원클릭으로
handoff
Create a durable repo-local handoff document with a portable prompt for a fresh agent.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
메뉴
Create a durable repo-local handoff document with a portable prompt for a fresh agent.
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
Improve, audit, simplify, rewrite, or migrate prompts, tool descriptions, agent instructions, and prompt stacks for GPT-5.6 Sol or the GPT-5.6 family. Use for outcome-first prompt design, autonomy boundaries, tool routing, PTC, grounding, verbosity, reasoning effort, and prompt evals.
Apply a deep, design-minded engineering workflow for complex work that needs careful planning, iteration, and simplification.
Use when preparing or running a Claude CLI agentic engineering review through claude -p for repository audits, architecture analysis, system design, security review, large refactors, specs, RFCs, or evidence-grounded long-form engineering analysis.
Use when preparing or running a ZCode GLM-5.2 agentic builder workflow through zcode --prompt for fast implementation, iterative code generation, focused fixes, and validation on an existing repository.
Apply adversarial opposite-analysis to plans, specs, architecture, code changes, and claims. Use when the user asks for adversarial review, opposing analysis, contrarian review, red-team reasoning, or Z.ai and MiniMax cross-checks through the Ralph MCP router.
Review and adjudicate Bugbot, Cursor, Seer, and similar automated PR feedback with local evidence before accepting, fixing, or dismissing findings.
| name | handoff |
| description | Create a durable repo-local handoff document with a portable prompt for a fresh agent. |
| argument-hint | What should the next session focus on? |
Use this skill when the user asks for $handoff, asks to write a handoff, wants
continuity before pausing work, or wants another agent to pick up a task.
Write a concise Markdown handoff that is useful without the old chat history. The handoff is a durable local artifact and must include a standalone prompt for the next agent. The next agent should treat the handoff as starting context, not as authority.
Save the handoff under .local-notes/ at the root of the repository where the
agent is currently working:
git rev-parse --show-toplevel.<repo-root>/.local-notes/ if it does not already exist.handoff-YYYY-MM-DD-HHMMSS.md in that directory.If a project has a stricter local-notes policy in its instructions, follow that policy's canonical repo root. For example, a secondary worktree may need to save the canonical copy under the primary checkout.
If the user passed arguments, treat them as the next session focus and tailor the handoff around that focus. If the user gives only a short label, infer the task conservatively from the current repo, recent discussion, branch, linked issue or PR, plans, docs, and obvious nearby context. If the focus is still ambiguous after local inspection, ask before writing.
Gather enough context for a fresh agent to orient: repo or product identity, branch and PR or issue anchors when relevant, modules likely involved, constraints, current status, known symptoms, and expected validation. Do not do the receiving agent's full independent review for them, and do not decide the final technical direction on their behalf.
Use this structure unless the project instructions require a stricter format:
# Handoff: <short title>
## Next Session Focus
<one or two paragraphs>
## Current State
<brief state of branch, worktree, PR/issue, runtime, or plan>
## Verified This Turn
- <facts actually checked in this session>
## From Conversation Only
- <useful context not independently rechecked>
## Needs Re-check
- <state likely to drift: branch, CI, deploy, live runtime, dependencies, dates>
## Relevant Artifacts
- <repo-relative paths, issue/PR URLs, plan paths, docs, commits, commands>
## Decisions And Constraints
- <decisions already made, user instructions, non-goals, ownership boundaries>
## Suggested Skills
- <skills the next agent should invoke, with a short reason>
## Next Agent Prompt
<standalone prompt following the rules below>
## Validation Expected
<focused checks, tests, live proof, or explicit no-heavy-validation boundary>
## Open Questions And Risks
<blockers, unknowns, and what should stop the work>
Keep the document compact. Do not duplicate content already captured in PRDs, plans, ADRs, issues, commits, diffs, or previous handoffs. Reference those artifacts by path, URL, commit, issue or PR number, command name, config key, or search term.
The Next Agent Prompt section must be portable enough to paste into a fresh
agent session. It must start a discussion and review, not a command-only work
order.
The prompt should ask the receiving agent to:
When relevant, tell the next agent to re-check live repo, GitHub, CI, deploy, or runtime state before trusting the handoff. Tell the next agent not to push, merge, close issues or PRs, label, post public comments, mutate external state, or run destructive local commands unless explicitly instructed.
Inside the handoff document, prefer repo-relative paths for code, docs, tests,
plans, and local notes. Use URLs for PRs, issues, specs, or external references.
Use absolute paths only when they materially identify local-only state such as a
canonical checkout, a secondary worktree, .local-notes/, a local runtime, or a
handoff destination.
Inside the Next Agent Prompt, avoid machine-specific absolute paths and
checkout names unless the user explicitly needs that local context. Prefer
portable anchors: repo owner/name, product or module names, issue or PR URLs,
branch names, package names, public symbols, command names, config keys, exact
error text, docs titles, and search terms.
Do not persist RED content. Follow the project's sensitivity policy instead of copying raw sensitive material into the handoff. Use sanitized symptoms, IDs, hashes, counts, and file names when that gives the next agent enough context.
If the task itself is sensitive, keep the handoff local, mark the sensitivity boundary clearly, and do not recommend external MCPs or web tools for RED material.
After writing the handoff, reply tersely with the file path, the next-session focus, the suggested skills, and unresolved open questions. Do not paste the full handoff unless the user asks for it. Do not copy to the clipboard unless the user explicitly requests a clipboard-ready prompt.