continual-learning

Continual Learning Skill

This skill enables agents to systematically capture lessons learned from tasks, building persistent knowledge that improves over time.

Core Principle

After completing any task, answer three questions:

1. What went right?
2. What went wrong?
3. What would I do differently next time?

Record these insights in the knowledge base for future reference.

When to Use This Skill

Use this skill when:

• Completing any non-trivial task (2+ steps)
• Debugging and fixing bugs
• Learning new tools or frameworks
• Discovering patterns in the codebase
• Making architectural decisions
• Encountering edge cases

Knowledge Base Structure

Lessons are stored in .opencode/memory/lessons_learned.md with these sections:

General Do's

• Patterns and practices that worked well
• Approaches worth reusing
• Good habits to maintain

General Don'ts

• Mistakes to avoid
• Pitfalls that caused problems
• Anti-patterns to stay away from

Edge Cases

• Unexpected scenarios encountered
• Special cases that need handling
• Platform/tool-specific quirks

Project-Specific Patterns

• This project's unique conventions
• Workflows specific to this codebase
• Custom tooling or scripts used here

Usage Workflow

1. After Task Completion

When a task is done, reflect on the experience and identify key learnings.

2. Categorize the Lesson

Decide which section the lesson belongs to:

• Do's: Good practices that worked
• Don'ts: Mistakes or bad approaches
• Edge Cases: Unexpected scenarios
• Project-Specific: Patterns unique to this project

3. Record the Lesson

Add the lesson to the appropriate section in .opencode/memory/lessons_learned.md

Format each lesson as:

- **[Topic]**: [Concise description of the lesson]
  - [Key insight or action item]
  - [Additional context or examples]

4. Cross-Reference (Optional)

If a lesson relates to other parts of the codebase, add references:

- **[Topic]**: [Description]
  - [Insight]
  - Related: `path/to/file`, `component-name`, or `pattern-name`

Examples

Recording a "Do"

- **LSP Tools for Navigation**: Use `lsp_goto_definition` to find symbol definitions
  - Jump directly to where a function/class is defined
  - More reliable than manual file searching
  - Related: `lsp_find_references` for usage analysis

Recording a "Don't"

- **Suppressing Type Errors**: Never use `@ts-ignore` or `as any` to silence errors
  - Errors are warnings about real issues
  - Always fix the root cause instead
  - Type safety catches bugs at compile time

Recording an Edge Case

- **OpenCode Skills Location**: Skills can be in two locations
  - `.opencode/skill/<name>/SKILL.md` (preferred)
  - `.claude/skills/<name>/SKILL.md` (legacy)
  - The `skill` tool searches both locations

Recording a Project Pattern

- **Smart Task Workflow**: This project uses `/smart-task` command for task management
  - Always invoke with `/smart-task` for systematic task handling
  - Combines reading, acting, and recording in one flow
  - Located at `.opencode/command/smart-task.md`

Best Practices

Be Specific

• Instead of "Test code is important", write "Use describe blocks for test organization"
• Concrete advice is more actionable than generalities

Include Context

• Note what project, language, or tool the lesson applies to
• Mention version numbers if relevant
• Link to specific files or components when possible

Keep It Concise

• One key insight per lesson entry
• Bullet points for multiple related points
• Avoid lengthy explanations—focus on the core learning

Update Existing Entries

• If you find a better approach, add it to existing lessons
• Mark outdated lessons with ~~strikethrough~~ and note the replacement
• Keep the knowledge base fresh and accurate

Review Before Starting New Tasks

• Check if similar tasks have been done before
• Look up relevant sections for guidance
• Apply documented patterns to your current work

Integration with OpenCode Commands

This skill works alongside the /smart-task command:

• /smart-task: Full workflow for user-invoked tasks (Read → Act → Record)
• This skill: Lightweight guidance for continual learning behavior

Agents can load this skill to make continual learning a natural part of their workflow, even when not explicitly using the /smart-task command.

Tips for Effective Learning

1. Reflect Immediately: Capture lessons right after completing a task while fresh
2. Look for Patterns: Notice recurring themes across different tasks
3. Share Learnings: Discuss lessons with team members to validate insights
4. Iterate and Improve: Update lessons as you discover better approaches
5. Apply Proactively: Before starting a task, review related lessons

Skill Loading

Load this skill using:

skill continual-learning

Once loaded, the agent will have continual learning guidance available for all tasks.