# longterm-progress

**Repository Path**: woipanda/longterm-progress

## Basic Information

- **Project Name**: longterm-progress
- **Description**: OpenClaw skill for long-running resumable local workflows
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: main
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2026-04-28
- **Last Updated**: 2026-04-28

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# longterm-progress

A self-contained OpenClaw skill for long-running, resumable local workflows.

It provides a small user-facing interface for requests like:

- 推进到完成
- 后台继续
- 继续推进
- 看一下状态
- 暂停 / 停止

Internally it maintains a local task ledger, checkpoints, worker metadata, logs, runtime results, and recovery state under:

```text
skill-state/longterm-progress/
```

## What changed in this release

This release upgrades the skill from a **Windows-first worker launcher** to a **cross-platform worker launcher**:

- keeps the original `.cmd` launchers for Windows
- adds native `.sh` launchers for Linux/macOS
- verifies `systemd-run --user` background execution on Linux
- keeps `statectl-v2.js` / `status-v2.js` / `sync-worker-result-v2.js` as the shared state layer
- supports Codex and Claude CLI workers on Linux through the new shell wrappers

## Install

Copy this folder into your OpenClaw workspace skills directory:

```text
.openclaw/workspace/skills/longterm-progress
```

Then restart/reload OpenClaw if your runtime does not hot-load skills.

## Platform support

### Windows

Use the original batch launchers:

```cmd
scripts\submit-worker.cmd
scripts\submit-codex-worker.cmd
scripts\submit-claude-worker.cmd
```

### Linux / macOS

Use the new shell launchers:

```bash
scripts/submit-worker.sh
scripts/submit-codex-worker.sh
scripts/submit-claude-worker.sh
```

On Linux, this release has been validated with:

- `systemd-run --user`
- `codex ask ...`
- `claude -p ...`
- result txt/json writeback
- `sync-worker-result-v2.js` state sync

## Main commands

Create a task:

```bash
node skills/longterm-progress/scripts/statectl-v2.js init "Task title" --goal "final outcome" --worker codex
```

Check status:

```bash
node skills/longterm-progress/scripts/status-v2.js
```

Submit a generic worker on Linux/macOS:

```bash
bash skills/longterm-progress/scripts/submit-worker.sh <task_id> <worker_type> [command...]
```

Submit Codex worker on Linux/macOS:

```bash
bash skills/longterm-progress/scripts/submit-codex-worker.sh <task_id> <prompt-file> [cwd]
```

Submit Claude worker on Linux/macOS:

```bash
bash skills/longterm-progress/scripts/submit-claude-worker.sh <task_id> <prompt-file> [cwd]
```

Sync worker result:

```bash
node skills/longterm-progress/scripts/sync-worker-result-v2.js <task_id>
```

## Notes for Codex / Claude on Linux

### Codex

The Linux launcher now forwards common auth env vars into `systemd-run --user`, including:

- `OPENAI_API_KEY`
- `OPENAI_API_BASE`
- `OPENAI_MODEL`

So Codex background workers can run under transient user services without manually inlining the key in every command.

### Claude

The Linux launcher now forwards common Anthropic env vars into `systemd-run --user`, including:

- `ANTHROPIC_API_KEY`
- `ANTHROPIC_AUTH_TOKEN`
- `ANTHROPIC_BASE_URL`

## Boundaries

- This is not a permanent daemon by itself.
- It does not bypass human approval for risky or external actions.
- CLI workers still require the corresponding local CLI auth to be available.
- State files and runtime logs are intentionally local and should not be committed.

## Verified local capability in this release

Validated on a Linux OpenClaw host:

- Codex CLI callable
- Claude CLI callable
- `submit-worker.sh` background submission via `systemd-run --user`
- `submit-codex-worker.sh` end-to-end success
- `submit-claude-worker.sh` end-to-end success
- result txt/json writing
- worker result sync back into v2 task state