agentsclimarketplace

Handoff

Skill anortham/goldfish/skills/handoff

Use when returning to a project after time away, switching harnesses, or handing work off to another agent, to produce a structured session-resumption summary from the active brief, recent checkpoints, and git delta.From its SKILL.md

Install
npx -y skills add anortham/goldfish --skill handoff

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

2 things to look at

  • no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
  • 3 stars3 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.

SKILL.md

4.5 KB, as published. Nobody here has run it

Handoff

When To Use

Three triggers:

  • Returning to a project after time away. Yesterday's work, last week's branch, something you parked.
  • Switching harnesses. You were on Claude Code and are now in Codex, or the reverse. Native harness memory does not survive the jump.
  • Handing off to a different agent. A parallel agent picks up where another left off in the same repo.

How /handoff differs from neighbors

  • /standup aggregates across projects for a daily-update audience (you, looking at what you did everywhere).
  • /recall returns raw checkpoints for an in-session agent that needs memory restored.
  • /handoff synthesizes within one project into a single resumption document for a different agent or a returning you, picking up cold.

Pick by audience and scope.

Workflow

Four steps. Run them in order, then compose the document.

1. Load the active brief

recall({ limit: 0 })

This returns the active brief without pulling checkpoints. If no brief is active, note that in the output and keep going.

2. Load recent checkpoints

recall({ days: 3, limit: 10, full: true })

Default window is 3 days. If the user passed a time argument (for example --since 2d or --since 4h), honour it:

recall({ since: "2d", limit: 10, full: true })

full: true is required so you see next, unknowns, and git context.

3. Capture git state

git rev-parse --abbrev-ref HEAD
git status -s
git log -1 --oneline

Three short commands, three short outputs. No fancy flags.

4. Synthesize the document

Compose one markdown document using the sections in Output Format below. Do not dump tool output verbatim; distill it.

Output Format

One markdown document, sections in this order:

## Direction

3-5 lines summarizing the active brief: goal, key constraints, current status. If no brief is active, write "No active brief" and lean on the checkpoint trail for direction.

## State at handoff

  • Current branch (from git rev-parse --abbrev-ref HEAD)
  • Last commit (from git log -1 --oneline)
  • Uncommitted changes (from git status -s), or "clean working tree"

Keep this compact. Three bullets maximum.

## Recent activity

Last 5-10 checkpoints, dense format. Group adjacent checkpoints into logical milestones when they tell a continuous story; otherwise list chronologically. For each checkpoint include:

  • One-line summary (use the checkpoint's summary field)
  • Associated commit or tag if the git context identifies one

When clustering, lead the cluster with a one-line milestone header, then nest the supporting checkpoints beneath it.

## Next steps

Pull from the most recent checkpoint's next field. Add any success criteria from the active brief that do not yet have checkpoint evidence. Short bullets; concrete actions.

## Open questions

Pull from recent checkpoints' unknowns fields and from any open items in the active brief. If there are no open questions, say so plainly.

## Source pointers

File paths the receiving agent can read deeper from:

  • The active brief file (.memories/briefs/<id>.md)
  • The 2-3 most recent checkpoint files (.memories/<date>/<time>_<hash>.md)
  • Any docs/plans/ documents referenced by the brief or a recent checkpoint

Time Scoping

The skill takes one optional argument: a time window (--since 2d, --since 4h).

Default behaviour:

  1. If the current branch has commits, start from the last commit on this branch and expand backwards until you have at least 5 checkpoints or hit the 3-day cap.
  2. Otherwise default to the last 3 days.

Users who want a fixed window should pass --since.

Why This Skill Exists

Native harness memory does not survive harness switches. Goldfish's evidence ledger (briefs plus checkpoints) does, and /handoff turns that ledger into a portable resumption document. It is the explicit answer to "how does work resume on a different harness or a different agent?"

Rules

  • Synthesize, do not dump. The receiving agent reads one document, not a transcript.
  • Lead each section with the signal, not the mechanics.
  • If a section is empty (no brief, no unknowns, clean tree), say so in one line and move on.
  • Keep the whole document short enough to read in one pass.

Keep looking

Skills are one crate of 609,856. 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.