agentsclimarketplace

Handoff

Skill cskwork/agent-handoff/handoff

A Claude Code / Codex skill for long-running tasks: writes a structured handoff doc so a fresh-context agent resumes with zero ambiguity. Docs: https://cskwork.github.io/agent-handoff/

Install
npx -y skills add cskwork/agent-handoff --skill handoff

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Writes a structured markdown handoff file so a fresh-context agent can resume a long-running task — objective, progress, key decisions, the exact next step, and how to verify. Use near the context limit, at a milestone or session end, or when asked to create a handoff / save state / "핸드오프 만들어".

SKILL.md

3.7 KB, 825 tokens by cl100k_base, as published. Nobody here has run it

handoff

Write the document your next agent opens before it works. A handoff compacts everything that matters about a long-running task into one structured markdown file, so an agent with an empty context window resumes exactly where you left off — without re-reading the conversation.

This skill always writes a markdown file as the durable cross-agent transport, and also emits a paste-ready resume block you can drop into a fresh chat.

When to use

Trigger on any of: context window ~80%+ full · a milestone or work session ending · after 5+ file edits · the user asks for a handoff / state save / "next plan of action".

Manual trigger (user present, steering): just write the file. Automatic/threshold trigger (context near full): also append the auto-continue note (step 7).

Pipeline (summary)

gather → structure → write → verify → resume. Full detail in ref/pipeline.md. Do NOT continue the task; produce only the artifact.

  1. Preflight. If essentially nothing happened, say "Nothing to hand off yet" and stop — never fabricate.
  2. Gather mechanically. Derive read/modified sets from tool history; capture commands, test output, and error text verbatim; confirm git branch/HEAD/dirty via read-only probes. Never trust recall.
  3. Preserve the open thread. Capture any unanswered user question VERBATIM (the most-failed item) → Open Questions, and Current State if it blocks the next step.
  4. Structure. Fill ref/templates/handoff.md in order: Current State first; Done labeled "DO NOT REDO" + quantified; ONE next step; decisions with rationale; concrete file:line / commands / real errors.
  5. Self-verify. Run ref/checklists/quality.md; fix any secret, placeholder, vagueness, missing verify command, or dropped open question before writing.
  6. Write. Save handoffs/handoff-<UTC>.md in the working project (ISO 8601, :/. → -); chain and merge a prior handoff if one exists.
  7. Emit + report. Print the path + a one-line summary, then the doc wrapped in <handoff-context>…</handoff-context> (exact wrapper in ref/pipeline.md) as the paste-ready resume block.
  8. (Auto trigger only) Add the auto-continue note: honor the user's LATEST intent; say so if nothing remains.

Reference files

Hard rules

  • Capture exact technical state, not abstractions — paths, symbols, commands, real errors.
  • Never include secrets, tokens, or keys.
  • Output ONLY the handoff artifact and where it lives; do not answer the conversation's questions.
  • The file is the source of truth; the <handoff-context> block is its transport.

What ships with it: 13 files

27.4 KB alongside SKILL.md, 1 of them executable

examples/

ref/

scripts/

Keep looking

Skills are one crate of 326,984. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.