feat(skill): support more customization to code execution #709
Conversation
Summary of ChangesHello @fang-tech, 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 the customization and flexibility of skill management and code execution within the AgentScope framework. It introduces a powerful builder pattern for configuring code execution environments, allowing developers to precisely control tool availability, working directories, and resource handling. Furthermore, it provides comprehensive tools and documentation to streamline the creation and packaging of new skills, fostering a more robust and extensible agent ecosystem. Highlights
Changelog
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 significant improvements to the skill execution customization. The refactoring of SkillBox to use a CodeExecutionBuilder provides a much more flexible and fluent API for configuring code execution environments. The simplification of AgentSkill and the added customizability to AgentSkillPromptProvider are also welcome changes. The new example for creating skills is very comprehensive and does a great job of showcasing the new features. The test coverage for the new functionality is excellent. I have a couple of suggestions to improve code conciseness and adherence to best practices.
|
|
||
| Always be thorough in your analysis and explain your reasoning clearly. | ||
| """; | ||
| scanner.close(); |
There was a problem hiding this comment.
Closing a Scanner that wraps System.in also closes the underlying System.in stream. This is generally discouraged because it can prevent other parts of an application from reading from standard input. While it's acceptable in this simple example that exits after the chat, it's a good practice to avoid closing scanners that wrap System.in. The JVM will handle closing the stream on exit.
fang-tech
force-pushed
the
feat/skill/support-more-customization-to-codeExecution
branch
from
2f8d242 to
04aa83d
Compare
Codecov Report❌ Patch coverage is 📢 Thoughts on this report? Let us know! |
There was a problem hiding this comment.
Pull request overview
This PR expands the Skill system’s “code execution” support to be more customizable and introduces a new example “skill-creator” skill + tooling.
Changes:
- Reworked SkillBox code execution into a fluent builder (selectively enable shell/read/write, configure workDir/uploadDir, customize ShellCommandTool, and upload skill files).
- Added new skill-creator bundled resources (SKILL.md, references, packaging/validation/init scripts) and updated docs (EN/ZH) to describe new capabilities.
- Added/updated tests around ShellCommandTool configuration/getters and SkillBox code execution/upload behavior, plus updated the quickstart example.
Reviewed changes
Copilot reviewed 15 out of 15 changed files in this pull request and generated 11 comments.
Show a summary per file
| File | Description |
|---|---|
| docs/zh/task/agent-skill.md | Documents unified skill loading + tool lifecycle + code execution capabilities (ZH). |
| docs/en/task/agent-skill.md | Documents unified skill loading + tool lifecycle + code execution capabilities (EN). |
| agentscope-examples/quickstart/src/main/resources/skills/skill-creator/scripts/quick_validate.py | Adds a minimal skill validator script for SKILL.md frontmatter. |
| agentscope-examples/quickstart/src/main/resources/skills/skill-creator/scripts/package_skill.py | Adds a packager script to create distributable .skill archives. |
| agentscope-examples/quickstart/src/main/resources/skills/skill-creator/scripts/init_skill.py | Adds a skill initializer script to generate a template skill folder. |
| agentscope-examples/quickstart/src/main/resources/skills/skill-creator/references/workflows.md | Adds workflow-pattern guidance for skill authors. |
| agentscope-examples/quickstart/src/main/resources/skills/skill-creator/references/output-patterns.md | Adds output-format pattern guidance for skill authors. |
| agentscope-examples/quickstart/src/main/resources/skills/skill-creator/SKILL.md | Introduces the “skill-creator” skill content. |
| agentscope-examples/quickstart/src/main/resources/skills/skill-creator/LICENSE.txt | Adds license text for the bundled skill. |
| agentscope-examples/quickstart/src/main/java/io/agentscope/examples/quickstart/AgentSkillExample.java | Updates the example to load/use the skill-creator skill and enable code execution tools. |
| agentscope-core/src/test/java/io/agentscope/core/tool/coding/ShellCommandToolTest.java | Adds tests for new ShellCommandTool getters and baseDir behavior. |
| agentscope-core/src/test/java/io/agentscope/core/skill/SkillBoxTest.java | Updates/extends tests for new SkillBox code execution builder and upload behavior. |
| agentscope-core/src/test/java/io/agentscope/core/skill/AgentSkillTest.java | Updates tests to match new AgentSkill resource-path APIs. |
| agentscope-core/src/main/java/io/agentscope/core/tool/coding/ShellCommandTool.java | Adds getters needed to clone ShellCommandTool configuration. |
| agentscope-core/src/main/java/io/agentscope/core/skill/SkillFileFilter.java | Introduces a filter interface for deciding which skill resources get uploaded. |
| agentscope-core/src/main/java/io/agentscope/core/skill/SkillBox.java | Implements code execution builder, upload directory handling, filtering, base64 decoding, and auto-upload. |
| agentscope-core/src/main/java/io/agentscope/core/skill/AgentSkillPromptProvider.java | Allows custom prompt instruction/template and adds code execution guidance text. |
| agentscope-core/src/main/java/io/agentscope/core/skill/AgentSkill.java | Adds resource path/content accessors and removes script-only filtering. |
| agentscope-core/src/main/java/io/agentscope/core/ReActAgent.java | Updates agent skill integration to upload skill files when auto-upload is enabled. |
Comments suppressed due to low confidence (1)
agentscope-core/src/main/java/io/agentscope/core/skill/SkillBox.java:1151
- When code execution is enabled without an explicit workDir,
workDirStris null and ReadFileTool/WriteFileTool are constructed without a baseDir restriction. In the current ReadFileTool/WriteFileTool implementation, a null baseDir means no filesystem restriction, which defeats the sandbox and allows reading/writing arbitrary paths. Consider creating a temporary workDir duringenable()(not lazily) and always passing that directory to shell/read/write tools, or otherwise ensuring baseDir is never null for these tools.
