Handoff

Generate a HANDOFF.md for sharing with teammates who need full context on what happened.

Process

1. •

   Check for existing file — Read ./HANDOFF.md if it exists. If found:

   • •Preserve previous versions under a ## Previous Handoff heading
   • •Add new content as the primary section
2. •

   Extract from conversation:

   • •The problem/task and why it matters (business context, not just technical)
   • •What was tried and what was learned
   • •Current state — what's done, what's not
   • •Recommendations with rationale
   • •Any risks, caveats, or "watch out for" items
   • •Links to relevant files, PRs, Jira tickets, dashboards
3. •

   Calibrate audience — This is for engineers who:

   • •Know the codebase but weren't in this session
   • •Need enough context to take over or review the work
   • •Don't need to know about Claude skills, prompt iterations, or session mechanics
4. •

   Write HANDOFF.md in the current working directory.

Output Format

# Handoff: {Descriptive title}

**Date:** {YYYY-MM-DD}
**Author:** {user}
**Ticket:** {Jira ticket if applicable}
**Branch:** {git branch}

---

## Background

{Why this work happened — the business problem or technical need. 2-3 sentences.}

## What Was Done

{Chronological or logical summary of the work. Include specifics.}

### Key Findings

- {Finding with evidence — data points, error messages, measurements}

### Changes Made

- `{file}`: {what and why}

## Current State

{What's working, what's not, what's partially done}

## Recommendations

1. {Action item with rationale}
2. {Action item with rationale}

## Risks & Caveats

- {Thing that could bite someone who picks this up}

## References

- [{Jira ticket}]({url})
- [{Dashboard/monitoring}]({url})
- [{Related PR}]({url})

---

_Handoff generated {date} — reach out to {author} for questions._

Rules

• •No Claude artifacts — Strip references to skills, prompts, session mechanics. This is for humans.
• •Business context first — Start with "why" before "what". Teammates need to understand importance.
• •Evidence-based — Include actual numbers, error messages, query results. Not "it seems slow" but "p99 latency hit 4.2s."
• •Actionable recommendations — Each recommendation should have enough context that someone can act on it without asking follow-up questions.
• •Link everything — Jira tickets, PRs, dashboards, relevant files. Make it easy to dig deeper.
• •Skip sections with no content — Don't include empty Risks or References sections.
• •Write to ./HANDOFF.md unless the user specifies a different path