AgentSkillsCN

reflect

借助 Reflexion 框架，对已完成的工作进行结构化反思，并沉淀持续学习的经验

SKILL.md
--- frontmatter
name: reflect
description: Structured reflection on completed work using Reflexion framework with persistent learnings
user-invocable: true
allowed-tools: Read, Write, Grep

Reflection: Session Analysis

Dynamic Context

  • •Session state: !cat .claude/memory/local/session-state.json 2>/dev/null | head -20 || echo "No session state"
  • •Recent learnings: !cat .claude/memory/local/learnings.md 2>/dev/null | tail -20 || echo "No previous learnings"
  • •Git activity: !git log --oneline --since="8 hours ago" 2>/dev/null | head -10 || echo "No recent commits"
  • •Failed commands: !cat .claude/memory/local/error-log.txt 2>/dev/null | tail -5 || echo "No error log"

Reflexion Framework

1. Result Evaluation

  • •Goal Achievement: What was requested vs what was delivered?
  • •Quality Assessment: Does the output meet professional standards?
  • •Unexpected Issues: What problems arose during execution?
  • •Deviation Analysis: Where did the approach diverge from the plan?

2. Process Analysis

  • •Efficiency: Were there unnecessary steps or redundant operations?
  • •Tool Usage: Were the right tools used for each task?
  • •Decision Points: What key decisions were made and why?
  • •Bottlenecks: What slowed down progress?

3. Pattern Extraction

Identify reusable patterns:

  • •

    What Worked Well (to repeat):

    • •Approach patterns that led to success
    • •Tool combinations that were effective
    • •Problem-solving strategies that worked
  • •

    What Didn't Work (to avoid):

    • •Anti-patterns encountered
    • •Common mistakes made
    • •Inefficient approaches used
  • •

    New Knowledge (to remember):

    • •Codebase patterns discovered
    • •Framework/library quirks found
    • •Project-specific conventions learned

4. Actionable Learnings

For each learning, format as:

markdown
## [YYYY-MM-DD] [Category]

**Context**: [When does this apply?]
**Learning**: [What was learned?]
**Action**: [What to do differently next time?]
**Confidence**: HIGH / MEDIUM / LOW

Categories: architecture, debugging, testing, tooling, performance, security, patterns

5. Persist Learnings

Append new learnings to .claude/memory/local/learnings.md, merging with existing entries:

  • •Deduplicate similar learnings (update confidence if repeated)
  • •Remove learnings contradicted by new evidence
  • •Keep the file under 100 entries (archive oldest LOW confidence items)

Output

Provide a concise reflection summary and confirm which learnings were persisted.