safe-pause
/safe-pause
Use when the user says "safe pause", "pause safely", "bring this to a stop", "wind down", "stop at the next boundary", or wants a running session and its subagents brought to a resumable stop at a task, milestone or project boundary. Also for "pause now" when work must stop at the last green point.
- Author
- mayank-io ↗
- Repository
- mstack ↗
- Licence
- MIT
- Last repository change
- 12.10.2026
Assessment and security
- Files in this package
- 1
- Scripts
- 0
- HTML comments
- 0
- URLs in SKILL.md
- 0
No signals detected. Repository changed on 12.10.2026, without references to retired models or APIs.
MIT, open: use and modify while retaining author attribution.
What it does and when to use it
Use when the user says "safe pause", "pause safely", "bring this to a stop", "wind down", "stop at the next boundary", or wants a running session and its subagents brought to a resumable stop at a task, milestone or project boundary. Also for "pause now" when work must stop at the last green point.
Triggers from the description
Original author description
Use when the user says "safe pause", "pause safely", "bring this to a stop", "wind down", "stop at the next boundary", or wants a running session and its subagents brought to a resumable stop at a task, milestone or project boundary. Also for "pause now" when work must stop at the last green point.
How it works
- Arguments
- Invariants (every mode)
- Procedure
- Red Flags: stop, you are not paused
SKILL.md preview
---
name: safe-pause
description: Use when the user says "safe pause", "pause safely", "bring this to a stop", "wind down", "stop at the next boundary", or wants a running session and its subagents brought to a resumable stop at a task, milestone or project boundary. Also for "pause now" when work must stop at the last green point.
---
# Safe Pause
Bring this session and every subagent it owns to a stop that `/dev:safe-resume` can continue from without asking what was happening.
**Violating the letter of these rules is violating their spirit.** A pause that leaves one worktree dirty, pushes one branch, or forgets one owed step is not a pause.
## Arguments
`/dev:safe-pause [task|milestone|project] [--now]`. Default `task`.
| Mode | Stop point | While running down |
|------|-----------|--------------------|
| `task` | Each live subagent finishes the unit it is on (the commit, the test run, the review pass), commits, writes its resume memory, exits. Bound ~20 min. `--now`: stop at the last green point instead of finishing the unit. | Session exits once the last subagent has. |
| `milestone` | Agents run to the end of the current milestone as its plan defines it: all its commits, its targeted tests, its review, READY reported. The next milestone is not started. | Session stays up to orchestrate, declines anything new, prints a progress line every 30 min. |
| `project` | The session completes the whole dispatched item: every milestone through LAND READY, and LANDED where a lander session exists. | Nothing new accepted during the run-down. |
## Invariants (every mode)
- No new dispatch or backlog item is taken from the moment the skill starts.
- Peer messages that arrive are answered `PAUSED (<mode>), resumes on MK's word` and logged into RESUME.md under *Peer messages logged*.
- Every worktree this session or its agents own ends with a commit. If the work is not a real commit, make a labelled one: `wip(pause): <what is built> / <what is owed>`.
- Nothing pushed, merged or landed. Never touch a primary main.
- One-full-suite rule: do not start a full suite to "finish". A suite already running may complete.
- Resume memory goes to `~/.local/state/mk-pause/<session-name>/RESUME.md`. Never `/tmp`.
## Procedure
Use `bash` with `set -euo pipefail` and `"${VAR:?}"` for every path variable. `PS` below is `python3 ${CLAUDE_PLUGIN_ROOT}/scripts/pause_state.py` (run `--help` for the subcommands).
1. **Freeze intake.** Record the mode and the wall-clock time. From here on decline new work.
2. **Detect topology, do not assume it.** The session has a coordinator only if its config or memory names a PM, or `ListAgents` shows a peer it has already exchanged messages with. Note the lander the same way. A session that runs independently reports to MK only.
3. **Signal every live subagent** with the mode. In `task` mode each one finishes its unit (or, with `--now`, stops at its last green point), commits, writes its own `RESUME.md` section (branch, worktree, tip, tests passed and owed, review state, exact next command), and exits. In `milestone` and `project` modes keep orchestrating until the mode's stop point, then do the same.
