feat: Add Dify Plugin Integration #158
Conversation
|
[APPROVALNOTIFIER] This PR is NOT APPROVED This pull-request has been approved by: The full list of commands accepted by this bot can be found here. DetailsNeeds approval from an approver in each of these files:Approvers can indicate their approval by writing |
|
Welcome @ifelseend! It looks like this is your first PR to volcano-sh/agentcube 🎉 |
Summary of ChangesHello @ifelseend, I'm Gemini Code Assist1! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed! This pull request significantly enhances Dify's capabilities by integrating AgentCube, an open-source project designed to support and manage AI Agent workloads. The new plugin provides Dify users with a secure and isolated environment to execute code and terminal commands, which is crucial for building advanced agents that require computational tasks. It ensures that users maintain full control over their data by being designed for self-hosted AgentCube instances, and comes with comprehensive documentation and an automated publishing workflow to streamline its adoption and maintenance. Highlights
🧠 New Feature in Public Preview: You can now enable Memory to help Gemini Code Assist learn from your team's feedback. This makes future code reviews more consistent and personalized to your project's style. Click here to enable Memory in your admin console. Using Gemini Code AssistThe full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips. Invoking Gemini You can request assistance from Gemini at any point by creating a comment using either
Customization To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a Limitations & Feedback Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here. You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension. Footnotes
|
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request introduces a Dify plugin for AgentCube, which is a valuable addition. The implementation is a good starting point, but I've identified several critical issues that need to be addressed before this can be merged. These include a thread-safety problem in the core tool logic, incorrect session handling that breaks the reuse feature, and a bug in the automated publishing workflow that will cause it to fail. Additionally, there are several high-severity configuration mistakes in the plugin's YAML files that need correction. My detailed comments below provide specific suggestions to resolve these issues.
|
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## main #158 +/- ##
=======================================
Coverage ? 34.34%
=======================================
Files ? 29
Lines ? 2533
Branches ? 0
=======================================
Hits ? 870
Misses ? 1540
Partials ? 123
Flags with carried forward coverage won't be shown. Click here to find out more. ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
Signed-off-by: ifelseend <1003324934@qq.com>
ifelseend
force-pushed
the
dify-plugin
branch
from
77cc687 to
aadd9ef
Compare
hzxuzhonghu
left a comment
hzxuzhonghu
left a comment
There was a problem hiding this comment.
seems it to sync the plugins to https://github.com/langgenius/dify-official-plugins/
My confusion: is that possible to just maintain the source there, keep it here looks not necessary
| on: | ||
| push: | ||
| tags: | ||
| - 'dify-plugin/v*' # Trigger only on plugin-specific tags |
There was a problem hiding this comment.
tags:
- "v*.*.*"
- "v*.*.*-*" # Support for pre-release tags like v1.2.3-alpha
There was a problem hiding this comment.
Pull request overview
This PR introduces a Dify plugin that integrates AgentCube's code execution capabilities into the Dify platform. The plugin enables users to execute Python code and shell commands in secure sandboxed environments managed by AgentCube.
Changes:
- Added a complete Dify plugin implementation in
integrations/dify-plugin/directory with tool provider, manifest, and documentation - Implemented the AgentCube Code Interpreter tool that wraps the agentcube-sdk for executing code and commands
- Added GitHub Actions workflow for automated plugin packaging and publishing to the Dify plugin marketplace
Reviewed changes
Copilot reviewed 14 out of 16 changed files in this pull request and generated 14 comments.
Show a summary per file
| File | Description |
|---|---|
integrations/dify-plugin/tools/agentcube-code-interpreter.yaml |
Tool definition with parameters for language selection, code/command execution, and session management |
integrations/dify-plugin/tools/agentcube-code-interpreter.py |
Python implementation of the code interpreter tool using the agentcube-sdk |
integrations/dify-plugin/provider/agentcube.yaml |
Provider configuration defining the AgentCube tool provider for Dify |
integrations/dify-plugin/provider/agentcube.py |
Provider class with credential validation stub (currently empty) |
integrations/dify-plugin/manifest.yaml |
Plugin manifest defining metadata, permissions, and runtime configuration |
integrations/dify-plugin/main.py |
Entry point that initializes the Dify plugin with 120s timeout |
integrations/dify-plugin/requirements.txt |
Python dependencies including dify-plugin and agentcube-sdk |
integrations/dify-plugin/README.md |
User-facing documentation explaining plugin capabilities and configuration |
integrations/dify-plugin/PRIVACY.md |
Privacy policy clarifying self-hosted deployment and data ownership |
integrations/dify-plugin/GUIDE.md |
Developer guide for Dify plugin development (general reference) |
integrations/dify-plugin/.gitignore |
Python-specific gitignore patterns |
integrations/dify-plugin/.difyignore |
Patterns for excluding files from plugin packaging |
integrations/dify-plugin/.env.example |
Example environment variables for plugin debugging |
integrations/dify-plugin/_assets/icon.png |
Plugin icon for light theme |
integrations/dify-plugin/_assets/icon-dark.png |
Plugin icon for dark theme |
.github/workflows/dify-plugin-publish.yml |
Automated workflow for packaging and publishing plugin on tag push |
| try: | ||
| """ | ||
| IMPLEMENT YOUR VALIDATION HERE | ||
| """ | ||
| except Exception as e: | ||
| raise ToolProviderCredentialValidationError(str(e)) | ||
|
|
There was a problem hiding this comment.
