Skip to content
Back to skills

Handoff

ASecurity

Write a curated, size-capped handoff file that carries task state from this session to a fresh one. Run only when the user invokes /handoff.

  • 4 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added November 7, 2026
ai-agentsgobashgit

Security analysis

A100/100

Pro scans all 3 files and shows the line behind each finding

Scanned November 7, 2026

npx -y skills add CowboyLogic/ai-dev --skill handoff --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Handoff?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Handoff
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/cowboylogic-handoff/badge)](https://www.skillsdirectory.com/skills/cowboylogic-handoff)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: handoff
description: Write a curated, size-capped handoff file that carries task state from this session to a fresh one. Run only when the user invokes /handoff.
license: MIT
disable-model-invocation: true
---

# Handoff

Write one live file, `.agent-output/handoff.md`, so a fresh session can pick up this
task where it stopped. Overwrite any previous handoff. The file holds task state only,
not durable facts: those belong in the harness's memory feature.

## 1. Check that the path is ignored

A handoff can hold proprietary context, so never write it to a path git would stage.
Run this before the first write, from inside the repository:

```bash
git check-ignore -q .agent-output/handoff.md
```

- **Exit 0 (ignored):** continue to step 2.
- **Exit 1 (not ignored):** stop and ask which the user prefers. Write nothing until they
  answer.
  - **A.** Create `.agent-output/` and add `.agent-output/` to the repository's
    `.gitignore`. This is a tracked change that every contributor receives.
  - **B.** Append `.agent-output/` to the file printed by
    `git rev-parse --git-path info/exclude`. This is untracked and applies to this clone
    only.
- **Not a git repository** (the command reports that): continue to step 2. There is
  nothing to ignore.

## 2. Gather the header facts

- Timestamp: `date -u +%Y-%m-%dT%H:%M:%SZ`
- Branch: `git symbolic-ref --short -q HEAD`. If that prints nothing, use `(detached)`.
  Outside a git repository, use `(none)`.

## 3. Write the file

Create `.agent-output/` if it does not exist, then write the file in this shape. The
header lines must keep this exact form, because the session-start hook reads the
`Branch:` line.

```markdown
# Handoff

Written: <timestamp>
Branch: <branch>

## Goal and current state
## Decisions
## Gotchas and dead ends
## Next step
## In-flight work
## Open questions
```

- **Goal and current state:** what the task is and how far it got.
- **Decisions:** each decision with the reason for it.
- **Gotchas and dead ends:** approaches already tried and why they failed.
- **Next step:** the exact next action, specific enough to start without re-deriving it.
- **In-flight work:** uncommitted or half-done changes, by file.
- **Open questions:** what still needs an answer, and from whom.

Rules:

- Keep the whole file to about 100 lines. Cut to fit.
- Leave out anything a fresh session can derive from `git log`, `git diff`, or the code.
- Leave out durable facts and preferences. They belong in memory, not here.
- Never include credentials, tokens, or secrets.
- Omit a section only if it is truly empty. Write `None` rather than inventing content.

## 4. Report back

Tell the user the path, the line count, and that the file is meant to be reviewed before
use. Suggest `/clear` when they are ready to continue in a fresh session. Do not run it
for them.

Files in this skill

  • README.md4.7 KB
  • SKILL.md2.8 KB
  • scripts/session-start.sh2.2 KB

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.