Run a fail-closed governance preflight before consequential AI-agent actions.
编程
Preflight Workflow
试用Stop before you break — four mandatory checks before every operation: search, rollback, test, scope
它能做什么
Stop before you break — four mandatory checks before every operation: search, rollback, test, scope
技能文档
🛡️ Preflight Workflow
A lightweight safety workflow that stops agents and humans from acting before they think. Four mandatory checks — if even one fails, stop and fix it first.
When to Use
Every time you (or your agent) are about to:
- Run a system command
- Deploy or modify code
- Change configuration
- Execute an unfamiliar or risky operation
Core rule: No search = no action. No preflight = no execution.
How It Works (for agents)
Step 1: User describes what they want to do
If it's a new, unfamiliar, or risky operation → run the full four-question checklist.
Step 2: Four mandatory questions
✅ Q1: Have you searched first?
- Look for existing solutions, known pitfalls, or prior experience
- Don't answer from memory or intuition
- Check at least 2 independent sources
✅ Q2: Is there a rollback plan?
- What state can you return to if it fails?
- Have you made a backup?
- Do you have clear rollback steps?
✅ Q3: Has it been tested or validated?
- Has the approach been verified on a small scale?
- Are you relying on "let's try and see"?
✅ Q4: Is the impact clear?
- What are you changing?
- Who/what will it affect?
- How long to recover if it goes wrong?
Step 3: All pass → proceed. Any fail → stop.
Four ✅ required. Not one less.
Step 4: After the operation
| Outcome | Action |
|---|---|
| ✅ Success | Log it (date + what + result) |
| ❌ Failure | Write to LEARNINGS.md (root cause + fix + prevention) |
| 💡 New insight | Update this SKILL.md |
CLI Mode (no agent, just a shell script)
chmod +x preflight.sh
./preflight.sh "deploy new cron job"
Answers four yes/no questions interactively. All ✅ → "All passed, go ahead." Any ❌ → "Blocked: ... Fix it first."
Common Pitfalls
- ❌ "I know this one" → No you don't. Search first.
- ❌ "This is too simple to check" → Small ops cause big outages.
- ❌ Trying repeatedly without documenting → Failure is a lesson. Write it down.
- ✅ Getting corrected? Accept → Fix → Remember. No excuses.
Included Files
| File | Purpose |
|---|---|
preflight.sh | Interactive CLI checklist (4 yes/no questions) |
SKILL.md | Loadable skill for any SKILL.md-compatible agent |
LEARNINGS.md | Post-mortem template (root cause + fix + prevention) |
install.sh | One-command install script |
README.md | Quick start guide |
Verification
Run the script and answer all four questions:
- All
y→ ✅ "All passed, proceed." - Any
n→ ❌ "Blocked by: [reason]. Fix before proceeding."
相关技能
Wrap multi-step agent workflows with pre-execution checks, side-effect queues, result validation, retry budgets, checkpointing, audit logs, and failure-rule accumulation. Prevents false successes, duplicate sends, unrecoverable crashes, and silent drift in LLM production systems. Use it when a workflow sends, publishes, pays, deletes, or writes to another system, runs unattended on a schedule, or must be safe to rerun after a mid-task failure. Trigger keywords: workflow safety, workflow guardian, agent guard, guardrails, pre-execution check, pre-flight check, retry budget, idempotency, false success, duplicate send, checkpoint recovery, rerun safety, audit log, drift detection, agent reliability, production guardrails, 工作流守护, Agent 护栏, 副作用队列, 漂移检测, 幂等, 防重复发送, 假成功, 断点恢复, 生产护栏, 重跑安全. 中文摘要:为多步骤 Agent 工作流加装七项护栏——执行前检查、检查点、副作用队列、预算 重试、结果验证、审计记录、规则沉淀,拦截假成功、重复发送与渐进漂移。触发词:工作流 守护、Agent 护栏、假成功拦截、防重复发送、重试预算、断点恢复、生产护栏、漂移检测.
Audit a public webpage for automation and integration readiness through the TinyOps paid x402 service. Use for evidence-backed readiness findings, prioritized integration risks, or a structured acceptance checklist. Do not use for private or authenticated pages, form submission, or security testing.
Preflight safety guard for shell and infrastructure commands. Use before running commands that delete, overwrite, move, deploy, rewrite git history, change p...
For novel, fiddly, or infrastructure-sensitive work, search prior notes and solved cases for a structurally similar problem. Reuse the method as a hypothesis, verify it against current state, test the real outcome falsifiably, and record only the reusable delta.
Prepare local documents for OpenClaw review without losing source anchors, redaction notes, or validation checkpoints.