General

Handoff

Package shaped context — plans, decisions, project state — into a structured document for handoff to another AI agent, your future self, or a coworker.

Input

Mode: $ARGUMENTS

Parse the first word to determine mode:

• "list" — Show all existing handoffs
• "update [slug]" — Append new context to an existing handoff
• "view [slug]" — Read and summarize an existing handoff
• Anything else (or empty) — Create a new handoff (remaining args = description)

---

Before Starting (Auto-Gather)

Silently gather this context before asking any questions:

1. Plan files — Read all .claude/plans/*.md files. These are the primary context source.
2. Project instructions — Read CLAUDE.md or claude.md in the repo root (if it exists).
3. Git state — Run git log --oneline -20, git status, and git diff --stat.
4. Directory structure — Run ls at the repo root.
5. Existing handoffs — Read the list of files in .handoffs/ (if the directory exists).

Do NOT display or summarize what you gathered. Just hold it as context.

---

Mode: Create (Default)

Step 1: Determine context source

• If plan file(s) were found in .claude/plans/, use them as primary context. Tell the user: "I found a plan in .claude/plans/ — I'll use that as the basis for this handoff."
• If NO plan files found, ask: "I don't see an active plan file. Can you describe the project state and what needs to happen next?"

Step 2: Ask focused questions

Ask these two questions (together, not sequentially):

1. "Who is this handoff for?"

• AI agent (another Claude Code session, Janet, OpenClaw, etc.)
• Myself later (picking this up tomorrow/next week)
• A coworker (human who needs context)

1. "Anything to add beyond what's in the plan?" — gotchas, failed approaches, environment setup notes, constraints for the receiving agent, things that aren't in the plan but matter.

If the user provided a description in the arguments (e.g., /handoff auth system migration), use that as the handoff title/topic. Still ask the two questions above.

Step 3: Generate the handoff

Use the appropriate template based on audience. Generate a slug from the description or plan title (2-4 words, lowercase, hyphenated).

Save to: .handoffs/YYYY-MM-DD-[slug].md

Create the .handoffs/ directory if it doesn't exist.

After saving, confirm the file path and show a brief summary of what was captured.

---

Template: AI Agent

Use when the audience is another AI agent or Claude Code session. Optimize for an agent to pick up and start working immediately with zero additional context needed.

# Handoff: [Title]

**Created:** YYYY-MM-DD
**Author:** [user or "Claude Code session"]
**For:** AI Agent
**Status:** Ready to execute

---

## Summary

[2-3 sentences: what this project/task is, current state, what needs to happen next. An agent reading only this section should understand the mission.]

## Project Context

[What the project is. Tech stack. Architecture notes. Repo structure highlights. Only what's needed to orient — not a full tour.]

## The Plan

[The shaped plan from .claude/plans/ — include the full plan content here. This is the core of the handoff. If no plan file existed, include whatever plan/approach was described by the user.]

## Key Files

| File | Why It Matters |
|------|---------------|
| [path] | [brief reason] |

[Only files that are directly relevant. Not every file in the repo.]

## Current State

**Done:**
- [Completed items]

**In Progress:**
- [Partially done items with notes on where they stand]

**Not Started:**
- [Items that haven't been touched]

## Decisions Made

[Decisions that were made during planning/implementation. Include the reasoning so the agent doesn't re-litigate them.]

- **[Decision]** — [Why. What was considered and rejected.]

## Important Context

[Gotchas, failed approaches, environment notes, things that aren't obvious from the code.]

- [Item]

## Next Steps

[Priority-ordered. Each step should be specific enough for an agent to execute without guessing.]

1. **[Step]** — [Acceptance criteria or definition of done]
2. **[Step]** — [Acceptance criteria or definition of done]

## Constraints

[Rules the receiving agent must follow. Files not to touch. Patterns to maintain. Performance requirements. Security considerations.]

- [Constraint]

---

Template: Self/Later

Use when the user is creating a handoff for themselves to pick up later. Optimize for fast context recovery — "where was I and what was I thinking?"

# Handoff: [Title]

**Created:** YYYY-MM-DD
**Status:** Paused — pick up from here

---

## Where I Left Off

[Specific breadcrumbs. "I was in the middle of..." / "The last thing I did was..." / "I stopped because..."]

## The Plan

[The shaped plan. Full content from .claude/plans/ or user description.]

## What's Working

- [Verified/done items — things that are solid and don't need revisiting]

## What's Not Working Yet

- [Bugs, failing tests, half-done implementations. Be specific about symptoms.]

## My Current Thinking

[The approach being taken. Hypotheses. What I was about to try. Mental model of the problem.]

## Decisions I've Made

[So I don't re-litigate them next time.]

- **[Decision]** — [Why]

## Things I Tried That Didn't Work

[Save future-me the time of going down these paths again.]

- **[Approach]** — [Why it didn't work]

## Next Time I Pick This Up

[Ordered list of what to do first when resuming.]

1. [First thing]
2. [Second thing]

## Open Questions

[Unresolved items. Things I need to figure out. Unknowns.]

- [Question]

---

Template: Coworker

Use when the audience is a human coworker. Optimize for readability — assume they have no context on this work.

# Handoff: [Title]

**Created:** YYYY-MM-DD
**Author:** [name]
**Status:** [Handing off / Pausing / FYI]

---

## TL;DR

[3-4 sentences max. What is this, what state is it in, what needs to happen.]

## Background

[Why this work exists. Business context. What problem it solves.]

## The Plan

[What's been planned. High-level approach.]

## What's Done

- [Completed items]

## What's Left

- [Remaining items with rough effort estimates if useful]

## Key Decisions

[Decisions made and why — so they aren't revisited without reason.]

- **[Decision]** — [Rationale]

## Things to Watch Out For

[Gotchas, fragile areas, known issues, technical debt.]

- [Item]

## Where to Get Help

[Who to ask. Relevant docs. Slack channels. Related PRs.]

- [Resource]

---

Mode: List

Read all files in .handoffs/ and display them as a table:

| Date | Slug | For | Status | Summary |

For each file, read just the frontmatter (first ~10 lines) to extract date, audience, and status. Show the first sentence of the Summary section.

If no .handoffs/ directory or no files exist, say: "No handoffs found in this repo. Run /handoff to create one."

---

Mode: Update

Parse the slug from arguments (everything after "update").

Find the matching file in .handoffs/. Match on slug (the part after the date). If no exact match, list available handoffs and ask which one.

Ask:

1. What's changed since the last handoff?
2. Any new decisions, blockers, or completed items?
3. Should the status change?

Append to the file — add an update section at the bottom. Never overwrite existing content.

---

## Update — YYYY-MM-DD

[New context, decisions, progress, blockers. Written in the same style as the original template.]

After saving, confirm what was added.

---

Mode: View

Parse the slug from arguments (everything after "view").

Find the matching file in .handoffs/. If no exact match, list available handoffs and ask which one.

Read the full file and present a conversational summary:

• What this handoff is about
• Who it's for
• Current status
• Key next steps
• Any updates that have been appended

Keep it brief — the user can read the full file if they want details.

---

After Delivering

For create mode, offer to:

• Create another handoff for a different audience (e.g., "Want an AI agent version too?")
• Review and adjust the handoff before sharing

For update mode, offer to:

• Continue updating
• View the full handoff

For list mode, offer to:

• View a specific handoff
• Create a new one

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: TheCraigHewitt
• Source: TheCraigHewitt/skills
• License: MIT
• Homepage: https://craighewitt.com

Install and usage instructions live in the source repository linked above.