Menu
Enforce safety guardrails on agent actions: permission gates, allowlists, audit
Instalar com Codex ou Claude Copie este prompt, cole no Codex, Claude ou outro assistente e deixe que ele revise a página da skill e instale para você.
Baseado na classificação ocupacional SOC
Build and maintain ZERO-COST, Hermes-automated online income systems — multiple independent "streams" (affiliate blog, Gumroad products, tools hub, Fiverr guides), each with a generator + self-running cron, hosted free on GitHub Pages. Use when the user wants to "make money online for free", "passive/automated income", or have "Hermes do everything". Encodes the proven architecture, the affiliate-token pattern, and the gotchas (Pages /docs, clean-on-build, f-string braces, MSYS paths, ad-hoc verification) that broke a real build.
Build and operate a zero-cost, fully-automated multi-stream money system (affiliate + digital products + POD + traffic + support) maintained by Hermes Agent crons. Use when the user wants to make money online for free with automation, or to extend/verify the money-engine project.
Configure, extend, or contribute to Hermes Agent.
Stand up a local autonomous "agent company" — Paperclip (the orchestrator) + Hermes Agent employees (via the hermes_local adapter) — on Windows via the Hermes terminal. Covers the verified pnpm/tsx build, the embedded-Postgres-vs-admin trap, and how the Hermes adapter is already bundled in Paperclip main (so you do NOT hand-wire it). Use when the user wants to "run a Paperclip company", "hire a Hermes agent in Paperclip", "build the agent-company stack", or references paperclipai/paperclip + NousResearch/hermes-paperclip-adapter.
Drive Autodesk Fusion 360 via its MCP server to programmatically build, simulate, and export parametric CAD / robot models. Also covers STATIC code+math engineering verification of a payload WITHOUT running Fusion (no MCP needed) — servo torque margins, mass/COM/stability, structural safety factors, 3D-printability, power budget. Use when a project sends Python payloads to Fusion 360 over MCP (e.g. a run_simulation.py controller + versioned optimus_*.py / model_*.py payloads), when asked to "run the vN simulation", to "analyze which version is best", or to "review/verify the engineering" of a CAD script.
Deploy, build, verify, and operate a Next.js site on Cloudflare Workers via the OpenNext adapter (@opennextjs/cloudflare), and set up the Cloudflare MCP server. Covers the build pipeline, multi-worker split configs, auth (wrangler login OAuth token OR CLOUDFLARE_API_TOKEN), headless-deploy pitfalls (stdin is not a tty, IPv6 DNS), the diff-based resumable asset upload, the Cloudflare MCP server (run <accountId> + OAuth token reuse), and how to verify a project is "completely working". Use when deploying/building/verifying for Cloudflare, wiring up the Cloudflare MCP, or hitting Cloudflare auth/deploy errors.
| name | agent-guardrails |
| description | Enforce safety guardrails on agent actions: permission gates, allowlists, audit |
Pre-flight safety gate for planned agent actions: permission checks, allowlists, and an audit trail.
Keywords: guardrails, safety, agent, cli, security, policy, python, open-source, automation, MIT
Part of the itsPremkumar Hermes / OpenClaw / Paperclip agent stack — 31 free, MIT-licensed, CI-tested agent-native tools.
Agents take irreversible actions (shell, deletes, network) with no pre-flight safety check. Agent Action Guardrails solves this: Pre-flight safety gate for planned agent actions: permission checks, allowlists, and an audit trail.
Best for: Agent platform teams and security reviewers who must approve or deny planned agent actions.
# Requires Python 3.8+. No pip install needed.
curl -O https://raw.githubusercontent.com/itsPremkumar/agent-guardrails/main/agent_guardrails.py
# Or copy the file anywhere — it's self-contained.
python agent_guardrails.py self-test # prove it works end-to-end
python agent_guardrails.py check --help # check subcommand
| Alternative | Why this skill is better |
|---|---|
| Trusting the prompt | guardrails evaluates the action, not the intent. |
| Post-hoc review | It blocks before execution, not after damage. |
| Manual code review | Policy-as-code makes safety repeatable across agents. |
Q: What does check do?
A: Evaluates a planned action against policy and returns allowed/denied + reason.
Q: Is it offline?
A: Yes. Pure stdlib, no network, no telemetry.
Q: How do allowlists work?
A: Define permitted command patterns; anything outside is denied by default.
Q: Does it log decisions?
A: Yes — every decision emits a structured audit entry.
Built and maintained by @itsPremkumar (Chennai, India · serving developers worldwide). Free for individuals and teams everywhere. Documentation in English; tool output is locale-neutral.
# .github/workflows/verify.yml
name: Verify
on: [push]
jobs:
verify:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
- name: Self-test agent-guardrails
run: python agent_guardrails.py self-test
Free + MIT-0 (free, modifiable, no attribution required). Sponsor if useful:
⭐ Star on GitHub