成功 Criteria
- Every mistake/friction point cites specific moment 和 impact
- Improvements actionable, prioritized, 和 <= 10 items
- 每个 skill audit proposes measurable changes (不 vague suggestions)
- 用户 asked 哪个 items 到 persist 到 memory
Process
1. 会话 Review
Scan the full conversation. For each finding, cite the specific exchange (quote or paraphrase) and its impact.
| Category | Signal |
|---|
| Mistakes | Wrong outputs, incorrect assumptions, hallucinated facts |
| Friction | Repeated clarifications, verbose responses, misread intent |
| Wasted effort | Work discarded, wrong approaches tried first |
| Wins | Approaches worth repeating, smooth interactions |
Skip one-time typos, external tool failures, and issues outside agent control.
2. Improvements
Numbered list of concrete improvements, ranked by impact. Each item: one sentence, imperative, actionable. Cap at 10.
Ask: "Which of these should I remember for future chats?"
Save approved items to memory files at ~/.claude/projects/.../memory/ using the Write tool with proper frontmatter (see MEMORY.md index).
3. Skill Audit (如果 skills 是 used)
For each skill invoked during the session:
. Self-check gate -- 如果 skill lacks 成功 criteria + verification 循环:
- 添加
## 成功 Criteria 在 top (3-5 measurable checks)
- 添加
## Self-Check 在 bottom: "验证 所有 成功 criteria met 之前 presenting 输出. 如果 不, iterate (max 5 乘以)."
B. 令牌 efficiency -- 标志: redundant phrasing, mergeable sections, oversized examples, "Claude 已经 knows " content, inert frontmatter metadata.
C. 其他 -- Missing edge cases, vague directives (rewrite 作为 measurable criteria 或 移除), naked negations (添加 "做 Y 代替" 或 移除).
Present proposed changes as diffs. Ask: "Apply these? (all / pick / skip)"
4. Pattern Detection
If 2+ similar tasks appear that no existing skill covers, suggest a new skill (1-2 sentence description). Create only after confirmation.
Proactive trigger: 当...时 用户 corrects 您, clarifies 相同 thing twice, 或 shows frustration, append: "Tip: 类型 /reflect 当...时 您're 就绪 -- I'll review 什么 我们 可以 improve."
Self-Check
Before presenting output, verify all success criteria are met. If any fail, revise (max 5 iterations).
Success Criteria
- Every mistake/friction point cites the specific moment and its impact
- Improvements are actionable, prioritized, and <= 10 items
- Each skill audit proposes measurable changes (not vague suggestions)
- User is asked which items to persist to memory
Process
1. Session Review
Scan the full conversation. For each finding, cite the specific exchange (quote or paraphrase) and its impact.
| Category | Signal |
|---|
| Mistakes | Wrong outputs, incorrect assumptions, hallucinated facts |
| Friction | Repeated clarifications, verbose responses, misread intent |
| Wasted effort | Work discarded, wrong approaches tried first |
| Wins | Approaches worth repeating, smooth interactions |
Skip one-time typos, external tool failures, and issues outside agent control.
2. Improvements
Numbered list of concrete improvements, ranked by impact. Each item: one sentence, imperative, actionable. Cap at 10.
Ask: "Which of these should I remember for future chats?"
Save approved items to memory files at ~/.claude/projects/.../memory/ using the Write tool with proper frontmatter (see MEMORY.md index).
3. Skill Audit (if skills were used)
For each skill invoked during the session:
A. Self-check gate -- If the skill lacks success criteria + verification loop:
- Add
## Success Criteria at top (3-5 measurable checks)
- Add
## Self-Check at bottom: "Verify all success criteria are met before presenting output. If not, iterate (max 5 times)."
B. Token efficiency -- Flag: redundant phrasing, mergeable sections, oversized examples, "Claude already knows this" content, inert frontmatter metadata.
C. Other -- Missing edge cases, vague directives (rewrite as measurable criteria or remove), naked negations (add "do Y instead" or remove).
Present proposed changes as diffs. Ask: "Apply these? (all / pick / skip)"
4. Pattern Detection
If 2+ similar tasks appear that no existing skill covers, suggest a new skill (1-2 sentence description). Create only after confirmation.
Proactive trigger: When the user corrects you, clarifies the same thing twice, or shows frustration, append: "Tip: Type /reflect when you're ready -- I'll review what we can improve."
Self-Check
Before presenting output, verify all success criteria are met. If any fail, revise (max 5 iterations).
by