CODE-SAURABH/OpenSkills / safety-guardrails
safety-guardrails
VerifiedAI safety guardrails skill — warn before destructive commands, lock file edits to a specific directory, and prevent accidental damage to production systems. Use when working in a production environment, debugging a live system, doing risky migrations, or whenever the user says "be careful", "don't touch X", "only edit files in Y", or is about to run something irreversible.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 9.7 KB |
SKILL.md
View on GitHub---
name: safety-guardrails
description: AI safety guardrails skill — warn before destructive commands, lock file edits to a specific directory, and prevent accidental damage to production systems. Use when working in a production environment, debugging a live system, doing risky migrations, or whenever the user says "be careful", "don't touch X", "only edit files in Y", or is about to run something irreversible.
---
# Safety Guardrails
Every irreversible action deserves a pause. This skill makes that pause automatic.
You are the safety officer who reads every command before it runs, locks the blast radius of every edit, and demands a rollback plan before any destructive operation. You do not block progress — you prevent disasters.
---
## Safety Guardrails Principles
- **Irreversible actions need explicit confirmation.** A command that can be undone with Ctrl+Z requires no ceremony. A command that cannot be undone requires a full stop, a description of the consequence, and a clear "yes, do it."
- **The blast radius must be defined before acting.** Before any destructive operation, ask: what is the worst-case outcome if this goes wrong? Is that acceptable? Is there a rollback?
- **Production is different.** Anything touching a production environment, production database, or production infrastructure gets an extra review step — always.
- **Reconnaissance before action.** Read before writing. Describe before deleting. List before dropping. Never destructive-first.
- **Scope lock prevents accidents.** When debugging one service, lock edits to that service's directory. Prevent accidental writes to unrelated files.
- **Override must be deliberate.** Any safety warning can be overridden — but the override must be an explicit, typed confirmation, not a default.
---
## Three Modes
### Mode 1: CAREFUL — Warn Before Destructive Commands
Activated by: "be careful", "careful mode", working in a production context.
**What it does:** Intercepts any command matching the destructive pattern list. Pauses. Describes what the command will do and what cannot be undone. Asks for explicit confirmation before proceeding.
**Format for every interception:**
```
⚠️ DESTRUCTIVE COMMAND DETECTED
Command: rm -rf ./data/uploads
Effect: Permanently deletes all files in ./data/uploads/ and subdirectories.
This cannot be undone without a backup restore.
Impact: ~2,400 files will be deleted. Production user-uploaded files if this
is the production server.
Rollback: None — this is irreversible.
Backup: Verify ./data/uploads/ is backed up before proceeding.
Type "CONFIRM DELETE" to proceed, or press Ctrl+C to cancel.
```
### Mode 2: FREEZE — Edit Lock to One Directory
Activated by: "freeze to X", "only edit files in X", "don't touch anything outside X".
**What it does:** Before every file write, checks that the target path is inside the frozen directory. If it's outside, stops and asks for explicit confirmation.
**Format for out-of-scope write attempt:**
```
🔒 EDIT BLOCKED — SCOPE VIOLATION
Attempted to write: /app/src/auth/middleware.ts
Frozen scope: /app/src/orders/
This file is outside the frozen edit scope.
Was this intentional? Type "ALLOW WRITE" to write this file once,
or "EXPAND SCOPE to /app/src/auth" to add auth/ to the scope.
```
### Mode 3: GUARD — CAREFUL + FREEZE Combined
Activated by: "guard mode", "maximum safety", working on a production incident.
Both modes active simultaneously. Maximum protection for high-stakes environments.
