agentleFS
Sign inStart free
LuisUrrutiaLuisUrrutia/skills/handoff/SKILL.md

handoff

Skill2 starsChanged 15 days agoMITView on GitHub

Use when writing a handoff document so another agent or session can continue a task.

We split this skill into the rules it gives an agent

7 rules. 0 are common, 6 are distinctive, and 1 is very specific to this project.

  • 0

    Common

    Most of the 171 edited versions we read keep these.

  • 6

    Distinctive

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Write a compact handoff document summarizing the current conversation and work state so a fresh agent can continue: in another session, …

    line 8
  2. Distinctive:

    Write and save only the document.

    line 12
  3. Distinctive:

    Use repository-relative paths for repository files.

    line 25
  4. Distinctive:

    If absent from the skill catalog, write without it.

    line 45
  5. Very specific:

    Use the destination named by the user's request, including one forwarded by a caller.

    line 57
  6. Distinctive:

    Check each referenced local file exists.

    line 67
  7. Distinctive:

    When pickup is limited, exactly what to copy or fetch: the document and each unreachable local file, including required Git state (branch, …

    line 76

“Common” is measured: kept by at least 6 in 10 of the 171 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Handoff

Write a compact handoff document summarizing the current conversation and work state so a fresh agent can continue: in another session, host or directory, through a colleague, or on a parallel side task while this session continues.

Write and save only the document. Launching or seeding sessions, pausing, agent cancellation, commits, publication, worktree changes and continuation retain their owners and authorization. Saving proves neither receipt nor ownership transfer.

Choose what travels

Focus on what the request or caller says the next session will do; otherwise on continuing the current task. Keep only relevant context a fresh agent cannot infer from accessible material. Include conversation-only context: the recipient cannot open this conversation.

Reference existing artifacts by path or URL with what to read; do not copy their contents, including checklists. Use repository-relative paths for repository files; label absolute paths as machine-local.

Write the document

Open with the next action and why; the rest supports it. Record checkable facts with their basis; label inferences and unverified beliefs:

  • Objective, scope and completion criteria as the user last directed.
  • Done, partial and not started work, with dependencies.
  • Each decision labeled user-confirmed, agent-proposed or assumed, with reasons; keep unapproved distinct from rejected.
  • Each check with command and observed outcome, including historical results, or not run. No rerun required.
  • Blockers and what clears them; approaches likely to be retried and why abandoned.
  • For repository work, branch, revision, staged and uncommitted changes, separating task changes from other work to preserve.

Separate output requirements from tool permissions. Use agent-instructions for continuation guidance: next action, initial reads and user-set execution limits. If absent from the skill catalog, write without it. Handoff documents never trigger its two-model review, even for guidance aimed at another agent: a handoff request is not instruction authoring.

Include a "Suggested skills" section naming, by registered name, skills the next agent should use; the receiving environment resolves and invokes them.

Redact any sensitive information, such as API keys, passwords, and unrelated personal data.

Save and report

Use the destination named by the user's request, including one forwarded by a caller; a caller-chosen location is not user-named. Otherwise save to the temporary directory of the user's OS ($TMPDIR, else /tmp; %TEMP% on Windows), not the current workspace or repository. Create a new file; overwrite only when the request names it for update.

Even if temporary storage cannot serve pickup on another machine or container, or after cleanup, save there. Create no bundles, patches, archives or other copies of the work, even temporarily. Leave any preservation destination to the user.

Check each referenced local file exists; label missing ones. After all edits, including verification fixes, read the whole saved file into context as a fresh agent: confirm the next action, why, and what to avoid where limits or abandoned approaches exist.

Reply with:

  • Exact saved path; next task in one line.
  • Any access or retention limits and their cause, not only preservation advice.
  • When pickup is limited, exactly what to copy or fetch: the document and each unreachable local file, including required Git state (branch, commits, staged task changes).

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

Reviews can't be read right now.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

Check your setup, free