agentleFS
rumoiirumoii/rumo-engineering-skills/skills/rumo-handoff/SKILL.md

rumo-handoff

Skill7 starsChanged 14 days agoMITView on GitHub

Use when the user asks to hand off, hand over, pause, or package the current task so another session, agent, or client can continue it, or when a new session must pick up earlier work, such as asking for the current progress, the next step, or where a topic stands after days away ("汇报一下当前进度", "下一步做什么", "接着上次做", "回忆一下 X 做到哪了") or to continue another agent's session. Writes a handoff document only when asked; picks up from this project's recent Claude Code, Codex, and Kimi Code sessions, checked against the repository. Do not use for an end-of-task summary within the same session.

We split this skill into the rules it gives an agent

10 rules. About 0 are common, 10 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈10

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use when this session starts on work that an earlier session or agent did, and the user asks where things stand, what to do next, or to …

    line 12
  2. Distinctive:

    Check the record against the repository: branch, HEAD, recent commits, dirty files, and any background runs it mentions.

    line 17
  3. Distinctive:

    Stop after the report unless the user asked to continue.

    line 19
  4. Distinctive:

    Write the document the next agent needs to continue without this conversation, only when the user asks for one.

    line 25
  5. Distinctive:

    Use the location the user names.

    line 29
  6. Distinctive:

    Verify each fact now; do not copy it from memory.

    line 42
  7. Distinctive:

    Write for a reader with no access to this conversation.

    line 76
  8. Distinctive:

    Never include passwords, tokens, keys, or private certificate material.

    line 77
  9. Distinctive:

    Keep authorizations explicit: say what the user approved and what still needs approval.

    line 78
  10. Distinctive:

    Prefer facts over narrative and keep it short enough to read in two minutes.

    line 79

≈ “Common” is an estimate: at least 1 in 5 of the 308,232 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Rumo Handoff

Carry a task across sessions. Write a handoff document when the user asks for one; in a new session, pick up from the earlier sessions' own records, which already exist, so nothing has to be written at the end of a session.

Pick Up

Use when this session starts on work that an earlier session or agent did, and the user asks where things stand, what to do next, or to continue.

  1. From the project directory, run scripts/session_digest.py list. It lists this project's recent Claude Code, Codex, and Kimi Code sessions, newest first, and never reads another project's. The newest may be this session: skip the one whose last_user is the current request. Sessions with headless: true were launched by another agent, such as a review lane; prefer the others.
    • When the user names a topic ("where is the recognition work"), has been away for days, or several lines of work may be open, list a window instead: list --since 14 --limit 30, adding --grep '<term|synonym|English term>' for a topic. --grep searches the user's messages, compaction summaries, and titles. Group the sessions into lines of work by their first and last user messages.
  2. Run scripts/session_digest.py show <id> for the latest relevant session, and for one more when it stopped mid-task or points to an earlier one. With several lines of work, show only the latest session of each unfinished line, at most three, with --budget 15000. It keeps the user's messages, each turn's final reply, and the latest compaction summary. When the user names a session, such as codex resume <id> or kimi --session <id>, show that one. Also read any handoff document or plan file the record names.
  3. Check the record against the repository: branch, HEAD, recent commits, dirty files, and any background runs it mentions. Where they disagree, the repository wins; say so.
  4. Report where the work stopped, what was done and verified, what is unfinished or unverified, the recommended next step, and which steps need the user's approval. Treat the record's decisions as settled and do not redo verified work; re-check only what the repository contradicts. With several lines of work, give each one line led by one status checked against git, in the user's language: pushed, committed but not pushed, uncommitted and verified, uncommitted and unverified, in progress on a branch, reverted, planned but not started, or dropped. Then list at most five recurring problems or dead ends, and end with the single most useful next step.
  5. Stop after the report unless the user asked to continue. Approvals in the record covered that session; ask again before commits, pushes, deployments, and other actions that need authorization. Never repeat credentials that appear in a record.

Without session files, such as work from another machine, ask for a handoff document or where the session's record is.

Write A Handoff Document

Write the document the next agent needs to continue without this conversation, only when the user asks for one.

Where It Goes

Use the location the user names. Otherwise follow the workspace's convention for task notes; if there is none, ask. Do not commit the document unless asked, and keep it out of product source trees by default.

Make the handoff discoverable without this conversation. When the workspace's own instructions name a task index, status document, or handoff registry that the next agent reads, add a concise link there with the current state and check that it resolves. Do not edit other tracked files, such as a product README, to point at an uncommitted handoff, and do not create a parallel documentation system. Otherwise give the user the exact path to pass to the receiving agent; ask where to link it when the user expects automatic discovery.

Gather Facts First

  • Re-read the original request and every later decision the user made.
  • For each repository involved: path, branch, HEAD, upstream, worktree status, and any worktrees or stashes this task created.
  • Commits made in this task, and anything staged, unstaged, or untracked that belongs to it.
  • Checks actually run, with commands and results.
  • Running processes, terminals, servers, or remote sessions the next agent must know about.
  • Approaches tried and abandoned. Copy the plan's decisions and dead ends when the task has a plan file, then add any the conversation shows that it does not.

Verify each fact now; do not copy it from memory.

Document Structure

# Handoff: <task>

## Goal
One or two sentences, plus acceptance criteria.

## Decisions
What the user decided, with the reason when it matters. Mark anything still open.

## State
Repositories, branches, commits, dirty files, processes, environments.

## Verified / Not Verified
What was checked and how; what remains unproven and why.

## Next Steps
Ordered, concrete actions. Mark steps that need user approval.

## Tried And Dropped
Approaches tried and abandoned, each with why and where the evidence is.

## Pitfalls
Traps already hit and how they were resolved.

## Resume
The exact commands or reads to get back to a known-good state.

Rules

  • Write for a reader with no access to this conversation; never say "as discussed above".
  • Never include passwords, tokens, keys, or private certificate material; name where credentials live instead.
  • Keep authorizations explicit: say what the user approved and what still needs approval. A handoff does not transfer approval for actions the user has not granted.
  • Prefer facts over narrative and keep it short enough to read in two minutes.
  • This skill writes the document and its necessary navigation link only. To start the receiving agent and confirm it took over, use rumo-agent-collaboration after the user authorizes that route. If the skill or route is unavailable, give the user the document and stop.

Report

Give the document path, how the next agent will discover it, and a three-line summary: where the task stands, the next step, and anything blocked.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.