The _validate_credentials method is not implemented (it only contains a pass statement inside the try block). Since this plugin doesn't require credentials (self-hosted configuration via tool parameters), either implement a proper validation or add a comment explaining why validation is not needed. Following Dify plugin conventions, if no validation is required, the method should still indicate this clearly.
| try: | |
| """ | |
| IMPLEMENT YOUR VALIDATION HERE | |
| """ | |
| except Exception as e: | |
| raise ToolProviderCredentialValidationError(str(e)) | |
| """ | |
| This provider does not use external credentials. | |
| AgentCube is configured via self-hosted parameters in the tool settings, | |
| so there is nothing to validate here. Following Dify plugin conventions, | |
| we still implement this method to make the expectation explicit: | |
| - if credentials are empty or None, accept them | |
| - if any credentials are provided, treat that as a configuration error | |
| """ | |
| if not credentials: | |
| # No credentials expected or required for this provider. | |
| return | |
| # Any supplied credentials indicate a misconfiguration. | |
| raise ToolProviderCredentialValidationError( | |
| "agentcube-code-interpreter does not require credentials; " | |
| "please leave the credential configuration empty" | |
| ) |
| error_msg = str(e) | ||
| finally: | ||
| if ci_client and not session_reuse: | ||
| ci_client.stop() |
There was a problem hiding this comment.
When an error occurs and ci_client has been created but session_reuse is False, the finally block will call ci_client.stop(). However, if the error occurred during session creation or initialization, the session might not have been fully created, potentially causing additional errors in the stop() method. Consider adding a check to ensure the client is in a valid state before calling stop(), or wrap the stop() call in a try-except block.
| ci_client.stop() | |
| try: | |
| ci_client.stop() | |
| except Exception: | |
| # ignore errors during cleanup to avoid masking original exceptions | |
| pass |
| "session_id": ci_client.session_id, | ||
| "code_interpreter_id": code_interpreter_id, |
There was a problem hiding this comment.
The success result includes code_interpreter_id from the input parameter rather than from the actual client. This is inconsistent because the client might be using a different name (either default or from the name parameter). Consider using ci_client.name to accurately reflect which code interpreter was actually used during execution.
| "session_id": ci_client.session_id, | |
| "code_interpreter_id": code_interpreter_id, | |
| "session_id": ci_client.session_id, | |
| "code_interpreter_id": getattr(ci_client, "name", code_interpreter_id), |
| client_kwargs = { | ||
| "router_url": router_url, | ||
| "workload_manager_url": workload_manager_url | ||
| } | ||
| if code_interpreter_id: | ||
| client_kwargs["name"] = code_interpreter_id | ||
| if session_id: | ||
| client_kwargs["session_id"] = session_id | ||
|
|
||
| ci_client = CodeInterpreterClient(**client_kwargs) |
There was a problem hiding this comment.
According to the CodeInterpreterClient SDK (see sdk-python/agentcube/code_interpreter.py:51-61), both workload_manager_url and router_url are optional parameters with None as default. However, the SDK will raise a ValueError if router_url is not provided. The tool should validate that these required URLs are provided before attempting to create the client, providing a clearer error message to users.
| version: "3.12" | ||
| entrypoint: main | ||
| minimum_dify_version: null | ||
| created_at: 2025-12-22T10:54:47.846104+08:00 |
There was a problem hiding this comment.