4. **Make every owned worktree clean.** List the session's repos (every repo it or its agents worked in) and the branches it owns. `PS dirty --repo <repo> [--repo <repo2>] --owned <branch> [--owned <branch2>]` lists what is still dirty. In each listed worktree commit the real work or a `wip(pause): <built> / <owed>` commit. Repeat until `dirty` exits 0. The primary checkout is never listed and never committed on: files already there, whoever left them, are recorded as a baseline instead.
5. **Write the resume body** to a scratch file with these sections, in this order:
- `## First message` — the exact text to re-send to the coordinator on resume, or `none`.
- `## First command` — the single exact command to run first on resume.
- `## In-flight` — one bullet per branch: worktree, tip, tests passed, tests owed, review state.
- `## Queue` — remaining work in order.
- `## Blockers` — each with who holds it.
- `## Subagents` — each subagent's section from step 3.
- `## Peer messages logged` — every message acknowledged during the run-down.
6. **Record the state.** `PS write --session <session-name> --session-id <id> --mode <mode> --repo <repo> [--repo <repo2>] --owned <branch> [--owned <branch2>] --body <file> [--coordinator <name>] [--lander <name>]`. Pass every repo the session touched, so resume verifies all of them. Pass `--owned` for each branch (or worktree path) the session owns; without it every non-primary worktree is treated as the session's, including other sessions'. It records each owned worktree's branch and tip, records each primary checkout as a baseline, and refuses while an owned worktree is dirty or an `--owned` name matches nothing. The session name is the one set with `/rename`; the id is `$CLAUDE_SESSION_ID` when set, else the id in this session's scratchpad path.
7. **Pointer in auto-memory.** Add one line to this project's `MEMORY.md`: `- [Paused <session-name>](...) — resume via /dev:safe-resume, state at <RESUME.md path>`.
8. **Report, by topology.** Coordinator present: send the pause report (mode, time, branches with tips, owed steps, RESUME.md path) to the PM, and to the lander if one of the branches is queued there. No coordinator: report to MK in the terminal only; RESUME.md already holds it. Never message a peer this session has never talked to. Never broadcast.
9. **Stop.** `task`: exit when the last subagent has. `milestone`: stop after READY is reported, session stays up. `project`: stop after LAND READY (or LANDED).
## Red Flags: stop, you are not paused
- "I'll push this one branch so it isn't lost" → nothing is pushed. The commit is the safety.
- "The suite will only take a few minutes" → no suite is started to finish.
- "This dispatch is tiny, I'll take it" → intake is frozen.
- "I'll write RESUME.md from memory later" → write it before exiting; `PS write` refuses dirty owned worktrees for this reason.
- "The tool refused, so I'll record a different repo instead" → no. Pass every repo with `--repo` and the session's branches with `--owned`; a refusal names an owned worktree to commit.
- "The PM probably wants to know" when no PM has been talked to → report to MK only.
Versions
The database has no other skills named safe-pause.
Database snapshot: 13.11.2026 · package version 9736e5c4ed69 (SHA-256 digest of the entire package)
Downloads in the last 30 days: 0 · ranking
Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.
Similar skills
- 90safe-resume
Use when the user says "safe resume", "resume from pause", "pick up where we left off", "continue the paused work", or when a session start hook reports a /dev:safe-pause state for this session. Also use when a fresh session is asked to continue and a RESUME.md exists for it. - 100make-camp
Appa's camp for pausing work safely. When the team sets down for the night, they leave camp so anyone can pick up in the morning. Use for pause-safely, "pause here", "stop for now", "save your place", "checkpoint this", going offline, a session restart, or imminent context… - 90pause-or-stop
Pause, resume, archive or stop HeyLead LinkedIn outreach, for one campaign, one person or everything at once. Use when the user says stop, pause, hold on, too many messages, or wants a person never contacted again. - 85session-continuity
Resume a prior session, reconstruct recent project context, pause safely on request, hand off to a fresh agent, audit a decision trail, or catch the user up in text or video. Preserve existing evidence and the exact authorized resume point. Use for "pick up where we left off",… - 86caretaker-pause
Pause or resume the caretaker's cloud tasks from a session (D74). Use when the user says "pause my caretaker", "stop the caretaker", "turn the caretaker back on", "is the caretaker running?", or asks how much Claude it uses. - 100ctx-pause
Pause context hooks for this session. Use when context nudges aren't needed for the current task.