String workDirStr = skillBox.workDir != null ? skillBox.workDir.toString() : null;
boolean shellEnabled = false;
// Shell Tool - check if withShell() was called
if (withShellCalled) {
ShellCommandTool shellTool;
if (customShellTool != null) {
// Clone custom tool with workDir override
shellTool = cloneShellToolWithWorkDir(customShellTool, workDirStr);
} else {
// Create default shell tool
shellTool =
new ShellCommandTool(
workDirStr,
Set.of("python", "python3", "node", "nodejs"),
null);
}
skillBox.toolkit
.registration()
.agentTool(shellTool)
.group("skill_code_execution_tool_group")
.apply();
shellEnabled = true;
}
// Read Tool
if (enableRead) {
ReadFileTool readTool = new ReadFileTool(workDirStr);
skillBox.toolkit
.registration()
.tool(readTool)
.group("skill_code_execution_tool_group")
.apply();
}
// Write Tool
if (enableWrite) {
WriteFileTool writeTool = new WriteFileTool(workDirStr);
skillBox.toolkit
.registration()
.tool(writeTool)
.group("skill_code_execution_tool_group")
.apply();
agentscope-core/src/main/java/io/agentscope/core/skill/AgentSkillPromptProvider.java
Show resolved
Hide resolved
agentscope-examples/quickstart/src/main/resources/skills/skill-creator/scripts/init_skill.py
Show resolved
Hide resolved
| public void uploadSkillFiles() { | ||
| Path targetDir = ensureUploadDirExists(); | ||
| SkillFileFilter filter = fileFilter != null ? fileFilter : SkillFileFilter.acceptAll(); | ||
| int fileCount = 0; |
There was a problem hiding this comment.
uploadSkillFiles() calls ensureUploadDirExists() up-front, which can create a temp workDir + register a shutdown hook even if there are no skills/resources (or everything is filtered out). To avoid unnecessary filesystem side effects, consider lazily creating the upload directory only when you’re about to write the first accepted resource.
agentscope-examples/quickstart/src/main/resources/skills/skill-creator/scripts/package_skill.py
Show resolved
Hide resolved
...tscope-examples/quickstart/src/main/resources/skills/skill-creator/scripts/quick_validate.py
Show resolved
Hide resolved
| private static final String SKILL_NAME = "skill-creator"; | ||
| private static final String RESOURCES_DIR = | ||
| "agentscope-examples/quickstart/src/main/resources/skills"; | ||
| private static final String OUTPUT_DIR = "agentscope-examples/quickstart/target/skill-output"; |
There was a problem hiding this comment.
This example hard-codes paths assuming the repo layout (agentscope-examples/quickstart/src/main/resources/...). That will break when running from a different working directory or from a packaged artifact. Consider resolving the skills directory via classpath (e.g., JarSkillRepositoryAdapter) or at least making the base path configurable via args/env.
| ShellCommandTool shellCommandTool = | ||
| new ShellCommandTool( | ||
| Set.of("python", "ls", "cat"), |
There was a problem hiding this comment.
The example shell whitelist includes Unix-only commands (ls, cat), so it will fail on Windows. If this example is intended to be cross-platform, consider using OS-conditional allowed commands (e.g., dir/type on Windows) or document that it’s Unix-only.
| ShellCommandTool shellCommandTool = | |
| new ShellCommandTool( | |
| Set.of("python", "ls", "cat"), | |
| String osName = System.getProperty("os.name").toLowerCase(); | |
| Set<String> allowedCommands = | |
| osName.contains("win") | |
| ? Set.of("python", "dir", "type") | |
| : Set.of("python", "ls", "cat"); | |
| ShellCommandTool shellCommandTool = | |
| new ShellCommandTool( | |
| allowedCommands, |
...tscope-examples/quickstart/src/main/resources/skills/skill-creator/scripts/quick_validate.py
Show resolved
Hide resolved
1 participant
AgentScope-Java Version
[The version of AgentScope-Java you are working on, e.g. 1.0.8, check your pom.xml dependency version or run
mvn dependency:tree | grep agentscope-parent:pom(only mac/linux)]Description
[Please describe the background, purpose, changes made, and how to test this PR]
Checklist
Please check the following items before code is ready to be reviewed.
mvn spotless:applymvn test)