Claude-skill-registry long-running-harness-kit

Design long-running agent harnesses with resumable checkpoints and initializer/coder handoffs. Use when tasks span multiple sessions or require recovery after interruption.

install

source · Clone the upstream repo

git clone https://github.com/majiayu000/claude-skill-registry

Claude Code · Install into ~/.claude/skills/

T=$(mktemp -d) && git clone --depth=1 https://github.com/majiayu000/claude-skill-registry "$T" && mkdir -p ~/.claude/skills && cp -r "$T/skills/data/long-running-harness-kit" ~/.claude/skills/majiayu000-claude-skill-registry-long-running-harness-kit && rm -rf "$T"

manifest: skills/data/long-running-harness-kit/SKILL.md

Long Running Harness Kit

Overview

Define a minimal harness that can pause and resume safely, with clear artifacts that survive context resets. Emphasize initializer + coder roles and persistent run state.

Quick start

Fill
```
templates/harness_plan.json
```
.
Write
```
progress.log
```
and
```
results.json
```
outside the skill directory.
Use the initializer to set up context; use the coder to make incremental progress.

Core Guidance

Use two roles: initializer (setup + context) and coder (incremental tasks).
Persist state each step (progress, decisions, artifacts).
Keep tasks small; checkpoint after every meaningful change.
On resume, read artifacts first, then continue with a narrow next step.

Resources

```
references/harness-checklist.md
```
: Resumability and handoff checklist.
```
templates/harness_plan.json
```
: Plan scaffold for long-running runs.

Validation

Confirm artifacts exist and a resume step can continue from the last checkpoint.