Agent skill
smalltalk-developer
Comprehensive Pharo Smalltalk development workflow guide with AI-driven Tonel editing. Provides expertise in Tonel file format syntax (class definitions with name, superclass, instVars, category, method categories, class comment placement), package structure (package.st placement, directory organization, BaselineOf dependencies), development workflow (Edit → Import → Test cycle with absolute paths, re-import timing, test execution), and Pharo best practices (CRC format documentation, method categorization conventions). Use when working with Pharo Smalltalk projects, creating or editing Tonel .st files, organizing packages and dependencies, resolving import order issues, writing class comments, implementing standard Pharo development patterns (Singleton, Settings, etc.), or troubleshooting Tonel syntax.
Install this agent skill to your Project
npx add-skill https://github.com/mumez/smalltalk-dev-plugin/tree/develop/skills/smalltalk-developer
SKILL.md
Smalltalk Developer Workflow
Implement the standard workflow for Pharo Smalltalk development using AI editors and the Tonel file format.
Core Development Cycle
The fundamental workflow for Smalltalk development consists of three steps that repeat:
1. Edit Tonel Files
Edit Tonel files directly in the AI editor. The AI editor is the single source of truth for code.
Key principle: All code changes happen in .st files, not in the Pharo image.
Tonel basics:
- One class per
.stfile - File name matches class name
- Each package directory contains
package.st
For detailed Tonel syntax and examples, see Tonel Format Reference.
Follow the Style Guide for idiomatic Smalltalk.
Refer to Implementation Patterns for common patterns (Singleton, Settings, etc.).
2. Import to Pharo
Before importing, run lint to check code quality:
/st-lint src/MyPackage # Check Smalltalk best practices
Then import the package into the running Pharo image using absolute paths:
mcp__smalltalk-interop__import_package: 'MyPackage' path: '/home/user/project/src'
Critical rules:
- ✅ Lint before import - Check code quality first
- ✅ Always use absolute paths (never relative)
- ✅ Re-import after every change
- ✅ Import packages in dependency order
- ✅ Import test packages after main packages
3. Run Tests
After import, run tests to verify changes:
mcp__smalltalk-interop__run_class_test: 'MyClassTest'
mcp__smalltalk-interop__run_package_test: 'MyPackage-Tests'
If tests fail, return to step 1, fix the Tonel file, and repeat.
Essential Best Practices
Path Management
Always use absolute paths for imports:
✅ /home/user/myproject/src
❌ ./src
❌ ../myproject/src
Finding the source directory:
- Check
.projectfile forsrcDirectoryfield - Common locations:
src/,repositories/ - Construct full path:
<project_root>/<srcDirectory>
See Best Practices Reference for details.
Package Dependencies
Check BaselineOf<ProjectName> to determine correct import order:
baseline: spec
<baseline>
spec for: #common do: [
spec
package: 'MyPackage-Core';
package: 'MyPackage-Json' with: [
spec requires: #('MyPackage-Core')
]
]
Import order: Dependencies first → MyPackage-Core → MyPackage-Json
See Best Practices: Dependencies for complete guide.
Import Timing
Re-import after every change - Pharo doesn't automatically reload files.
Standard sequence:
- Edit
MyPackage/MyClass.st - Import
MyPackage - Edit
MyPackage-Tests/MyClassTest.st - Import
MyPackage-Tests - Run tests
File Editing Philosophy
The AI editor is the source of truth:
- ✅ Edit
.stfiles → Import to Pharo - ❌ Edit in Pharo → Export to
.stfiles
Use export_package only for:
- Initial project setup
- Emergency recovery
- Exploring existing code
See Best Practices: File Editing for rationale.
Common Patterns
Pattern 1: Creating New Class
1. Ensure .project file exists in project root (create if missing - see Best Practices)
2. Create src/MyPackage/MyClass.st with class definition
3. import_package: 'MyPackage' path: '/absolute/path/src'
4. run_class_test: 'MyClassTest'
Pattern 2: Adding Methods
1. Read existing src/MyPackage/MyClass.st
2. Add new method to file
3. import_package: 'MyPackage' path: '/absolute/path/src'
4. run_class_test: 'MyClassTest'
Pattern 3: Multi-Package Development
1. Check BaselineOf for dependency order
2. Import packages in correct sequence
3. Import test packages last
4. Run comprehensive tests
Pattern 4: Installing External Packages from GitHub
When a required package is not present locally, use install_project — never use Metacello directly via eval:
mcp__smalltalk-interop__install_project
project_name: 'MyLib'
repository_url: 'github://owner/MyLib:branch/src'
Parameters:
project_name(required): the baseline/project namerepository_url(required): Metacello-style URL (e.g.github://owner/repo:branch/src)load_groups(optional): comma-separated groups to load
Critical rule: If a class or package is missing and is not in the local source tree, always reach for install_project first. Do NOT try to load it with Metacello new baseline:... load via eval — that approach bypasses the managed workflow and can corrupt the image state.
For complete examples, see Development Session Examples.
