Claude Skill

handover

Write the handover a fresh session starts from: the objective, what is true now, the evidence actually obtained, what is left, the open decisions and the next concrete action, reconciled with the disk first. Started by name only: /agent-harness:handover in Claude Code, $agent-har

LLM Mart · 0 points · 0 views 0 listing impressions 0 install-command copies
Download phuryn-product-engineering-agent-harness_skills_handover-39e839d.zip · 2 KB

Install

skills CLI npx skills add https://github.com/phuryn/product-engineering/tree/main/agent-harness/skills/handover
Claude Code claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install phuryn-product-engineering@llmmart
Git git clone https://github.com/phuryn/product-engineering.git

The skills CLI installs just this skill, for any of its supported agents. Claude Code installs the whole phuryn/product-engineering collection as a plugin from our marketplace. Git is the plain clone.

Skill manifest

Handover — what the next session starts from

Write the one note the next session reads first. The session is ending, or its context is; work continues tomorrow, in a fresh session, or with another agent. Start that session from what is true rather than what this session remembers.

Invocation

/agent-harness:handover                          Claude Code, plugin installed
/agent-harness:handover the next session finishes the billing stage
$agent-harness:handover                          Codex, plugin installed
/handover   ·   $handover                        the skill folder copied into a repository

Steps

1. Reconcile with reality first

Check what this session believes against reality before writing a word.

  • Check the disk and name uncommitted work as uncommitted. Read git status, the branch, commits made this session, and files the work touched.
  • Check every background job, server, watcher and build this session started. Each is still running, finished with its exit code read from its log, or gone.
  • Correct project state files where the disk contradicts them. Check the plan's status, status board, backlog and earlier handover; correct those lines rather than copying them.

An inherited report is a claim, this session's own memory included.

2. Write what the next session needs

Section Holds
Objective what the work is for, in a line, and who asked
Where it stands what is true now: built, committed, pushed, deployed, each said plainly
Evidence what was actually run and observed: the command, its exit code, the number, the verdict. Not what was intended. Anything unverified says so
Left to do in order, each small enough to start on
Open decisions each with its options, a recommendation, and whose decision it is
Where things are branches, worktrees, files, logs, URLs, processes still running and how to stop them
Next action one concrete first step: a command to run, a file to open. Never "continue"

Keep it short enough to stay true, preserving the evidence. Evidence costs most to lose. A re-run experiment costs money; restating a number from memory is how a wrong one reaches the owner.

3. Put it where the project keeps such things

Use the project's own handover convention before creating a new file. Use its status board, plan's status lines or existing handoff file. With none, keep one HANDOVER.md at the repository root, rewritten in place each time. It says what is true now, not the history.

Use the flight-deck skill for hours, multiple sessions or background jobs. That work wants its status board and log instead of a single note. A few lines are enough for a short session.

Name where secrets are kept without putting them in the handover. No keys, tokens, passwords or connection strings.

4. Tell the owner how to start the next session

End with the next session's starting line and anything left running. For example: Read HANDOVER.md and start from its next action.

Each rule's origin, the case that taught it, is in receipts.md beside this file; read it when a rule's edge is unclear.

How you can tell it worked

A fresh session given only the note can say what the work is for, what is verified and what is not, and run the next action without asking what happened.

Files (product-engineering)
  • agents
    • openai.yaml 43 B
      Loading…
  • receipts.md 263 B
    Loading…
  • SKILL.md 3.7 KB
    ---
    name: handover
    description: "Write the handover a fresh session starts from: the objective, what is true now, the evidence actually obtained, what is left, the open decisions and the next concrete action, reconciled with the disk first. Started by name only: /agent-harness:handover in Claude Code, $agent-harness:handover in Codex."
    argument-hint: "<optional: what the next session will do>"
    disable-model-invocation: true
    ---
    
    # Handover — what the next session starts from
    
    **Write the one note the next session reads first.** The session is ending, or its context is; work continues tomorrow, in a fresh session, or with another agent. Start that session from what is true rather than what this session remembers.
    
    ## Invocation
    
    ```
    /agent-harness:handover                          Claude Code, plugin installed
    /agent-harness:handover the next session finishes the billing stage
    $agent-harness:handover                          Codex, plugin installed
    /handover   ·   $handover                        the skill folder copied into a repository
    ```
    
    ## Steps
    
    ### 1. Reconcile with reality first
    
    **Check what this session believes against reality before writing a word.**
    
    - **Check the disk and name uncommitted work as uncommitted.** Read `git status`, the branch, commits made this session, and files the work touched.
    - **Check every background job, server, watcher and build this session started.** Each is still running, finished with its exit code read from its log, or gone.
    - **Correct project state files where the disk contradicts them.** Check the plan's status, status board, backlog and earlier handover; correct those lines rather than copying them.
    
    **An inherited report is a claim, this session's own memory included.**
    
    ### 2. Write what the next session needs
    
    | Section | Holds |
    |---|---|
    | Objective | what the work is for, in a line, and who asked |
    | Where it stands | what is true now: built, committed, pushed, deployed, each said plainly |
    | Evidence | what was actually run and observed: the command, its exit code, the number, the verdict. Not what was intended. Anything unverified says so |
    | Left to do | in order, each small enough to start on |
    | Open decisions | each with its options, a recommendation, and whose decision it is |
    | Where things are | branches, worktrees, files, logs, URLs, processes still running and how to stop them |
    | Next action | one concrete first step: a command to run, a file to open. Never "continue" |
    
    **Keep it short enough to stay true, preserving the evidence.** Evidence costs most to lose. A re-run experiment costs money; restating a number from memory is how a wrong one reaches the owner.
    
    ### 3. Put it where the project keeps such things
    
    **Use the project's own handover convention before creating a new file.** Use its status board, plan's status lines or existing handoff file. With none, keep one `HANDOVER.md` at the repository root, rewritten in place each time. It says what is true now, not the history.
    
    **Use the `flight-deck` skill for hours, multiple sessions or background jobs.** That work wants its status board and log instead of a single note. A few lines are enough for a short session.
    
    **Name where secrets are kept without putting them in the handover.** No keys, tokens, passwords or connection strings.
    
    ### 4. Tell the owner how to start the next session
    
    **End with the next session's starting line and anything left running.** For example: *Read `HANDOVER.md` and start from its next action.*
    
    Each rule's origin, the case that taught it, is in `receipts.md` beside this file; read it when a rule's edge is unclear.
    
    ## How you can tell it worked
    
    A fresh session given only the note can say what the work is for, what is verified and what is not, and run the next action without asking what happened.
    

Comments (0)

Sign in to join the conversation.

No comments yet.

Reviews (0)

No reviews yet.