AgentSkillsCN

continuous-learning

从会话中提取并应用模式。在实施结束时使用，以记录学习心得。

SKILL.md
--- frontmatter
name: continuous-learning
description: "Extract and apply patterns from sessions. Use at end of implementation to capture learnings."

Continuous Learning

Extract reusable patterns, anti-patterns, and decisions from implementation sessions.

When to Use

  • •At the end of a feature implementation
  • •After resolving a non-trivial bug
  • •After a refactoring session or failed approach that taught something valuable

What to Capture

Patterns (Reusable Solutions)

  • •Architectural patterns applied (e.g., "Repository pattern for DB access")
  • •Testing strategies that caught real bugs
  • •Error handling approaches that improved reliability
  • •Code organization decisions that improved readability

Anti-Patterns (Mistakes to Avoid)

  • •Approaches that seemed correct but failed in practice
  • •Assumptions that turned out to be wrong
  • •Configurations that caused subtle bugs
  • •Test patterns that produced flaky results

Decisions (Context-Dependent Choices)

  • •Why one approach was chosen over alternatives
  • •Trade-offs accepted and their justification
  • •What would trigger revisiting this decision

Learning Format

code
### [Category]: [Short Title]

**Context:** When/where this applies
**Learning:** What was discovered
**Evidence:** What proved this (test result, error, metric)
**Action:** What to do differently next time

Where to Store Learnings

Project-level: Add to progress.txt under ## Codebase Patterns -- stack-specific patterns, framework gotchas, naming conventions.

Session-level: Append to progress.txt under the session entry -- files changed, problems encountered, how they were resolved.

Knowledge base: If Obsidian or similar is configured -- architecture decisions, language insights, and debugging techniques that apply across projects.

How to Apply Learnings

  1. •At session start: Read ## Codebase Patterns in progress.txt before implementing
  2. •During implementation: Note patterns immediately; reference known anti-patterns to avoid them
  3. •At session end: Run the extraction checklist below

Extraction Checklist

  • • Did I discover a reusable pattern? Record it.
  • • Did I hit a problem that cost significant time? Record the anti-pattern.
  • • Did I make a non-obvious decision? Record the context and reasoning.
  • • Did I learn something about the framework or language? Record it.
  • • Would a future session benefit from knowing this? If yes, record it.

Quality Criteria

Good learnings are:

  • •Specific -- includes the exact context where it applies
  • •Actionable -- tells you what to do, not just what happened
  • •Evidenced -- based on an observed outcome, not a guess
  • •Discoverable -- stored where future sessions will naturally find it