The created_at timestamp is set to 2025-12-22, which is in the past relative to the PRIVACY.md date (2026-01-22). This inconsistency should be resolved. Additionally, verify if the December date is correct or if it should be updated to reflect the actual creation/last update time.
| created_at: 2025-12-22T10:54:47.846104+08:00 | |
| created_at: 2026-01-22T10:54:47.846104+08:00 |
| - name: language | ||
| type: select | ||
| required: true | ||
| options: | ||
| - value: python | ||
| label: | ||
| en_US: python | ||
| - value: javascript | ||
| label: | ||
| en_US: javascript | ||
| - value: typescript | ||
| label: | ||
| en_US: typescript | ||
| label: | ||
| en_US: language | ||
| human_description: | ||
| en_US: programming language of the code (required if code is provided) | ||
| llm_description: The programming language of the code. Agentcube supports python by default. If you use javascript or typescript, you must use a custom code interpreter and the code_interpreter_id parameter is required. | ||
| form: llm |
There was a problem hiding this comment.
The language parameter is marked as required (line 39), but the logic in tools/agentcube-code-interpreter.py:49 only uses language when both language and code are provided. If a user only wants to execute a command without code, they would still be forced to provide a language parameter unnecessarily. Consider making the language parameter optional and only requiring it when code is provided.
| ci_client.stop() | ||
|
|
||
| if error_msg: | ||
| result = {"status": "error", "reason": error_msg} |
There was a problem hiding this comment.
When an error occurs and the error result is returned (line 62), the result only contains "status" and "reason" fields. However, if a session was partially created before the error, the session_id would still be valuable information for debugging or cleanup. Consider including the session_id in the error response when available, similar to how it's included in the success response.
| result = {"status": "error", "reason": error_msg} | |
| result = {"status": "error", "reason": error_msg} | |
| if ci_client and getattr(ci_client, "session_id", None): | |
| result["session_id"] = ci_client.session_id |
| PLUGIN_NAME=$(grep "^name:" manifest.yaml | cut -d' ' -f2) | ||
| echo "Plugin name: $PLUGIN_NAME" | ||
| echo "plugin_name=$PLUGIN_NAME" >> $GITHUB_OUTPUT | ||
|
|
||
| VERSION=$(grep "^version:" manifest.yaml | cut -d' ' -f2) | ||
| echo "Plugin version: $VERSION" | ||
| echo "version=$VERSION" >> $GITHUB_OUTPUT | ||
|
|
||
| AUTHOR=$(grep "^author:" manifest.yaml | cut -d' ' -f2) | ||
| echo "Plugin author: $AUTHOR" | ||
| echo "author=$AUTHOR" >> $GITHUB_OUTPUT |
There was a problem hiding this comment.
The parsing of the manifest.yaml file uses basic grep and cut commands which may be fragile. For example, if there's any extra whitespace or formatting variation in the YAML file, the parsing could fail. Consider using a proper YAML parser like yq for more robust parsing, or add validation to ensure the parsed values are not empty before proceeding.
| PLUGIN_NAME=$(grep "^name:" manifest.yaml | cut -d' ' -f2) | |
| echo "Plugin name: $PLUGIN_NAME" | |
| echo "plugin_name=$PLUGIN_NAME" >> $GITHUB_OUTPUT | |
| VERSION=$(grep "^version:" manifest.yaml | cut -d' ' -f2) | |
| echo "Plugin version: $VERSION" | |
| echo "version=$VERSION" >> $GITHUB_OUTPUT | |
| AUTHOR=$(grep "^author:" manifest.yaml | cut -d' ' -f2) | |
| echo "Plugin author: $AUTHOR" | |
| echo "author=$AUTHOR" >> $GITHUB_OUTPUT | |
| PLUGIN_NAME=$(awk -F': *' '$1=="name"{print $2; exit}' manifest.yaml | tr -d '"' | xargs) | |
| if [ -z "$PLUGIN_NAME" ]; then | |
| echo "error: plugin name not found in manifest.yaml" >&2 | |
| exit 1 | |
| fi | |
| echo "Plugin name: $PLUGIN_NAME" | |
| echo "plugin_name=$PLUGIN_NAME" >>"$GITHUB_OUTPUT" | |
| VERSION=$(awk -F': *' '$1=="version"{print $2; exit}' manifest.yaml | tr -d '"' | xargs) | |
| if [ -z "$VERSION" ]; then | |
| echo "error: plugin version not found in manifest.yaml" >&2 | |
| exit 1 | |
| fi | |
| echo "Plugin version: $VERSION" | |
| echo "version=$VERSION" >>"$GITHUB_OUTPUT" | |
| AUTHOR=$(awk -F': *' '$1=="author"{print $2; exit}' manifest.yaml | tr -d '"' | xargs) | |
| if [ -z "$AUTHOR" ]; then | |
| echo "error: plugin author not found in manifest.yaml" >&2 | |
| exit 1 | |
| fi | |
| echo "Plugin author: $AUTHOR" | |
| echo "author=$AUTHOR" >>"$GITHUB_OUTPUT" |
| # See the License for the specific language governing permissions and | ||
| # limitations under the License. | ||
|
|
||
| import os |
There was a problem hiding this comment.
