feat: critic agent

.github/pull_request_template.md

@@ -69,5 +69,5 @@ Version format: MAJOR.MINOR.PATCH — each segment may be 2 digits (e.g. 10.11.2
 Python venv aligned with `manifest.yaml`, `pyproject.toml`, and `uv.lock` (or `requirements.txt` for legacy plugins).
 -->
 
-- [ ] Local deployment — Dify version: <!-- e.g. 1.2.0 -->
+- [ ] Local deployment — Dify version: 1.7.0
 - [ ] SaaS (cloud.dify.ai)

agent-strategies/self_refine_agent/.env.example

@@ -0,0 +1,4 @@
+INSTALL_METHOD=remote
+REMOTE_INSTALL_HOST=debug-plugin.dify.dev
+REMOTE_INSTALL_PORT=5003
+REMOTE_INSTALL_KEY=your-plugin-key-here

agent-strategies/self_refine_agent/README.md

@@ -0,0 +1,104 @@
+# Self-Refine Agent Strategy
+
+A Dify agent strategy plugin that improves output quality through iterative self-critique and refinement.
+
+## What It Does
+
+Self-Refine Agent Strategy implements the [Self-Refine](https://arxiv.org/abs/2303.17651) (NeurIPS 2023) algorithm, enabling agents to automatically evaluate and improve their own outputs through multiple refinement cycles.
+
+The strategy works by having the LLM act as its own critic:
+1. **Execute** - Agent generates an initial response
+2. **Evaluate** - LLM judges the output quality (0-100 score)
+3. **Critique** - LLM identifies specific issues and improvement opportunities
+4. **Refine** - Critique is injected into context for the next iteration
+5. **Repeat** - Process continues until quality threshold is met or max refinements reached
+
+This approach is particularly effective for tasks requiring high-quality outputs such as writing, analysis, code generation, and complex reasoning.
+
+## Key Features
+
+- **Automatic Quality Control** - LLM evaluates its own outputs without human intervention
+- **Iterative Improvement** - Each refinement cycle builds on previous critiques
+- **Transparent Process** - All intermediate steps (execution, evaluation, critique) are visible
+- **Configurable** - Control maximum iterations and refinement cycles
+
+## Usage
+
+### Parameters
+
+| Parameter | Type | Default | Description |
+|-----------|------|---------|-------------|
+| `model` | model-selector | - | LLM model to use |
+| `tools` | array[tools] | - | Available tools for the agent |
+| `instruction` | string | - | System instructions for the agent |
+| `query` | string | - | User query to process |
+| `context` | array[object] | - | Optional context information |
+| `maximum_iterations` | number | 5 | Max iterations per agent execution |
+| `max_refinements` | number | 2 | Max refinement cycles (0-5) |
+
+### Configuration Example
+
+When configuring the agent in Dify:
+
+1. Select **Self-Refine Agent** as your strategy
+2. Choose your preferred LLM model
+3. Set `max_refinements` to control quality vs. speed tradeoff:
+   - `0` - No refinement (standard agent behavior)
+   - `1-2` - Balanced (recommended for most cases)
+   - `3-5` - High quality (best for critical outputs)
+
+### Local Development
+
+To test the plugin locally:
+
+1. Create a `.env` file:
+```bash
+INSTALL_METHOD=remote
+DIFY_PLUGIN_DAEMON_URL=http://localhost:5003
+```
+
+2. Ensure Dify is running locally (via Docker Compose)
+
+3. Start the plugin:
+```bash
+python main.py
+```
+
+## How It Works
+
+The refinement loop follows this pattern:
+
+```
+Initial Query → Agent Execution → Output
+                        ↓
+                   Evaluation (score)
+                        ↓
+                   [Score ≥ 80?] → Yes → Return Final Output
+                        ↓ No
+                   Generate Critique
+                        ↓
+           Inject Critique into Context
+                        ↓
+              Re-execute Agent → Output
+                        ↓
+           [Max refinements reached?] → Yes → Return Best Output
+                        ↓ No
+                   Repeat Cycle
+```
+
+The agent automatically stops refining when:
+- Output quality score reaches 80 or above
+- Maximum refinement cycles are reached
+- No significant improvements are detected
+
+## Reference
+
+Based on the paper:
+
+Madaan, A., Tandon, N., Gupta, P., Hallinan, S., Gao, L., Wiegreffe, S., ... & Clark, P. (2023).
+**Self-Refine: Iterative Refinement with Self-Feedback.**
+NeurIPS 2023. [arXiv:2303.17651](https://arxiv.org/abs/2303.17651)
+
+## License
+
+MIT License

agent-strategies/self_refine_agent/main.py

@@ -0,0 +1,8 @@
+import sys
+
+from dify_plugin import DifyPluginEnv, Plugin
+
+plugin = Plugin(DifyPluginEnv(MAX_REQUEST_TIMEOUT=240, MAX_INVOCATION_TIMEOUT=900))
+
+if __name__ == "__main__":
+    plugin.run()

