Audit websites for accessibility issues and WCAG compliance. Use when checking accessibility, fixing a11y issues, or ensuring WCAG compliance.
中文简介正在准备中。当前显示原文说明。
Use when a working session is ending, the context window is nearly full, work needs to pass to a teammate or a fresh session, or the user says "hand this off", "write this up before we lose it", "I'm starting a new session", or "summarize where we are". Produces a handoff document that lets the next session resume without re-reading the transcript.
中文简介正在准备中。当前显示原文说明。
查看安装方法安装前可以查看交给 AI 代理的指令内容。
A session ends three ways: the context fills up, the person leaves, or the work moves to someone else. All three lose the same thing — the reasoning that never made it into a file. This skill writes that reasoning down before it evaporates.
Write a handoff file, do not summarize in chat. Chat scrolls away; a file survives a /clear, a crash, and a different person opening the repo tomorrow.
Default path: HANDOFF.md at the repo root, or docs/handoff/<yyyy-mm-dd>-<topic>.md if the repo already has a docs convention. If a handoff file exists, append a new dated section rather than overwriting — the history of what was tried is half the value.
Write these sections in this order. Skip a section only when it is genuinely empty.
Goal. One sentence: what this work is supposed to achieve, in the user's terms, not the ticket's.
State. What is done, what is half-done, what has not been started. Be specific about half-done — "auth works, but the refresh-token path is stubbed and returns a fixed string" beats "auth in progress".
Files touched. Path plus one line on what changed and why. Include files you read and decided not to change when the decision was non-obvious.
Decisions and why. The choices a reasonable person would question. Each one: what was chosen, what was rejected, and the reason. This is the section that saves the next session an hour.
Dead ends. What was tried and did not work. Name the failure mode. Without this, the next session repeats it.
Next step. The single next action, concrete enough to start on immediately. Not "continue the refactor" — "extract parseInvoice from billing.ts:210 into its own module, then point the three callers at it".
Open questions. Anything waiting on a human. Name who, and what the question is.
Prefer evidence over memory:
git status and git diff --stat for what actually changed on disk.git log --oneline -15 for what landed.If the transcript disagrees with the repo, the repo wins. Say so in the handoff.
# Handoff — <topic> — <date>
**Goal:** <one sentence>
**Branch:** <name> (base: <name>) · **Uncommitted:** <yes, list / no>
## State
- Done: ...
- In progress: ...
- Not started: ...
## Files
- `path/to/file.ts` — <what and why>
## Decisions
- <chose X over Y because Z>
## Dead ends
- <tried X, failed with Y — don't repeat>
## Next step
<one concrete action>
## Open questions
- <question> — needs <person>
## Getting back to work
```bash
<commands>
## When the Context Window Is the Reason
If the handoff is being written because context is nearly full, write the file first, before any other work. Then tell the user to start a fresh session and open the file. Do not try to squeeze in one more fix — a truncated handoff is worse than an early one.
还没有评价。欢迎分享你的使用感受。
无需登录即可发表。欢迎分享使用感受、提问,或说明在哪些环境下无法运行。
还没有评论。来写第一条吧。
Audit websites for accessibility issues and WCAG compliance. Use when checking accessibility, fixing a11y issues, or ensuring WCAG compliance.
中文简介正在准备中。当前显示原文说明。
Deploy a 2-layer parallel agent hierarchy for large, parallelizable work — big refactors, multi-file migrations, codebase-wide audits, bulk generation. A top-tier commander (Fable or Opus) orchestrates the swarms; the user picks a power level (Max Power / Heavy / Balanced / Economy) that sets the Opus/Sonnet/Haiku model mix per layer. Layer 1 is 3-50+ specialist agents, each with its own full context window; Layer 2 is 2+ sub-agents per member. Includes git safety, tiered sizing, a pre-deploy gate, phantom-completion checks, and multi-wave follow-up.
中文简介正在准备中。当前显示原文说明。
Use to test whether an AI agent, prompt, or skill actually works before or after it ships — build a small eval set, score runs, and catch regressions when the prompt or model changes. Trigger when the user asks whether a prompt is good, wants to compare models or versions, sees inconsistent agent output, or is about to put an agent in front of customers.
中文简介正在准备中。当前显示原文说明。
Deploys swarms of sub-agents for massive parallel data processing tasks. Unlike agent-army (which is for code changes), this is for DATA tasks -- processing 1000 documents, analyzing datasets, bulk content generation. Configurable swarm size, task distribution, result aggregation, progress tracking, and error recovery.
中文简介正在准备中。当前显示原文说明。
Designs and deploys custom agent teams for specific business workflows. Interactive discovery of business processes, then generates complete team configurations with specialized agent roles, tool access, communication protocols, and handoff rules.
中文简介正在准备中。当前显示原文说明。
Agent-to-Agent (A2A) communication protocol. Connect two or more Claude agents that pass messages, share context, delegate tasks, and collaborate. Implements structured handoffs, shared memory, and multi-agent conversations.
中文简介正在准备中。当前显示原文说明。