Import of 'os' is not used.
| import os |
| try: | ||
| """ | ||
| IMPLEMENT YOUR VALIDATION HERE | ||
| """ | ||
| except Exception as e: | ||
| raise ToolProviderCredentialValidationError(str(e)) |
There was a problem hiding this comment.
This statement is unreachable.
| try: | |
| """ | |
| IMPLEMENT YOUR VALIDATION HERE | |
| """ | |
| except Exception as e: | |
| raise ToolProviderCredentialValidationError(str(e)) | |
| """ | |
| IMPLEMENT YOUR VALIDATION HERE | |
| """ | |
| pass |
The Dify plugin packages are maintained at https://github.com/langgenius/dify-plugins, but the source code for each plugin is maintained by its respective owner. |
There was a problem hiding this comment.
I donot think we need this, because we have added in the root dir of this repo
| label: # Required: Multi-language names | ||
| en_US: Plugin Name | ||
| zh_Hans: 插件名称 | ||
| created_at: 2023-01-01T00:00:00Z # Required: Creation time (RFC3339) |
| <details> | ||
| <summary><b>3. Testing & Debugging</b></summary> | ||
| 1. Copy `.env.example` to `.env` and configure: |
There was a problem hiding this comment.
configure what? where is REMOTE_INSTALL_KEY from
| - name: Checkout target repo | ||
| uses: actions/checkout@v3 |
There was a problem hiding this comment.
The workflow uses deprecated actions/checkout@v3. As of 2026, actions/checkout@v4 is available and recommended. Consider updating to v4 for better performance and security updates, unless there's a specific reason to use v3.
| except Exception as e: | ||
| error_msg = str(e) |
There was a problem hiding this comment.
The error handling in the execute method catches all exceptions but doesn't log them or provide meaningful context. When an exception occurs, the error message is captured but there's no logging mechanism to help with debugging. Consider using a logger to record exceptions before returning the error result. This aligns with the Kubernetes coding convention to use proper logging (klog/logr) instead of just capturing error strings.
| "status": "success", | ||
| "session_id": ci_client.session_id, | ||
| "code_interpreter_id": code_interpreter_id, | ||
| "results": results | ||
| } | ||
|
|
There was a problem hiding this comment.
When an error occurs during execution, the function returns code_interpreter_id in the result dictionary even though it may be None (line 67). This could be misleading if the client was not successfully created. Consider only including code_interpreter_id in the success case, or handling the None case explicitly.
| "status": "success", | |
| "session_id": ci_client.session_id, | |
| "code_interpreter_id": code_interpreter_id, | |
| "results": results | |
| } | |
| "status": "success", | |
| "session_id": ci_client.session_id, | |
| "results": results, | |
| } | |
| if code_interpreter_id is not None: | |
| result["code_interpreter_id"] = code_interpreter_id |
| if command: | ||
| command_result = ci_client.execute_command(command) | ||
| results.append({"type": "command", "result": command_result}) | ||
|
|
||
| if language and code: | ||
| code_result = ci_client.run_code(language, code) | ||
| results.append({"type": "code", "result": code_result}) | ||
|
|
||
| if not command and not code: | ||
| raise ValueError("Either command or code must be provided") |
There was a problem hiding this comment.
The validation check on line 53-54 only triggers if both command and code are missing, but the code on line 49 requires both language AND code to be present for code execution. If language is provided but code is not (or vice versa), the tool will silently skip code execution without raising an error. This could lead to confusing behavior where a user provides a language but forgets the code, and the tool succeeds without executing anything. Consider adding validation to ensure that if language is provided, code must also be provided.
| if command: | |
| command_result = ci_client.execute_command(command) | |
| results.append({"type": "command", "result": command_result}) | |
| if language and code: | |
| code_result = ci_client.run_code(language, code) | |
| results.append({"type": "code", "result": code_result}) | |
| if not command and not code: | |
| raise ValueError("Either command or code must be provided") | |
| if (language and not code) or (code and not language): | |
| raise ValueError("both language and code must be provided together") | |
| if command: | |
| command_result = ci_client.execute_command(command) | |
| results.append({"type": "command", "result": command_result}) | |
| if language and code: | |
| code_result = ci_client.run_code(language, code) | |
| results.append({"type": "code", "result": code_result}) | |
| if not command and not language and not code: | |
| raise ValueError("either command or code must be provided") |
| @@ -0,0 +1,2 @@ | |||
| dify-plugin>=0.4.0,<0.7.0 | |||
There was a problem hiding this comment.
