mcpbeat

M0 Handoff

coco-research/m0-handoff

Session handoff and resume for M0. Write a compact_checkpoint before a session ends or context is compacted, and resume from the latest one at the start of a session in any tool. Turns a cold start into a continuation. Triggers on: 'm0 handoff', 'hand off', 'session handoff', 'save session state', 'resume where we left off', 'pick up where we left off', 'before I compact', 'end of session', 'continue in Cursor', 'continue in Claude Code'.

2k tokens
context cost
the whole folder, loaded on every use
1
files
instructions only
0
copies elsewhere
how many repositories repackaged it
196
stars on the repo
on the repository, not the skill itself

Install

one command, takes just this skill from the repository
npx skills add https://github.com/coco-research/coco --skill m0-handoff

The instruction itself

7 sections, as written by the author

/m0-handoff — Hand Off and Resume

Two halves of one flow, both against the M0 operational thread:

  • Hand off — write a compact_checkpoint that a session with no other context

could act on.

  • Resume — read the latest one and continue, in this tool or another.

This is what M0 exists for. Everything else in the bundle supports it.

Quick Reference

M0="${M0_BASE_URL:-http://127.0.0.1:8787}"
M0S="$HOME/.claude/skills/m0/scripts"
PROJECT="${M0_PROJECT:-$(basename "$PWD")}"

# Hand off
curl -s -X POST "$M0/api/brain/checkpoint" -H 'Content-Type: application/json' -d "$(python3 - <<'PY'
import json, os, subprocess
def git(*a):
    try: return subprocess.check_output(("git",)+a, text=True, stderr=subprocess.DEVNULL).strip()
    except Exception: return None
print(json.dumps({
  "project": os.environ.get("M0_PROJECT") or os.path.basename(os.getcwd()),
  "kind": "compact_checkpoint",
  "source_tool": "claude-code",
  "branch": git("rev-parse", "--abbrev-ref", "HEAD"),
  "head_sha": git("rev-parse", "--short", "HEAD"),
  "text": "<what state the work is in>",
  "next_step": "<the single next action>",
  "last_verified": "<command and its actual result>",
  "meta_json": {"open_questions": ["<anything unresolved>"]},
}))
PY
)"

# Resume
curl -s "$M0/api/brain/thread?project=$PROJECT&kind=compact_checkpoint&limit=1" | python3 -m json.tool

Serverless equivalents, same store:

python3 "$M0S/m0_server.py" write --project "$PROJECT" --kind compact_checkpoint \
  --text "…" --next-step "…" --last-verified "…" \
  --branch "$(git rev-parse --abbrev-ref HEAD 2>/dev/null)" \
  --head-sha "$(git rev-parse --short HEAD 2>/dev/null)"

python3 "$M0S/m0_server.py" read --project "$PROJECT" --kind compact_checkpoint --limit 1

Handing off

Write one before the context window is compacted, before a session ends, and

before switching tools. Do not wait to be asked.

  • Resolve the project key$M0_PROJECT, else the repository or directory name.

The same value the earlier entries used; check with /m0-recall if unsure.

  • Write the four things that a stranger needs:

| Field | What belongs there |

|-------|--------------------|

| text | Where the work stands. What is done, what is half-done, what is broken. Prose, no jargon from this conversation. |

| next_step | The single next action, concrete enough to start on. One action, not a plan. |

| last_verified | The command that was run and what it actually printed. Empty if nothing was verified. |

| meta_json | Anything structured worth keeping: {"open_questions": [...], "files": [...], "pr": 42}. |

  • Include branch and head_sha. Without them a reader cannot tell whether

the checkpoint describes the tree in front of them.

  • Set source_tool to the tool you are running in, so the next reader knows

where the work happened.

  • Confirm the response. deferred: true means the store was busy and the entry

is spooled to a sidecar file: real, readable, and landing on the next drain, but

say so rather than reporting a clean write.

What a good handoff looks like

{
  "kind": "compact_checkpoint",
  "text": "Auth middleware refactor is done and green. The admin router still uses the old wrapper, so admin requests bypass the new nonce check.",
  "next_step": "Replace the old wrapper in routers/admin.py with require_session, then re-run tests/admin.",
  "last_verified": "pytest tests/auth tests/api -q: 214 passed, 0 failed",
  "branch": "feat/auth",
  "head_sha": "9f2c1ab",
  "meta_json": {"open_questions": ["Should legacy tokens keep working past the cutover?"]}
}

The failure mode to avoid is a checkpoint that only makes sense to the session

that wrote it: "continued the refactor, tests mostly fine, next step as

discussed". Assume the reader has nothing but this row.

Resuming

  • Fetch the latest checkpoint for the project (kind=compact_checkpoint,

limit=1). If there is none, fall back to the full thread via /m0-recall.

  • Reconcile it with reality before acting. The checkpoint is a claim from an

earlier moment:

   git rev-parse --abbrev-ref HEAD; git rev-parse --short HEAD; git status --short

Same branch and sha means the description probably still holds. Different, or a

dirty tree, means the world moved — read the newer entries too, and re-run

whatever last_verified claims if you are about to build on it.

  • Tell the user what you found, in three lines: where the work stands, what

was last verified, what the recorded next step is. Then say what you will do.

  • Ask before acting on a stale next step. If the recorded next step no longer

fits the tree, say so and propose the alternative rather than following it

mechanically.

  • Write the next entry as you go, with /m0-remember, so the thread does not

go quiet again until the following handoff.

Crossing tools

The thread is one local store, keyed by project, with source_tool on every row.

So "finish this in the other editor" is just: hand off here, resume there.

  • Use the same project key in both tools. M0_PROJECT in each tool's

environment is the reliable way; a mismatch produces two threads that each look

empty.

  • Set source_tool honestly in both, so a reader can tell which tool made a

claim.

  • Nothing syncs between machines. One machine, one thread.

Automating the handoff

A session-end hook makes a handoff unconditional rather than remembered — see

/m0 hooks for the recipe. It writes a session_end entry with the branch and

sha even if the agent forgot to write a checkpoint, and because a shutdown hook

runs at a bad moment, this is precisely the case the deferred-write path was built

for: a locked store spools the entry instead of dropping it.

A hook cannot know what the work state was, only that the session ended. It is a

floor, not a substitute for an explicit compact_checkpoint.

How to use it

Copy the folder

Take coco-research/m0-handoff from the repository into ~/.claude/skills for personal use, or into .claude/skills inside a project.

Check the name does not clash

The agent identifies a skill by the name field in its header. Two skills with the same name cannot sit side by side — one of them will be ignored.