菜单
Prepare or receive a professional XMemo handoff between Claude sessions or compatible AI agents such as Claude, Codex, GitHub Copilot, or Kiro. Use when switching agents, transferring unfinished work, reconciling a received handoff, or preserving a restart-ready package with provenance, evidence, blockers, and one exact next action.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
Audit implementation or project progress against the approved XMemo plan, checkpoint, decisions, and current evidence. Use when Claude must verify another agent's completion report, reconcile multi-agent claims, identify drift or blockers, and determine the exact next action. Never promote a claim to verified state without evidence.
Run a structured XMemo-grounded brainstorm that recalls prior goals, constraints, decisions, and failed approaches; generates distinct options; tests them; and converges on a user-approved direction. Use when exploring ideas, product directions, architectures, strategies, or plans with Claude. Keep proposals tentative until the user approves them, and save only durable outcomes.
Turn an important Claude conversation or work session into concise, retrieval-ready XMemo outcomes. Use when summarizing key points, preserving decisions and rationale, extracting TODOs and unresolved questions, or saving unfinished progress. Do not archive raw transcripts, filler, secrets, or tentative ideas as confirmed facts.
Use XMemo to recall and preserve durable project context, decisions, plans, TODOs, checkpoints, and cross-agent handoffs. Use when resuming work, distilling an important conversation, reviewing a brainstorm or implementation plan against saved context, recording progress or blockers, handing work between Claude and another compatible agent, or managing memory lifecycle actions. Never save raw transcripts, secrets, or tentative ideas as confirmed decisions.
Design or revise an evidence-gated project plan and executable Markdown plan using XMemo context and the live workspace. Use when starting a project, turning goals or a brainstorm into PROJECT_PLAN.md and EXECUTION_PLAN.md, decomposing work into phases and acceptance gates, assigning multi-agent ownership, or correcting a plan after review. Keep proposals distinct from approved decisions and implemented progress.
Resume a project or unfinished task from XMemo by recovering the latest verified checkpoint, relevant decisions, TODOs, blockers, and exact next action. Use at the start of a new Claude session, after an interruption, or when continuing work previously handled by another compatible agent. Reconcile saved state with current evidence before acting.
| name | handoff-work |
| description | Prepare or receive a professional XMemo handoff between Claude sessions or compatible AI agents such as Claude, Codex, GitHub Copilot, or Kiro. Use when switching agents, transferring unfinished work, reconciling a received handoff, or preserving a restart-ready package with provenance, evidence, blockers, and one exact next action. |
Transfer verified continuity, not confidence or chat history. Follow scope → verify → package → persist → receive → acknowledge.
Identify the objective, source, recipient, project or scope, artifacts, handoff direction, and whether work is paused or complete. Never assume the recipient exposes the same tools or can access local-only artifacts.
Recover the latest context with recall_context or get_project_context, then compare it with current files, diffs, tests, deployment state, or user confirmation. Treat another agent's completion statement as a claim until evidence supports it.
Include:
Exclude credentials, authorization data, private traces, raw logs, raw transcripts, and irrelevant internal identifiers.
update_state;record_event;todo only for a concrete follow-up;remember or update_memory;create_pending_decision.Preview a multi-record handoff before writing. Return a receipt with the records changed and what remains unresolved.
When receiving, recall the relevant saved state, compare the incoming claims with current evidence, and classify conflicts by scope, recency, provenance, and verification. Do not overwrite newer verified state with an older handoff. Produce an acceptance response: accepted, accepted with gaps, blocked, or rejected, plus the exact next action.
XMemo can preserve shared context only for agents connected to the same authorized account and scope. It does not guarantee that another agent saved its work, exposes the same tool surface, or can reopen the original conversation.
Read ../memory-steward/references/workflows.md, ../memory-steward/references/review-playbooks.md, ../memory-steward/references/memory-policy.md, and ../memory-steward/references/tool-routing.md for detailed handoff and reconciliation rules.