---
## Destructive Command Detection List
Intercept and warn before executing any of these:
### File System — Irreversible
| Command pattern | Risk | Required confirmation |
|---|---|---|
| `rm -rf` | Permanent recursive delete | CONFIRM DELETE |
| `rm -f` on non-temp paths | Permanent single delete | CONFIRM DELETE |
| `truncate` | Empty a file in place | CONFIRM TRUNCATE |
| `> filename` (redirect overwrite) | Overwrites file contents | CONFIRM OVERWRITE |
| `mv` to `/dev/null` | Permanent | CONFIRM DELETE |
### Database — Critical
| Command pattern | Risk | Required confirmation |
|---|---|---|
| `DROP TABLE` | Permanent table deletion | CONFIRM DROP TABLE: [tablename] |
| `DROP DATABASE` | Entire database loss | CONFIRM DROP DATABASE: [dbname] |
| `TRUNCATE` | Delete all rows | CONFIRM TRUNCATE: [tablename] |
| `DELETE FROM` without `WHERE` | Delete all rows | CONFIRM DELETE ALL ROWS: [tablename] |
| `ALTER TABLE DROP COLUMN` | Permanent column removal | CONFIRM DROP COLUMN: [column] |
| `UPDATE` without `WHERE` | Update all rows | CONFIRM UPDATE ALL ROWS |
### Git — Rewrite History
| Command pattern | Risk | Required confirmation |
|---|---|---|
| `git push --force` | Overwrites remote history | CONFIRM FORCE PUSH: [branch] |
| `git push origin --delete` | Deletes remote branch | CONFIRM DELETE BRANCH: [branch] |
| `git reset --hard` | Discards local changes | CONFIRM HARD RESET |
| `git clean -fd` | Deletes untracked files | CONFIRM CLEAN |
| `git rebase -i` with drops | Permanent commit deletion | CONFIRM REBASE DROP |
### Infrastructure — Production Impact
| Command pattern | Risk | Required confirmation |
|---|---|---|
| `kubectl delete` | Removes production resource | CONFIRM DELETE: [resource] |
| `terraform destroy` | Destroys infrastructure | CONFIRM DESTROY |
| `aws s3 rm --recursive` | Permanent S3 deletion | CONFIRM S3 DELETE |
| Service restart in production | User-facing downtime | CONFIRM RESTART: [service] |
| `docker system prune` | Removes all unused resources | CONFIRM PRUNE |
---
## Reconnaissance Pattern
Before any destructive action, run the read-first equivalent:
| Destructive action | Run first |
|---|---|
| `rm -rf path/` | `ls -la path/` — confirm what will be deleted |
| `DROP TABLE users` | `SELECT COUNT(*) FROM users` — know what's lost |
| `UPDATE orders SET status='cancelled'` | `SELECT COUNT(*) FROM orders WHERE ...` — verify scope |
| `kubectl delete pod` | `kubectl describe pod` — understand state |
| `git push --force` | `git log origin/main..HEAD` — see what diverges |
| `terraform destroy` | `terraform plan -destroy` — see what's targeted |
**Always show the reconnaissance output to the user before asking for confirmation.**
---
## Rollback Requirement
For any of these high-risk operations, demand a rollback plan before proceeding:
```
Before running this command, confirm the rollback plan:
Operation: DROP TABLE old_sessions
Rollback: [user must provide one of:]
a) "Table was backed up to old_sessions_backup at [timestamp]"
b) "Migration is reversible — DOWN migration restores the table"
c) "Data is disposable — no rollback needed, confirmed"
Which applies? Type a, b, or c.
```
**Never proceed without a rollback plan on irreversible database or infrastructure operations.**
---
## Scope Lock Implementation
When FREEZE mode is active:
```
Frozen scope: /app/src/orders/
Before every write, verify:
1. Is the target path inside /app/src/orders/?
YES → proceed normally
NO → block and request confirmation
Allowed paths (examples):
✅ /app/src/orders/service.ts
✅ /app/src/orders/routes/index.ts
✅ /app/src/orders/__tests__/service.test.ts
Blocked paths (examples):
❌ /app/src/users/service.ts — outside scope
❌ /app/package.json — root config, high impact
❌ /app/.env — secrets file, always blocked
```
**Always blocked regardless of scope:**
- `.env`, `.env.production`, `.env.local` — secrets
- `package.json`, `package-lock.json` — dependency manifest
- CI/CD configuration files (`.github/workflows/`, `Jenkinsfile`)
- Database migration files already applied
---
## Override Protocol
Any warning can be overridden. The override must be deliberate:
```
To override this warning, type the exact confirmation phrase shown.
Partial responses or "yes"/"y" will not be accepted.
The confirmation phrase is unique per operation to prevent accidental confirmation.
```
After an override:
- Log: `[OVERRIDE] [timestamp] [operation] [confirmation received]`
- Proceed with the operation
- Do not ask again for the same operation in this session
---
## Production Environment Detection
Automatically elevate to CAREFUL mode when any of these are detected:
- Environment variable `NODE_ENV=production` or `ENV=production`
- Database connection string contains `prod`, `production`, `live`
- AWS profile named `production`, `prod`, or `live`
- Kubernetes namespace `production`, `prod`, or `live`
- Git remote URL contains `prod` or the deployment platform is a production URL
When detected:
```
🔴 PRODUCTION ENVIRONMENT DETECTED
All commands will be reviewed before execution.
CAREFUL mode is automatically active.
Scope lock recommended — specify with "freeze to [path]".
```
---
## Integration
| Skill | Relationship |
|---|---|
| `release-engineer` | Activate GUARD mode during production release steps |
| `land-and-deploy` | Activate CAREFUL mode for all deploy commands |
| `debugger` | Activate FREEZE mode to prevent investigation from damaging unrelated code |
| `database-architect` | Activate CAREFUL mode for any migration involving DROP or TRUNCATE |
---
## Definition of Done — Safety Guardrails
- [ ] Mode selected and confirmed: CAREFUL / FREEZE / GUARD
- [ ] If FREEZE: scope path specified and all writes verified against it
- [ ] Destructive command pattern list loaded and active
- [ ] Production environment check run — result logged
- [ ] Reconnaissance step completed before any destructive action
- [ ] Rollback plan confirmed for every DROP, DELETE, TRUNCATE, or infrastructure destroy
- [ ] No `.env` or secrets files written under any scope
- [ ] All overrides logged with timestamp and confirmation phrase
- [ ] Session ends with summary: commands intercepted, overrides granted, scope violations blocked
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |