Home / All Skills / Workflow & Core Tools / handoff
English (EN)简体中文 (ZH)日本語 (JA)
Workflow & Core Tools · mattpocock/skills

handoff

Compact the current conversation into a handoff document for another agent to pick up.

L2 Local · Restricted workspaceV1 Commit-PinnedWrite FileDual-pass Audit
Author mattpocockGitHub mattpocock/skills ↗Stars 253948License MITCommit 3cca18b368
Control levels describe skill capabilities and suggested control measures; they do not represent quality or safety certifications, nor do they dictate usability. Usage decisions rest entirely with users and administrators. Different agent hosts apply varying constraints; do not assume hosts provide automatic protection. Third-party skill copyrights belong to their original authors. Use at your own risk.

1How It Works & Mechanism

Pure prompt orchestration, zero code: Functionality is executed entirely by the host model—reading the current conversation, summarizing it, and writing the document to the OS temporary directory as instructed. Other than the main instructions in SKILL.md and metadata in agents/openai.yaml, there are no scripts, templates, or referenced files in the directory.

productivity/handoff/SKILL.md
Write a handoff document summarising the current conversation so a fresh agent can continue the work. Save to the temporary directory of the user's OS - not the current workspace.
The entire text consists of only 5 imperative instructions (confirmed byte-by-byte via cat -A to have no code blocks or hidden content). Generating a 'document summarising the current conversation' relies on the host model's summarization capabilities; the skill itself only enforces behavioral constraints.

Output location discipline: Deliberately writes to the OS temporary directory rather than the current workspace to avoid mixing the handoff document into the user's repository or project directory.

productivity/handoff/SKILL.md
Save to the temporary directory of the user's OS - not the current workspace.
What it does: Isolates output side-effects to the temporary directory to prevent repo pollution; however, the filename/subpath is unspecified and left to the host model's discretion.

Anti-duplication design: Only references paths/URLs for existing artifacts (specs/plans/ADRs/issues/commits/diffs) without copying content, keeping the handoff compact and preventing drift from the source artifacts.

productivity/handoff/SKILL.md
Do not duplicate content already captured in other artifacts (specs, plans, ADRs, issues, commits, diffs). Reference them by path or URL instead.
The handoff document is positioned as a 'set of pointers + summary' rather than a 'content snapshot'; the next agent reads the original artifacts back as needed.

Parameterized customization: The user-provided argument (prompted by argument-hint: 'What will the next session be used for?') is treated as the purpose description for the next session, tailoring the focus of the document accordingly.

productivity/handoff/SKILL.md
If the user passed arguments, treat them as a description of what the next session will focus on and tailor the doc accordingly.
argument-hint only appears in the front metadata as a trigger prompt for the host; there is no schema constraint, and the argument is free-form text.

Dual-ecosystem explicit invocation: Both Claude's disable-model-invocation: true and OpenAI's allow_implicit_invocation: false prohibit automatic invocation by the model, running only when explicitly initiated by the user.

productivity/handoff/agents/openai.yaml
policy: allow_implicit_invocation: false
The frontmatter also includes 'disable-model-invocation: true'. Handoff is an intentional wrap-up action by the user and does not allow implicit triggering.

2Core Capabilities

01Compress the current conversation into a handoff document for a new agent to take over
02Write documents to the OS temporary directory, keeping them out of the workspace
03Generate a 'suggested skills' section specifying the Skills the next agent should invoke
04Reference existing artifacts by path/URL instead of duplicating them
05Sanitize sensitive information (API keys/passwords/PII)
06Tailor document focus based on user arguments

4Recommended Governance & Evidence

Suggested Controls for Safe Operation
For Agent Runtime
Restrict working directory
For Administrators
Retain operation logs
Evidence & Triggers (File Paths & Excerpts)
  • 本地工作目录: File read/write and artifact output: A handoff document is written to the user's OS temporary directory (instruction explicitly states: 'Save to the temporary directory of the use
Audit Notes: No script execution, no outbound network calls, no credential reading; the sole behavioral surface is writing a document to the OS temporary directory (local write, explicitly excluding the workspace)—classified under the 'Local Read/Write Only' tier according to the six-tier paradigm. The only consideration is that temp files are visible to other processes on the same machine and lack cleanup instructions.

5Second-Pass Verification

  • [ok] 外部依赖：无（external_deps 为空） — 第二遍逐词重扫两文件正文：无任何 CLI/API/网络端点/第三方包引用；唯一提及的 'Skill tool' 是宿主自身工具而非外部依赖
  • [ok] 无脚本/无可执行代码 — find productivity/handoff -type f 仅返回 SKILL.md 与 agents/openai.yaml；cat -A 显示正文为 5 段纯文本，无代码围栏
  • [ok] 文件写入：仅 OS 临时目录 — quote 'Save to the temporary directory of the user's OS - not the current workspace.' 逐字节核实；无其他写路径指令
  • [ok] 无网络调用/凭证读取 — 正文无 curl/gh/API 字样；未引用任何 env/keychain/配置文件
  • [ok] 触发方式：仅显式 — SKILL.md frontmatter 'disable-model-invocation: true' 与 openai.yaml 'allow_implicit_invocation: false' 两处独立证据一致
  • [ok] 功能声明 vs 实际能力相符性 — description 'Compact the current conversation into a handoff document' 与正文指令一致；无夸大——能力边界即宿主模型的总结+写文件
  • [ok] 元数据：license/stars/last_push — GitHub API：MIT、stargazers_count 253948、pushed_at 2026-09-04（与 pin commit 日期吻合）

6Verdict & Scenarios

  • Zero side-effect surface: No scripts, no network, no credentials; the sole action is writing a temporary document
  • Disciplined output: Explicitly requires writing to the OS temporary directory 'not the current workspace', avoiding repository pollution
  • Anti-duplication design: References existing artifacts by path/URL instead of copying them, keeping documents concise and preventing drift from source artifacts
  • Built-in privacy instructions: Explicitly mandates redacting API keys/passwords/PII
  • Controlled triggering: Metadata in both Claude and OpenAI ecosystems disables automatic invocation, running only when explicitly requested by the user
    • Best for: Individual developers needing to hand off complex work across sessions/agents (such as Matt Pocock-style workflows); teams looking to standardize how 'the next agent takes over'; scenarios where handoff documents are preferred to stay in the temp directory rather than being committed to the repository.
    Not for: Teams requiring a fixed handoff template or structured schema (this skill imposes no format constraints and is pure free-form text); workflows requiring automated handoff triggers (implicit invocation is disabled); scenarios needing handoff documents to be committed to a repository/shared location, or requiring document lifecycle (deletion/archival) management.
    Install Agent-Ready
    ① Verified Mirror 2026-09-06
    Option A · Direct DownloadDownload handoff.tar.gz
    sha256: fa26a009eb0d810a…
    Option B · Copy Guide for Agent
    install.json
    Agents automatically download and verify mirrors from this JSON guide.
    ② Upstream GitHub
    GitHub Repo ↗
    Pinned to commit 3cca18b368.