handoff

Write a HANDOFF.md that captures the current session's full context so another agent can continue seamlessly. Use this skill whenever the user says things like "write a handoff", "save progress for the next agent", "create a handoff doc", "hand this off", "write context for the next session", "document what we've done so the next agent can pick up", or any variation suggesting they want to preserve session state for a successor agent. Also trigger when the user mentions HANDOFF.md by name. Even if the user just says something like "I'm done for now, save the state" or "wrap up so someone else can continue", this skill should activate.

Safety Notice

This listing is imported from skills.sh public index metadata. Review upstream SKILL.md and repository scripts before running.

Copy this and send it to your AI assistant to learn

Install skill "handoff" with this command: npx skills add yigegongjiang/jj-skills/yigegongjiang-jj-skills-handoff

Handoff

Generate a HANDOFF.md in the project root that gives a fresh agent everything it needs to continue the current task without re-reading the entire conversation.

Why this matters

Context windows are finite. When a session ends or gets too long, the next agent starts cold. A good handoff document is the difference between the next agent picking up in seconds versus spending 10 minutes re-discovering what you already know. Write it as if you're briefing a colleague who is smart but has zero context about this project or task.

When triggered

  1. Read the full conversation history and any files you modified or read during the session.
  2. Write HANDOFF.md to the project root. If the file already exists, overwrite it completely.

What to write

Organize the document under these sections. Omit any section that genuinely has nothing to report — an empty section is worse than no section.

1. Objective

One or two sentences: what is the user trying to accomplish? State the end goal, not the intermediate steps.

2. Current Status

Where things stand right now. Be specific: "Tests pass on module X but module Y has a failing assertion on line 42" is useful; "making progress" is not.

3. What Was Done

A concise list of actions taken in this session. For each item, note the outcome:

  • What you tried
  • Whether it worked, partially worked, or failed
  • If it failed, why (error messages, root cause if known)

This is the most valuable section for the next agent. Don't just list successes — the failures and dead ends save the next agent from repeating them.

4. Key Decisions & Reasoning

Any non-obvious choices made during the session and why. Architecture decisions, library choices, trade-offs accepted. The next agent needs to know what's intentional vs. accidental.

5. Open Issues & Blockers

Anything unresolved: bugs, unanswered questions, external dependencies, things the user needs to decide. Be specific enough that the next agent can act on these.

6. Next Steps

Concrete, actionable items the next agent should do first. Prioritize them. If there's a specific order that matters, say so and explain why.

7. Relevant Files & Paths

List files that are central to the task — files modified, config files that matter, test files to run. Use relative paths from the project root. Don't dump every file you glanced at; only include ones the next agent will actually need to open.

Writing principles

  • Be specific, not vague. File names, line numbers, error messages, command outputs. Concrete details are what make a handoff useful.
  • Capture the why, not just the what. "Changed the retry logic from 3 to 5 attempts because the API has intermittent 502s under load" is far more useful than "updated retry count".
  • Include failures. Dead ends are valuable knowledge. If you spent time investigating something that turned out to be irrelevant, say so — it prevents the next agent from going down the same path.
  • Keep it scannable. Use bullet points and short paragraphs. The next agent should be able to skim the document in under a minute and know exactly where to start.
  • No filler. Don't pad sections. Don't add generic advice. Every sentence should carry information the next agent can act on.

Source Transparency

This detail page is rendered from real SKILL.md content. Trust labels are metadata-based hints, not a safety guarantee.

Related Skills

Related by shared tags or category signals.

General

polish-technical-docs

No summary provided by upstream source.

Repository SourceNeeds Review
Coding

code-path-analyzer

No summary provided by upstream source.

Repository SourceNeeds Review
General

tech-content-optimizer

No summary provided by upstream source.

Repository SourceNeeds Review
Automation

handoff

No summary provided by upstream source.

Repository SourceNeeds Review
243-boshu2