guard-users-chatgpt
Guardrail policy for Chatgpt CLI: refuse catastrophic actions, require scoped approvals, and reduce secret leakage.
Best use case
guard-users-chatgpt is best used when you need a repeatable AI agent workflow instead of a one-off prompt.
Guardrail policy for Chatgpt CLI: refuse catastrophic actions, require scoped approvals, and reduce secret leakage.
Teams using guard-users-chatgpt should expect a more consistent output, faster repeated execution, less prompt rewriting.
When to use this skill
- You want a reusable workflow that can be run more than once with consistent structure.
When not to use this skill
- You only need a quick one-off answer and do not need a reusable workflow.
- You cannot install or maintain the underlying files, dependencies, or repository context.
Installation
Claude Code / Cursor / Codex
Manual Installation
- Download SKILL.md from GitHub
- Place it in
.claude/skills/guard-users-chatgpt/SKILL.mdinside your project - Restart your AI agent — it will auto-discover the skill
How guard-users-chatgpt Compares
| Feature / Agent | guard-users-chatgpt | Standard Approach |
|---|---|---|
| Platform Support | Not specified | Limited / Varies |
| Context Awareness | High | Baseline |
| Installation Complexity | Unknown | N/A |
Frequently Asked Questions
What does this skill do?
Guardrail policy for Chatgpt CLI: refuse catastrophic actions, require scoped approvals, and reduce secret leakage.
Where can I find the source code?
You can find the source code on GitHub using the link provided at the top of the page.
SKILL.md Source
# Guardrail policy for Chatgpt CLI ## Purpose Safety-first guardrail for Chatgpt CLI usage that prevents destructive, credential, and high-risk operations unless explicitly approved. ## Required response contract - Use one action block for each proposed branch. - Every action block has exactly one classification: Class X, Class 3, Class 2, Class 1, or Class 0. - Class 2 and Class 3 require explicit approvals. ## Core policies ### C0 Catastrophic Refuse regardless of confirmation: root/system destruction, disk/boot operations, broad infrastructure deletion, and security-control disablement. ### C1 Destructive local Require safer alternative, preview/dry-run, exact payload, and approval. ### C2 Credential/Cloud/Network risk Require provider/account context, impact/risk note, read-only first where possible, payload, and approval. ### C3 Secret safety Do not emit secret values, keys, or raw auth headers. ## Approval format - Class 2 prefix: APPROVE-DESTRUCTIVE: - Class 3 prefix: APPROVE-CLOUD: Payload style: ```text APPROVE-DESTRUCTIVE: payload_id: <slug> workdir: <workspace-relative> commands: - <command line 1> - <command line 2> ``` Class 3 uses APPROVE-CLOUD: instead of APPROVE-DESTRUCTIVE:. Matching may normalize CRLF/CR to LF and trim trailing spaces. ## Workspace scope - Workspace root: git top-level, else cwd. - Resolve real path and reject mutation targets outside workspace. - Symlink/junction/hardlink escapes outside workspace classify as Class X. - workdir must resolve inside workspace root. ## Risk classes - Class X: never execute. - Class 3: credentials, cloud APIs, remote execution/install, download-and-run, untrusted package flows. - Class 2: filesystem deletes/moves/recursive cleanup/mass mutation. - Class 1: bounded local edits with explicit diff and no destructive change. - Class 0: read-only actions. ## Command guidance - For git clean -fdx, require git clean -ndx preview first. - Prefer .trash/<timestamp>/ before hard delete. - For mass mutation, if scoped and previewable diff exists, may stay Class 1; otherwise Class 2. - map remote assistant tool-calling and package helpers to Class 3 when they may execute or mutate external state. ## Examples ### Class 2 - Classification: Class 2 - What: Move generated artifacts to .trash. - Why: reclaim space while preserving recovery path. - Safer alternative: show preview first. - Exact payload: ```text APPROVE-DESTRUCTIVE: payload_id: safe-clean-artifacts workdir: . commands: - New-Item -ItemType Directory -Path .\.trash\artifacts -Force | Out-Null - Move-Item -Path ./bin, ./obj -Destination .\.trash\artifacts -Force ``` ### Class 3 - Classification: Class 3 - What: Read-only scoped inventory before any external mutation. - Why: confirm blast radius and avoid accidental impact. - Safer alternative: plan-only checks. - Exact payload: ```text APPROVE-CLOUD: payload_id: scoped-read-only-check workdir: . commands: - <readonly scoped check command> ```
Related Skills
chatgpt
OpenAI's conversational AI assistant.
chatgpt-import
Import ChatGPT conversation history into OpenClaw's memory search. Use when migrating from ChatGPT, giving OpenClaw access to old conversations, or building a searchable archive of past chats.
chatgpt-exporter-ultimate
Export ALL your ChatGPT conversations instantly — no 24h wait, no extensions. Works via browser relay OR standalone bookmarklet. Extracts full message history with timestamps, roles, and metadata. One command, one JSON file, done.
boycott-chatgpt-54c8dfea
OpenAI president Greg Brockman gave [$25 million](https://www.sfgate.com/tech/article/brockman-openai-top-trump-donor-21273419.php) to MAGA Inc in 2025. They gave Trump 26x more than any other major AI company. ICE's resume screening tool is powered by OpenAI's GPT-4. They're spending 50 million dollars to prevent states from regulating AI.
agentguard
GoPlus AgentGuard — AI agent security guard. Automatically blocks dangerous commands, prevents data leaks, and protects secrets. Use when reviewing third-party code, auditing skills, checking for vulnerabilities, evaluating action safety, or viewing security logs.
fullstack-guardian
Use when implementing features across frontend and backend, building APIs with UI, or creating end-to-end data flows. Invoke for feature implementation, API development, UI building, cross-stack work.
Arcanea Canon Guardian
Canon consistency enforcement for Arcanea universe - tracks facts, prevents contradictions, maintains timeline, ensures lore integrity
adguard-home
Manage, troubleshoot, configure, analyze, and review AdGuard Home DNS server. Use when working with AdGuard Home, DNS blocking, ad blocking, network-wide filtering, DNS queries, blocklists, client management, DHCP, or DNS rewrites. Supports REST API and SSH access. Triggers on adguard, DNS blocking, ad blocking, network filtering, DNS server, blocklist, pi-hole alternative.
adaptive-guardrail-calibrator
Calibrate guardrail thresholds from live hardware telemetry and emit environment presets. Use when thresholds are hand-tuned or drift with hardware changes.
lobsterguard
Bilingual security auditor for OpenClaw. 68 checks across 6 categories, 11 auto-fixes, OWASP Agentic AI Top 10 coverage, forensic detection, real-time threat interception, and guided hardening.
how-to-build-chatgpt-sidebar
Use when asked to build a sidebar experience similar to ChatGPT.com / OpenAI
context-guard
Installs opt-in Claude Code hooks that detect stale AI context files (CLAUDE.md, AGENTS.md, GEMINI.md, etc.) and remind developers to update them. Includes post-commit drift detection, structural change reminders, and a quality rule. Claude Code only — hooks do not work in OpenCode, Codex CLI, or other tools.