Install
openclaw skills install @ciklopentan/long-task-context-gptResume multi-step work from verified checkpoints
openclaw skills install @ciklopentan/long-task-context-gptUse this workflow when project decisions or progress need to survive a later stage or session, or when the user asks to resume or hand work off. Skip it for self-contained one-shot tasks. Respond in the user's language.
A checkpoint is a compact, source-linked navigation note—not a transcript, source of truth, larger context window, or automatic access to past chats.
Continue with a useful next action; do not equate activity with completion. Update the same checkpoint after a verified milestone, material decision or constraint change, new blocker, or before pausing or handing work off.
Keep it brief and actionable. Include only durable information needed to continue:
Do not copy chat transcripts, repeat the same sources in a separate list when they are already cited inline, or keep detailed history in the current-state note. Put history in the project's existing log and link to it. Preserve exact identifiers, units, revisions, and test status when they affect a decision; never shorten away an acceptance condition or caveat.
Before a meaningful pause, save and read the checkpoint back. Using only that note and its source references, a fresh session should be able to identify the goal and completion criteria, what is done and outstanding, the key constraints or unknowns, and the next action. If not, add only the missing decision-critical context. Skip this check when no checkpoint is warranted; do not inflate a note just to fill a template.