handoff

Writes self-contained end-of-session handoffs documenting unresolved failures after all reachable fixes.

64|11|Updated May 24, 2026
One-click install
npx skills add https://github.com/AlexanderMattTurner/agent-glovebox --skill handoff-alexandermattturner
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: handoff
Source: https://github.com/AlexanderMattTurner/agent-glovebox/tree/main/.claude/skills/handoff
Command: npx skills add https://github.com/AlexanderMattTurner/agent-glovebox --skill handoff-alexandermattturner

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? When a coding session ends with failures still standing, context about what was tried, what was reproduced, and what remains is lost. This Skill enforces a disciplined process for writing end-of-session handoffs so the next session can act on them without re-investigating from scratch. ## Core Features & Use Cases - Fix-first doctrine: Requires fixing everything reachable on the branch before writing any handoff entry, and requires reproducing failures on the unmodified base before calling them pre-existing. - Deduplication against issues and PRs: Searches GitHub issues and pull requests via MCP tools so an existing OPEN issue or PR owns the item instead of a duplicate filing. - Self-contained output: Handoffs carry the failing command, reproduction output, and observed-versus-hypothesised labels inline, since the next session cannot open scratch files or worktree paths. - Sizing to one session: Estimates units, fleet width, dependency depth, and CI rounds so a handoff is cut at a seam one session can finish. - Use Case: A session ends with two failing tests it did not cause. The Skill guides reproducing them on origin/main, finding the tracking issue, and pasting a complete handoff in chat naming the covering PR. ## Quick Start Ask the AI to write the end-of-session handoff for the failures that remain unresolved on this branch.

Frequently Asked Questions about handoff

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I write a handoff for unfinished work at the end of a coding session?

Fix everything reachable on your branch first, then document only what remains. Each handoff item needs the full path, exact command, reproduction output, and labels marking each claim as observed or hypothesised, pasted in full in chat rather than left as a file path.

How do I prove a test failure is pre-existing and not caused by my changes?

Create a detached worktree of the unmodified base with git worktree add --detach, run the exact failing command there, and quote its output. Also merge origin/main into your branch and rerun, since a failure only the merge produces is still yours.

Should I file a new GitHub issue for a failure I found during my session?

Search issues and pull requests first using the GitHub MCP search tools. An OPEN tracking issue owns the item, so edit that issue to add your row; an OPEN PR gets one line naming it, and a MERGED PR means the item is dropped.

When is an intermittent test failure not a handoff item?

Intermittent failures belong to CI triage, not the handoff. Only two flake cases hand back: one an OPEN PR already covers, and one blocked by a credential you lack that prevents committing a runnable reproduction.

How much work should a single handoff contain?

Size it to what one session can finish: roughly two fleet rounds of units, a dependency chain of about three, and about three CI rounds. Past those bounds, cut at the seam where the next unit depends on the previous one landing, and name that seam in both handoffs.