AgentSkillsCN

learn-from-feedback

针对 PR 反馈、用户修正，或代码审查发现的问题，及时更新项目记忆、规则与标准。在收到 PR 审查意见、用户修正，或代码审查中逐渐显现规律时，主动运用此功能加以应对。

SKILL.md
--- frontmatter
name: learn-from-feedback
description: Process PR feedback, user corrections, or code review findings and update project memory, rules, and standards accordingly. Use proactively after receiving PR review comments, user corrections, or when patterns emerge from code review.

Learn from Feedback

When receiving feedback (PR comments, user corrections, code review findings), update the project's knowledge base so the same issues don't recur.

Process

  1. •Identify the feedback: Read the PR comments (gh pr view $ARGUMENTS --comments), or parse the user's correction from context
  2. •Classify the learning:
    • •Coding standard → update .claude/skills/coding-standards/SKILL.md
    • •Domain knowledge → update .claude/rules/domain.md
    • •Testing practice → update .claude/rules/testing.md
    • •Workflow/process → update CLAUDE.md (Git Workflow section) or relevant skill
    • •Project-specific pattern → update auto memory (MEMORY.md)
  3. •Apply the fix: If the feedback points to a code issue, fix it
  4. •Update the knowledge base: Write the learning to the appropriate file so it persists
  5. •Summarize: Report what was learned and where it was saved

Rules

  • •Be conservative: only save verified, stable patterns — not one-off corrections
  • •Keep entries concise: one bullet point per learning
  • •Don't duplicate: check if the learning already exists before adding
  • •If unsure where a learning belongs, prefer auto memory (lowest commitment)