Ops Framework

A 0-token jobs + monitoring framework for OpenClaw: run long-running read tasks via scripts, checkpoint/resume safely, and send periodic progress + immediate alerts to Telegram. Write jobs are blocked by default and must be explicitly approved and verified.

Ops Framework（Jobs + Ops Monitor）— OpenClaw Skill (MVP)

目标：把“长任务执行 / 断点续跑 / 进度汇报 / 异常告警”做成 0-token 的可复用能力。

这套技能由两部分组成：

• ops-monitor.py：一个纯本地脚本，负责跑 status / 检测卡住 / 发送 Telegram 快报
• ops-jobs.json：一个声明式 job 配置（包含 kind/risk/命令/策略）

推荐作为“外挂”存在：长任务尽量用脚本跑，避免让模型持续盯进度烧 token。

What this solves

• Long-running read jobs: scans, inventories, large syncs, periodic polling; support pause/resume and stall detection.
• One-shot read jobs: health checks or lint-like checks; only report when they emit ACTION REQUIRED / ALERT or fail.
• One-shot write jobs: blocked by default; must be explicitly approved and must chain to a read-only verification job.

Quickstart (local)

1. Copy files to your OpenClaw host (suggested layout):

• ~/.openclaw/net/tools/ops-monitor.py
• ~/.openclaw/net/config/ops-jobs.json
• ~/.openclaw/net/state/ops-monitor.json (auto-created)

You can also run the script from any directory as long as OPENCLAW_HOME points to your OpenClaw state dir (default ~/.openclaw).

2. Start from the example config:

• ops-jobs.example.json

3. Validate:

python3 ops-monitor.py validate-config --config-file ~/.openclaw/net/config/ops-jobs.json
python3 ops-monitor.py selftest

4. Run one monitoring tick (prints only; does not send):

python3 ops-monitor.py tick --print-only

5. Run periodic ticks via your OS scheduler (launchd/systemd/cron). The script is designed to be called frequently; it decides whether to report based on policy and state.

Job kinds and safety

kind is one of:

• long_running_read
• one_shot_read
• one_shot_write (never auto-executed by ops-monitor)

risk is one of:

• read_only
• write_local
• write_external

Rules (MVP):

• long_running_read may auto-resume only when risk=read_only and policy.autoResume=true.
• one_shot_read may run explicitly or via queue (read-only only).
• one_shot_write is always blocked from auto-run; it exists as a declarative “approval + verification chain” placeholder.

Status contract (for long_running_read)

Your commands.status must print JSON to stdout, with at least:

• running (boolean)
• completed (boolean)

Recommended:

• pid (number)
• stopReason (string)
• progress (object)
• progressKey (string) — stable key used for stall detection
• level (ok|warn|alert)
• message (string)

Commands

# Validate config
python3 ops-monitor.py validate-config --config-file ~/.openclaw/net/config/ops-jobs.json

# Print current statuses (no Telegram)
python3 ops-monitor.py status --config-file ~/.openclaw/net/config/ops-jobs.json

# One monitoring tick
python3 ops-monitor.py tick --config-file ~/.openclaw/net/config/ops-jobs.json

# Explicitly start/stop a long job
python3 ops-monitor.py start <job_id> --config-file ~/.openclaw/net/config/ops-jobs.json
python3 ops-monitor.py stop <job_id>  --config-file ~/.openclaw/net/config/ops-jobs.json

# Run one one_shot_read job explicitly
python3 ops-monitor.py run <job_id> --config-file ~/.openclaw/net/config/ops-jobs.json

Reference docs

• Spec (Chinese, detailed): OPS_FRAMEWORK.md