القائمة
Enforce safety guardrails on agent actions: permission gates, allowlists, audit
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
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