 +1,62 @@
+# AI Agents: AgentScope Provider
+
+Thin AiProvider bridge that connects Drupal's `drupal/ai` system to an [AgentScope](https://github.com/modelscope/agentscope) Python HTTP runtime.
+
+## What This Module Does
+
+Provides **one thing**: an `AiProvider` plugin (`agentscope`) that speaks AgentScope's REST API.
+
+Everything else is handled by contrib:
+- **Agent plugin derivation** -- `ai_agents_ossa` (reads OSSA manifests, creates AiAgent plugins)
+- **Tool registration** -- `drupal/ai` Tool API
+- **Manifest management** -- `ai_agents_ossa` config entities
+
+## Requirements
+
+- Drupal 10.4+ or 11.x
+- PHP 8.3+
+- `drupal/ai` module (provides AiProvider plugin system)
+- `ai_agents_ossa` module (OSSA manifest bridge)
+- AgentScope Python runtime running as HTTP service
+
+## Installation
+
+```bash
+drush en ai_agents_agentscope
+drush cr
+```
+
+## Configuration
+
+Settings at `ai_agents_agentscope.settings`:
+
+| Setting | Default | Description |
+|---------|---------|-------------|
+| `endpoint` | `http://127.0.0.1:12310` | AgentScope HTTP API URL |
+| `timeout` | `60` | Request timeout (seconds) |
+| `api_key` | `` | Optional API key for runtime auth |
+
+## AgentScope Runtime
+
+```bash
+pip install agentscope
+agentscope serve --port 12310
+```
+
+## Architecture
+
+```
+ai_agents_ossa (config entity + deriver + agent plugin)
+    |
+    v
+AgentScopeProvider (THIS MODULE -- AiProvider, HTTP transport only)
+    |
+    v
+AgentScope Python Runtime (POST /api/v1/agent/message)
+```
+
+## Links
+
+- [AgentScope](https://github.com/modelscope/agentscope)
+- [OSSA Specification](https://openstandardagents.org/)
+- [Drupal AI Module](https://www.drupal.org/project/ai)

package/examples/drupal/ai_agents_agentscope/ai_agents_agentscope.info.yml

@@ -0,0 +1,13 @@
+name: 'AI Agents: AgentScope Provider'
+type: module
+description: 'Thin AiProvider bridge to AgentScope Python HTTP runtime. Agent plugins are handled by ai_agents_ossa.'
+core_version_requirement: ^10.4 || ^11
+package: AI
+php: 8.3
+
+dependencies:
+  - drupal:ai
+  - ai_agents_ossa:ai_agents_ossa
+
+lifecycle: experimental
+configure: ai_agents_agentscope.settings

package/examples/drupal/ai_agents_agentscope/ai_agents_agentscope.services.yml

@@ -0,0 +1,4 @@
+services:
+  logger.channel.ai_agents_agentscope:
+    parent: logger.channel_base
+    arguments: ['ai_agents_agentscope']

package/examples/drupal/ai_agents_agentscope/config/install/ai_agents_agentscope.settings.yml

@@ -0,0 +1,3 @@
+endpoint: 'http://127.0.0.1:12310'
+timeout: 60
+api_key: ''

package/examples/drupal/ai_agents_agentscope/config/schema/ai_agents_agentscope.schema.yml

@@ -0,0 +1,14 @@
+ai_agents_agentscope.settings:
+  type: config_object
+  label: 'AI Agents AgentScope settings'
+  mapping:
+    endpoint:
+      type: string
+      label: 'AgentScope HTTP endpoint'
+      description: 'URL of the AgentScope Python runtime API.'
+    timeout:
+      type: integer
+      label: 'Request timeout in seconds'
+    api_key:
+      type: string
+      label: 'API key for AgentScope runtime (optional)'

package/examples/drupal/ai_agents_agentscope/src/AgentScopeRuntime.php

@@ -0,0 +1,11 @@
+<?php
+
+/**
+ * @file
+ * DELETED — This file is no longer needed.
+ *
+ * Runtime HTTP communication is now handled directly by
+ * AgentScopeProvider (src/Plugin/AiProvider/AgentScopeProvider.php)
+ * which extends AiProviderClientBase and uses its built-in httpClient.
+ * Remove this file.
+ */

package/examples/drupal/ai_agents_agentscope/src/Plugin/AiAgent/AgentScopeAgent.php

@@ -0,0 +1,10 @@
+<?php
+
+/**
+ * @file
+ * DELETED — This file is no longer needed.
+ *
+ * The ai_agents_ossa module provides OssaAgentPlugin which handles all agent
+ * plugin functionality. This module only needs to provide an AiProvider.
+ * Remove this file.
+ */