Index of every page on this site: llms.txt
All skills
openclaw avatar

/handoff

@0a5ba4b
by openclawopenclaw/agent-skills1.1k stars
126
Run it yourself

Clipboard-ready handoff prompt for another agent to investigate or continue a task.

Use this Skill: https://skilld.dev/gh/openclaw/agent-skills/handoff

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

SKILL.md

≈23 tokens for metadata: the name and description. ≈1.2k when used: this file.

Description uses 1.2% of example budget

Before choosing a Skill, your Agent reads its name and description. All available Skills share that space.

  • A shorter description leaves more room for other Skills. This entry exceeds our 1% size suggestion.

In our Claude Code example, all Skill names and descriptions share 8,000 characters. This Skill uses ≈93 characters, or 1.2%.

The 1% threshold is a size suggestion. Longer descriptions can still fit.

Your model, settings, and other Skills decide how much text your Agent can read.

Example settings and source

The example uses a 200k-token context and default Claude Code settings. The count includes the name, description, separators, and when_to_use when present. Codex also counts local file paths.

Skit's source and limits: Codex 0.160.1, Claude Code 2.1.292.

Handoff

Write a clipboard-ready prompt for another agent to investigate, discuss, or work on a specific task.

Use when the user asks for handoff <task>, "write a handoff", "delegate this", or wants a prompt for another agent.

Workflow

  1. Identify the task from the user text. If the user gives only a short label, infer from the current repo, recent discussion, branch name, linked issue/PR, docs, and obvious nearby context.
  2. Gather enough context to write a useful handoff: repo/product identity, relevant issue/PR/branch names, likely modules, constraints, and known symptoms. Do not perform the receiving agent's full independent review or decide the final technical direction for them.
  3. Write a standalone prompt preserving the requested mode, existing authority, accepted decisions, completed proof, remaining work, and completion condition.
  4. Copy the full prompt to the clipboard.
  5. Final reply: terse confirmation with the task title. Do not paste the full prompt unless the user asks.

Handoff Prompt Rules

The prompt must:

  • Preserve the user's requested mode: investigation, discussion, or execution.
  • Carry existing authorization and its limits without granting new authority.
  • Ask the receiving agent to verify live state and unresolved assumptions. Size independent review to the remaining uncertainty and repository requirements.
  • Preserve accepted decisions and completed work unless new evidence invalidates them. An execution handoff must not reset the task to a proposal.
  • Assume the agent starts in the repo, a parent directory, a workspace directory, or a home directory and can find the repo itself.
  • Avoid filesystem paths. No absolute paths, home-directory paths, checkout names, or repo-relative file paths unless the user explicitly requests them.
  • Use portable anchors instead: repo owner/name, product/module names, issue/PR URLs, branch names, package/plugin names, public symbols, command names, config keys, exact error text, docs titles, and search terms.
  • Include enough context for the receiving agent to get the right repo, boundary, and desired outcome.
  • Include constraints, non-goals, validation expectations, and the desired output shape.
  • Tell the receiving agent to re-check live repo/GitHub/CI state where relevant.
  • Tell the receiving agent not to push, merge, close issues/PRs, label, or post public comments unless the handoff explicitly asks for it.

Prompt Template

Adapt this shape to the requested mode:

<Investigate, discuss, or continue execution of>: <short task title>

Context:
- <portable repo/product context>
- <what triggered this task>
- <known current state, branch/issue/PR names or URLs if relevant>
- <important constraints and ownership boundaries>
- <existing authority, accepted decisions, completed work, and remaining gaps>

Resume:
- Find the right repository from the current directory, a parent directory, or the usual workspace.
- Read the local agent/repo instructions.
- Inspect the relevant code, docs, tests, recent commits, and linked issue/PR state.
- Verify unresolved assumptions without repeating completed work or reopening settled decisions absent new evidence.
- Call out stale assumptions, hidden risks, and anything that should stop the work.

Task:
- <next investigation or already-authorized implementation>
- <expected behavior or decision criteria>
- <non-goals>
- <completion condition and concrete stop conditions>

Validation:
- <focused tests/checks/live proof expected>
- <what evidence should be included>
- <what is explicitly not required>

Output:
- For investigation, report findings and recommendation.
- For execution, complete the authorized work and report the result or precise blocker.
- If you edit code, keep changes scoped and report exact proof run.
- Do not push, merge, close issues/PRs, label, or post public comments unless explicitly told.

Clipboard

On macOS:

pbcopy < /tmp/handoff-prompt.txt

Use a temp file or pipe. Avoid inline shell quoting for prompts containing backticks, $, quotes, or user text.

If pbcopy is unavailable, use the obvious platform clipboard tool (wl-copy, xclip, clip.exe) or print the prompt and say clipboard copy was unavailable.

Quality Bar

  • No invented facts. Mark reviewed facts as such only after checking them.
  • No path leakage. Rewrite any accidental path as a symbol, module, command, issue/PR URL, or search term.
  • Enough context for a fresh agent to orient; no giant brain dump.
  • A fresh agent can continue in the requested mode without inventing authority or losing existing authorization.

Source: SKILL.md on GitHub

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

1 warning15d3 checks · Risk SAFE
  • Gen Agent Trust Hub17d

    The skill is a utility for generating structured handoff prompts for task delegation. It includes proactive security measures to prevent path leakage and avoid command injection during clipboard operations.

  • Socket15d

    No alerts

  • Snyk17d

    Risk: MEDIUM · 1 issue

Signed by skilld at 0a5ba4b. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub yesterday.

Activeupdated 3 weeks ago

README badge

README badge for openclaw/agent-skills/handoff