Automation
When this skill is active, automatically suggest:
- Import commands after editing Tonel files
- Test commands after successful import
- Debugging procedures when tests fail
Quick Reference
MCP Tools
Install external package (GitHub/remote):
mcp__smalltalk-interop__install_project
project_name: 'MyLib'
repository_url: 'github://owner/MyLib:branch/src'
Use this when a package is not present locally. Never use Metacello via eval for this purpose.
Import local package:
mcp__smalltalk-interop__import_package: 'PackageName' path: '/absolute/path'
Test:
mcp__smalltalk-interop__run_class_test: 'TestClassName'
mcp__smalltalk-interop__run_package_test: 'PackageName-Tests'
Debug:
mcp__smalltalk-interop__eval: 'Smalltalk code here'
Validate (optional):
mcp__smalltalk-validator__validate_tonel_smalltalk_from_file: '/path/to/file.st'
Common Commands
/st-import PackageName /path- Import package/st-test TestClass- Run tests/st-eval code- Execute Smalltalk snippet/st-validate file.st- Validate syntax
Troubleshooting
Import Fails
"Package not found":
- Verify absolute path is correct (in Docker, use container-side path: typically
/root/repos— checkcompose.ymlfor volume mounts) - Check
package.stexists - Ensure package name matches directory
"Syntax error":
- Run
validate_tonel_smalltalk_from_filefirst - Check Tonel syntax (brackets, quotes, periods)
"Dependency not found" / missing external package:
- If the package is not in the local source tree, install it from GitHub:
mcp__smalltalk-interop__install_project project_name: 'MyLib' repository_url: 'github://owner/MyLib:branch/src' - Do NOT try
Metacello new baseline: ... loadviaeval— useinstall_projectinstead - If it is a local package, check Baseline for required packages and import dependencies first
Tests Fail
- Read error message carefully
- Use
/st-evalto debug incrementally - Fix in Tonel file (not Pharo)
- Re-import and re-test
See Best Practices: Error Handling for complete guide.
Complete Documentation
This skill provides focused guidance for the core workflow. For comprehensive information:
- Tonel Format Reference - Complete Tonel syntax guide
- Style Guide - Idiomatic Smalltalk patterns and concise coding practices
- Best Practices - Detailed practices and patterns
- Development Examples - Real-world session workflows
- Implementation Patterns - Common patterns (Singleton, etc.)
Summary Workflow
Edit .st file
↓
Lint code (/st-lint)
↓
Import package (absolute path)
↓
Run tests
↓
Tests pass? → Done
↓
Tests fail? → Debug with /st-eval → Fix .st file → Re-import
Remember: The cycle is Edit → Lint → Import → Test. Never skip lint, import, or tests.
Recommended Agent Skills
Expand your agent's capabilities with these related and highly-rated skills.
smalltalk-implementation-finder
Method implementation finder and analyzer for Pharo Smalltalk. Provides expertise in discovering implementors across class hierarchies (search_implementors), analyzing implementation patterns, learning coding idioms from existing implementations, assessing refactoring impact (counting implementors and references), finding duplicate code for consolidation, understanding abstract method implementations (subclassResponsibility), and tracing method overrides through inheritance chains. Use when analyzing method implementations across classes, learning implementation idioms, assessing refactoring risk before changes, finding duplicate implementations for consolidation, understanding how abstract methods are implemented in concrete classes, or tracing which classes override specific methods.
smalltalk-usage-finder
Class and method usage analyzer for Pharo Smalltalk. Provides expertise in understanding class responsibilities through class comments (get_class_comment), discovering usage patterns via references (search_references_to_class), finding example methods (exampleXXX patterns), analyzing method usage in context (search_references with polymorphism handling), generating package overviews (list_classes with comment analysis), and resolving ambiguous names (search_classes_like, search_methods_like). Use when understanding what a class does, finding usage examples for classes or methods, discovering how to use an API, analyzing package structure and purpose, resolving unclear class or method names, or learning usage patterns from real-world code.
smalltalk-commenter
Generates CRC-style class comments for Smalltalk classes. Use after creating or modifying Tonel files to add or improve class documentation.
smalltalk-debugger
Systematic debugging guide for Pharo Smalltalk development. Provides expertise in error diagnosis (MessageNotUnderstood, KeyNotFound, SubscriptOutOfBounds, AssertionFailure), incremental code execution with eval tool, intermediate value inspection, error handling patterns (`on:do:` blocks), stack trace analysis, UI debugger window detection (read_screen for hung operations), and debug-fix-reimport workflow. Use when encountering Pharo test failures, Smalltalk exceptions, unexpected behavior, timeout or non-responsive operations, need to verify intermediate values, execute code incrementally for diagnosis, or troubleshoot Tonel import errors.
verl-rl-training
Provides guidance for training LLMs with reinforcement learning using verl (Volcano Engine RL). Use when implementing RLHF, GRPO, PPO, or other RL algorithms for LLM post-training at scale with flexible infrastructure backends.
openrlhf-training
High-performance RLHF framework with Ray+vLLM acceleration. Use for PPO, GRPO, RLOO, DPO training of large models (7B-70B+). Built on Ray, vLLM, ZeRO-3. 2× faster than DeepSpeedChat with distributed architecture and GPU resource sharing.
Didn't find tool you were looking for?