The dify-plugin dependency allows a wide version range (0.4.0 to <0.7.0), spanning multiple major feature releases. This broad range might introduce compatibility issues if breaking changes occur in minor versions. Additionally, the provider code comments reference SDK version 0.4.2 features (line 33 in agentcube.py), but the minimum is 0.4.0. Consider tightening the version range to ensure compatibility, or add comments documenting which versions have been tested.
| dify-plugin>=0.4.0,<0.7.0 | |
| dify-plugin>=0.4.2,<0.5.0 |
| - name: Checkout code | ||
| uses: actions/checkout@v3 |
There was a problem hiding this comment.
The workflow uses deprecated actions/checkout@v3. As of 2026, actions/checkout@v4 is available and recommended. Consider updating to v4 for better performance and security updates, unless there's a specific reason to use v3.
| - name: code | ||
| type: string | ||
| required: false |
There was a problem hiding this comment.
The 'code' parameter is marked as required: false in the YAML, but the README.md on line 28 indicates it is "Yes" (required). There's an inconsistency between the tool definition and the documentation about whether code is required. Based on the Python implementation (line 53), either command or code can be provided, making code conditionally required. The documentation should be updated to reflect this OR the YAML should mark it as required with appropriate validation in the code.
| form: llm | ||
| - name: language | ||
| type: select | ||
| required: true |
There was a problem hiding this comment.
The 'language' parameter is marked as required: true, but according to the Python implementation (line 49), code execution only happens when both language AND code are provided. If only command execution is needed, language is not actually required. This creates a situation where users must provide a language even when only executing commands. Consider making language required: false, or update the implementation to require language in all cases.
| required: true | |
| required: false |
| run: | | ||
| mkdir -p $RUNNER_TEMP/bin | ||
| cd $RUNNER_TEMP/bin | ||
| wget https://github.com/langgenius/dify-plugin-daemon/releases/download/0.0.6/dify-plugin-linux-amd64 |
There was a problem hiding this comment.
The workflow downloads a specific version (0.0.6) of the dify-plugin CLI tool. This hard-coded version could become outdated. Consider either: (1) documenting why this specific version is required, (2) using a version variable that can be easily updated, or (3) checking for and using the latest release. This will make maintenance easier when newer versions of the CLI tool are available.
| try: | ||
| """ | ||
| IMPLEMENT YOUR VALIDATION HERE | ||
| """ | ||
| except Exception as e: | ||
| raise ToolProviderCredentialValidationError(str(e)) | ||
|
|
There was a problem hiding this comment.
This statement is unreachable.
| try: | |
| """ | |
| IMPLEMENT YOUR VALIDATION HERE | |
| """ | |
| except Exception as e: | |
| raise ToolProviderCredentialValidationError(str(e)) | |
| """ | |
| Validate the provided credentials for the Agentcube code interpreter. | |
| Raises: | |
| ToolProviderCredentialValidationError: If the credentials are missing or invalid. | |
| """ | |
| if not credentials: | |
| raise ToolProviderCredentialValidationError("credentials are required") |
4 participants
Description
This PR introduces the AgentCube Dify Plugin, enabling users to integrate AgentCube's powerful code execution capabilities directly into Dify applications and workflows.
With this plugin, Dify users can leverage AgentCube's secure sandboxes to execute code and terminal commands, making it easier to build advanced agents that require computational capabilities.
Key Features
code_interpreter_idconfigurations.session_id, allowing variables and files to persist across multiple execution steps.router_urlandworkload_manager_url.Changes
intergrations/dify-plugindirectory containing the full plugin source.manifest.yamlandprovider/agentcube.yamlfor Dify plugin definition.tools/agentcube-code-interpreter.yamlandtools/agentcube-code-interpreter.pyimplementing the tool logic usingagentcube-sdk.README.md: Usage guide and configuration details.PRIVACY.md: Privacy policy clarifying data handling for self-hosted deployments.How to Test
intergrationsdirectory.