Skip to content

TsukiNi22/ai-utils / skills/ai/pause

Stop everything cleanly in the middle of a task so that it can go on later - no new step started, the step under way brought to a consistent state (never a file half edited or a build left broken), every background job / monitor / subagent / workflow / build / loop / scheduled wake-up started for the task stopped, nothing committed or pushed that the rules of the user do not allow, then a checkpoint written (goal, rules, done, interrupted, the exact next steps, what was stopped and how to start it again, files changed, decisions) and one short message - the work resumes at the next "continue" (continue-work). Use whenever the user says /pause, "stop tout proprement", "arrête-toi proprement", "mets en pause", "on reprendra plus tard", "stop et reprends au prochain continue", "pause the work", "stop cleanly".

Pause: stop cleanly, resume at the next "continue"

SKILL_DIR = the directory of this file. The user wants the work stopped now, cleanly, and taken up again later by a plain continue (the continue-work skill), in this session or a later one, without re-explaining anything. It is not a handover to another machine or model (that is export), and not the end of the task.

1. Stop starting things

  • From now on, start no new step: no new edit, command, build, agent or question.
  • The step under way is brought to the nearest consistent state, the shortest way:
    • a file edit in progress: finish it so the file is valid (syntax, matching blocks), or put the file back as it was before the step; never leave a file half written;
    • a change spread over several files (a rename, an API change): finish it if it is a few edits, otherwise revert the part of this step and note it in the checkpoint as not started;
    • a long command (build, test run, download): stop it (below) rather than wait, unless it ends within a minute.
  • No question to the user (the popup included): what would need an answer goes in the checkpoint.

2. Stop what runs for the task

Only what this session started for the task, never a process of the user:

  • background shell commands and monitors: TaskStop each one still running;
  • subagents, forks, workflows: TaskStop (a workflow keeps its run id: note it, it can be resumed);
  • a /loop running: ScheduleWakeup with stop: true; crons created for the task: CronDelete;
  • processes it launched outside the task tools (a server, a build in another folder): stop them by their exact command line (pgrep -fa to check first), never a broad pkill;
  • a temporary branch or worktree it made: leave it, write its path in the checkpoint (never delete work). For each one, note how to start it again (the command), or that it is not needed any more.

3. Leave the repository as the rules say

  • Commit or push only what the rules of the user already allow for this repository (a project rule granting commits after each batch: commit the finished part, never a half step). Otherwise nothing is committed: the changes stay in the working tree and are listed in the checkpoint.
  • Nothing destructive to "clean up" (no git reset, no deleted file the user did not ask to delete).

4. Write the checkpoint

bash SKILL_DIR/scripts/checkpoint.sh path      # the file of this project (~/.claude/pauses/<project>.md)

Fill templates/checkpoint.md into that file (one per project, a new pause replaces the old one), terse like rtk (facts, one per line, no prose), every path and command exact:

  • the goal in the user's words, the rules that apply (git, language, style, what needs a confirmation);
  • done (verified how), interrupted (its exact state), next in order - the first item is what continue runs first;
  • what was stopped and how to restart it, the files changed and not committed, the temp files, decisions, open points. A secret is never written in it (a token, a key, a password: say where it comes from instead).

5. Answer and stop

One short message in the user's language, then no more tool call:

  • En pause. + what was stopped (one line), the state of the step under way, where the checkpoint is;
  • Dis « continue » pour reprendre à : <next step>. Other checkpoints waiting on this computer: checkpoint.sh list.
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 464kUpdated Claude CodeCodexVerifiedInstall
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 464kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 318kUpdated Claude CodeCodexUnverifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 318kUpdated Claude CodeCodexUnverifiedInstall