메뉴
Write a detailed handover note so a fresh Claude Code session can pick up cognitively dense work without context loss. Captures decisions, rationale, discoveries, and next steps.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
| name | handover |
| description | Write a detailed handover note so a fresh Claude Code session can pick up cognitively dense work without context loss. Captures decisions, rationale, discoveries, and next steps. |
Write a detailed handover note for a fresh Claude Code session to pick up this work without context loss. This is for cognitively dense work where decisions, rationale, and discoveries are expensive to re-derive.
Review the full conversation — scan all turns, not just recent ones. Identify decisions, pivots, corrections, discoveries, and current state.
Write the handover note to docs/current/handover/ as a markdown file. Use the current date and a short slug for the filename (e.g. 2-Apr-26-cohort-forecast-conditioning.md). Use d-MMM-yy date format per project conventions.
Structure the note using these sections (all required):
What we are trying to achieve and why. Not just the task name — the underlying goal, the problem being solved, and any constraints or scope boundaries agreed with the user.
What is done, what is partially done, what is untouched. Be specific — name files changed, tests written, features working/broken. Use a status indicator for each item: DONE / IN PROGRESS / NOT STARTED / BLOCKED.
Decisions made during the session that a fresh agent would not derive from reading the code alone. For each decision, state:
This is the most important section. A fresh agent will re-derive naive solutions unless told what was already considered and rejected.
Things learned during the session that are not obvious from the codebase:
A map of the files that matter for this work, with a one-line description of each file's role. Group by area (e.g. "Backend", "Frontend", "Tests", "Docs"). Include files that were read for context, not just files that were changed.
Concrete, ordered, actionable steps for the next session. Each step should be specific enough that a fresh agent can execute it without asking clarifying questions. Include:
Unresolved questions that need the user's input or further investigation. Flag whether each is blocking or non-blocking.
cohort_forecast.py:245 — rows and columns are swapped when n_periods > horizon" is useful.Capture a complete snapshot of the current working tree (modified, deleted, AND untracked files) into a named git stash, then restore the working tree exactly as it was — same files modified, same files deleted, same files untracked, same staged/unstaged layout. Result is a reusable safety net the user can `git stash apply` later if they need to recover, while in the meantime nothing has changed. Triggered when the user types `/photocopy [name]` or asks to "photocopy", "snapshot", or "safety-stash" the current working dir without disturbing it.
Self-administering stage runner for multi-stage implementation plans. Picks the first incomplete stage from the plan's progress section, executes it with strict discipline (briefing receipt, atoms, commits, tests, schema parity), records completion in the plan, and returns control. Use when the user wants to advance a plan one stage at a time without manually specifying which stage. Triggered by `/implement-carefully <plan-path>`, `/implement-carefully warm-up <plan-path>` (preparation only — briefing receipt and recon report, then halts before any edits), or by the user asking to "run the next stage", "advance the plan", or "do the next bit of <plan>".
Load disciplined per-stage execution instructions for working through ONE stage of a multi-stage implementation plan. Use when the user is starting a session to implement a specific stage and wants strict stage-by-stage discipline (bounded scope, deliberate context budget, explicit stop conditions, briefing-receipt compliance, schema parity, no scope creep). Triggered by `/dispatch-stage <stage-id> [<plan-path>] [<companion-paths>...]` or by the user explicitly asking to "dispatch a stage agent" or "load the stage skill".
Reframe a comment, question, or passage to be succinct, lucid, and free of jargon — for when something the user or someone else has written is hard to follow. Also applies when the agent is asking the user for direction: the question must give enough context that an intelligent reader can actually respond, framed as a senior engineer seeking strategic guidance rather than a junior dev throwing tactical minutiae back. Explains or writes around project-specific terms so an outsider could understand. Triggered when the user types `/huh`, "huh?", "eh?", "what?", or similar confusion cues about a piece of text.
Review the full conversation and ensure all decisions, designs, and knowledge are properly documented in codebase docs and/or public docs. Creates or updates docs as needed.