agent-strategies/self_refine_agent/manifest.yaml

@@ -0,0 +1,35 @@
+version: 0.0.1
+type: plugin
+author: "KERVIN-FARMER"
+name: "self_refine_agent"
+label:
+  en_US: "Self-Refine Agent Strategy"
+  zh_Hans: "自我优化 Agent 策略"
+created_at: "2025-06-18T00:00:00.000000000Z"
+icon: icon.svg
+description:
+  en_US: "Agent strategy with automatic self-critique and iterative refinement. Based on Self-Refine (NeurIPS 2023)."
+  zh_Hans: "支持自动自我批判和迭代优化的 Agent 策略。基于 Self-Refine (NeurIPS 2023)。"
+tags:
+  - "agent"
+resource:
+  memory: 1048576
+  permission:
+    tool:
+      enabled: true
+    model:
+      enabled: true
+      llm: true
+plugins:
+  agent_strategies:
+    - "provider/self_refine.yaml"
+meta:
+  version: 0.0.1
+  minimum_dify_version: "1.7.0"
+  arch:
+    - "amd64"
+    - "arm64"
+  runner:
+    language: "python"
+    version: "3.12"
+    entrypoint: "main"

agent-strategies/self_refine_agent/prompt/templates.py

@@ -0,0 +1,65 @@
+"""
+Prompt templates for Self-Refine Agent Strategy
+
+Based on Self-Refine: Iterative Refinement with Self-Feedback (NeurIPS 2023)
+https://arxiv.org/abs/2303.17651
+"""
+
+# System prompt for initial execution
+EXECUTION_SYSTEM_PROMPT = """You are a helpful AI assistant. Follow the user's instructions carefully and provide detailed, accurate responses.
+
+{{instruction}}
+
+Available tools:
+{{tools}}
+
+Use the tools when necessary to gather information and complete the task. Provide thorough and well-reasoned answers."""
+
+# System prompt for execution with refinement context
+REFINEMENT_EXECUTION_PROMPT = """You are a helpful AI assistant. Follow the user's instructions carefully and provide detailed, accurate responses.
+
+{{instruction}}
+
+Available tools:
+{{tools}}
+
+Use the tools when necessary to gather information and complete the task.
+
+[SELF-REFINEMENT CONTEXT]
+Your previous attempt had the following issues:
+{{critique}}
+
+Please address these issues in this attempt and provide an improved response."""
+
+# Evaluation prompt for assessing output quality
+EVALUATION_PROMPT = """You are an expert evaluator. Assess the quality of the following agent output for the given query.
+
+Query: {{query}}
+
+Agent Output:
+{{output}}
+
+Evaluate the output based on these criteria:
+1. Accuracy: Is the information correct and factual?
+2. Completeness: Does it fully address the query?
+3. Clarity: Is it clear and well-structured?
+4. Relevance: Does it stay focused on the query?
+
+Respond in the following JSON format:
+{
+  "is_satisfactory": true/false,
+  "issues": "Specific description of issues if not satisfactory, or empty string if satisfactory",
+  "score": 0-100
+}
+
+Only set is_satisfactory to false if there are significant issues that require correction."""
+
+# Fallback critique when evaluation fails
+FALLBACK_CRITIQUE = "Unable to evaluate output quality. Please review and improve the response for accuracy, completeness, and clarity."
+
+SELF_REFINE_TEMPLATES = {
+    "execution_system": EXECUTION_SYSTEM_PROMPT,
+    "refinement_execution": REFINEMENT_EXECUTION_PROMPT,
+    "evaluation": EVALUATION_PROMPT,
+    "fallback_critique": FALLBACK_CRITIQUE
+}

agent-strategies/self_refine_agent/provider/self_refine.py

@@ -0,0 +1,5 @@
+from dify_plugin.interfaces.agent import AgentProvider
+
+
+class SelfRefineAgentProvider(AgentProvider):
+    pass

agent-strategies/self_refine_agent/provider/self_refine.yaml

@@ -0,0 +1,15 @@
+identity:
+  author: KERVIN-FARMER
+  name: self_refine
+  label:
+    en_US: Self-Refine Agent
+    zh_Hans: 自我优化 Agent
+  description:
+    en_US: Self-refining agent with automatic critique and iterative improvement
+    zh_Hans: 支持自动批判和迭代改进的自我优化 Agent
+  icon: icon.svg
+strategies:
+  - strategies/self_refine.yaml
+extra:
+  python:
+    source: provider/self_refine.py

agent-strategies/self_refine_agent/pyproject.toml

@@ -0,0 +1,13 @@
+[project]
+name = "self_refine_agent"
+version = "0.0.1"
+description = "Self-Refine Agent Strategy for Dify - Automatic critique and iterative refinement"
+readme = "README.md"
+requires-python = ">=3.12"
+
+dependencies = [
+    "dify_plugin>=0.9.0",
+]
+
+[dependency-groups]
+dev = []