用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt;运行前请先检查来源。
npx skills add https://github.com/majiayu000/claude-skill-registry --skill handoff命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地?可下载 SkillsMP 当前能够提供的文件。
基于 SOC 职业分类
正在显示 SKILL.md
| name | handoff |
| description | Write compact session handoffs. |
Quick Ref: Create structured handoff for session continuation. Output:
.agents/handoff/YYYY-MM-DD-<topic>.md+ continuation prompt.
YOU MUST EXECUTE THIS WORKFLOW. Do not just describe it.
Create a handoff document that enables seamless session continuation.
Given $handoff [topic]:
mkdir -p .agents/handoff
If topic provided: Use it as the handoff identifier.
If no topic: Derive from recent activity:
# Recent commits
git log --oneline -5 --format="%s" | head -1
# Check current issue
bd current 2>/dev/null | head -1
# Check ratchet state
ao ratchet status 2>/dev/null | head -3
Use the most descriptive source as the topic slug.
Topic slug format: 2-4 words, lowercase, hyphen-separated (e.g., auth-refactor, api-validation).
Fallback: If no good topic found, use session-$(date +%H%M) (e.g., session-1430).
Review what was done this session:
# Recent commits this session (last 2 hours)
git log --oneline --since="2 hours ago" 2>/dev/null
# Recent file changes
git diff --stat HEAD~5 2>/dev/null | head -20
# Research produced
ls -lt .agents/research/*.md 2>/dev/null | head -3
# Plans created
ls -lt .agents/plans/*.md 2>/dev/null | head -3
# Issues closed
bd list --status closed --since "2 hours ago" 2>/dev/null | head -5
Determine where we stopped:
Check for in-progress work:
bd list --status in_progress 2>/dev/null | head -5
List files the next session should read first:
# Recently modified
git diff --name-only HEAD~5 2>/dev/null | head -10
# Key artifacts
ls .agents/research/*.md .agents/plans/*.md 2>/dev/null | tail -5
Write to: .agents/handoff/YYYY-MM-DD-<topic-slug>.md (use date +%Y-%m-%d)
# Handoff: <Topic>
**Date:** YYYY-MM-DDTHH:MM:SSZ
**Session:** <brief session description>
**Status:** <Paused mid-task | Between tasks | Blocked on X>
---
## What We Accomplished This Session
### 1. <Accomplishment 1>
<Brief description with file:line citations>
**Files changed:**
- `path/to/file.py` - Description
### 2. <Accomplishment 2>
...
---
## Where We Paused
<Clear description of pause point>
**Last action:** <what was just done>
**Next action:** <what should happen >
---
-
-
---
---
path/to/critical-file.py .agents/research/YYYY-MM-DD-topic.md
path/to/related-file.py
### Step 7: Write Continuation Prompt
**Write to:** `.agents/handoff/YYYY-MM-DD-<topic-slug>-prompt.md` (use `date +%Y-%m-%d`)
```markdown
# Continuation Prompt for New Session
Copy/paste this to start the next session:
---
## Context
<2-3 sentences describing the work and where we paused>
## Read First
1. The handoff doc: `.agents/handoff/YYYY-MM-DD-<topic-slug>.md`
2. <Other critical files>
## What I Need Help With
<Clear statement of what the next session should accomplish>
## Key Files
```
<Suggested skill to invoke, e.g., "Use $implement to continue">
### Step 8: Extract Learnings (Optional)
If significant learnings occurred this session, also run post-mortem:
```bash
# Check if post-mortem skill should be invoked
# (if >3 commits or major decisions made)
git log --oneline --since="2 hours ago" 2>/dev/null | wc -l
If ≥3 commits: Suggest running $post-mortem --quick to extract learnings.
If <3 commits: Handoff alone is sufficient; learnings are likely minimal.
Tell the user:
$post-mortem --quickIf this handoff is ending a Codex hookless thread, run:
ao codex ensure-stop --auto-extract 2>/dev/null || true
ao codex ensure-stop is idempotent for the same Codex thread, so duplicate handoff
invocations are safe. Use ao codex status only when you need to confirm
lifecycle health.
Output completion marker:
<promise>DONE</promise>
If no context to capture (no commits, no changes):
<promise>EMPTY</promise>
Reason: No session activity found to hand off
Handoff created:
.agents/handoff/20260131T143000Z-auth-refactor.md
.agents/handoff/20260131T143000Z-auth-refactor-prompt.md
Session captured:
- 5 commits, 12 files changed
- Paused: mid-implementation of OAuth flow
- Next: Complete token refresh logic
To continue: Copy the prompt from auth-refactor-prompt.md
<promise>DONE</promise>
Handoff captures state for continuation. Post-mortem captures learnings for the flywheel (full knowledge lifecycle).
For a clean session end:
$handoff # Capture state for continuation
$post-mortem --quick # Extract learnings for future
Both should be run when ending a productive session.
If ao CLI not available:
ao ratchet status check in Step 2.agents/handoff/User says: $handoff (after working on OAuth flow for 2 hours, need to stop)
What happens:
bd list (issue #42 still open)$post-mortem --quick to extract learningsResult: Handoff captures state, continuation prompt ready, post-mortem suggested.
User says: $handoff (just closed issue #40, about to start #41 next session)
What happens:
$implement #41 suggestionResult: Handoff captures clean boundary, continuation is simple.
User says: $handoff (no topic provided, agent derives from commits)
What happens:
Result: Topic auto-derived from git history, no user input needed.
| Problem | Cause | Solution |
|---|---|---|
| "No session activity found to hand off" | No commits, no file changes detected | Expected for idle sessions. Nothing to hand off. Start new work or skip handoff. |
| Handoff files not written | .agents/handoff/ directory does not exist or not writable | Run mkdir -p .agents/handoff or check directory permissions |
| Topic slug is generic "session-1430" | No descriptive commits or issues to derive topic from | Provide explicit topic: $handoff auth-refactor for better naming |
| Continuation prompt missing key context | Recent files or artifacts not listed in handoff | Manually add missing files to handoff document or re-run with explicit topic |
| Post-mortem suggested but no learnings | Agent sees ≥3 commits and auto-suggests $post-mortem --quick | Run $post-mortem --quick or skip if commits are trivial (agent can't judge learning quality, only commit count) |
../post-mortem/SKILL.md — Full validation + knowledge lifecycle (council + extraction + activation + retirement)../retro/SKILL.md — Quick-capture a learningscripts/validate.sh