ONLINEAGENT_OPS 2026.Q3 HOME HUBSMODELSBUILDINGCOMPANIESFILMMAKINGSECURITYLIVING WITH ITSKILLSALL HUBS, ONE TERMINAL → ARTICLES CRAFT RECORD BLOG TOOLS SKILLS SEARCH /
HOME›ARTICLES›BEFORE YOU CONNECT AN AI AGENT OR PLUGIN
ARTICLES · CHEAT SHEETINTERMEDIATE

Before You Connect an AI Agent or Plugin: Permissions Checklist

A printable checklist to run before you connect an AI agent, MCP server, plugin or skill: who made it, what it can read and do, and what access to grant.

READ3 min
WORDS606
SECTIONS6
TYPECHEAT SHEET
CHECKED13 OCT 26

Connecting an AI agent, plugin or skill, on one page. This is the short version of permission tiers for agents, skills are a supply chain, what MCP is and prompt injection: why there is no fix. Print it (Ctrl+P, or Cmd+P on a Mac) and tick it before you click “connect”, “allow” or “install”.

TL;DR — THE SHORT VERSION
  • Connecting hands the AI real keys. What it is allowed to do matters more than how carefully you word your request.
  • Anything it reads, a stranger may have written. Hidden instructions in that text can steer it, and there is no complete fix.
  • Grant the lowest level the job needs, keep a human yes on anything that cannot be undone, and take access back afterwards.

What you are connecting

  • Connector or MCP server: a link to another app, so the AI can act there. MCP is a shared standard for such links. 1
  • Skill: a text file the AI reads and follows as instructions. A bad one only has to be read, not run.
  • Plugin or extension: the same questions apply.

Before you connect

  1. ☐ Who made it? A known maker, with a history? Servers from unknown sources are a named risk. 2
  2. ☐ Popular is not checked. One scan found a critical issue in 13.4% of published agent skills. 3
  3. ☐ Skill? Read all of it. Too long to read is itself the finding.
  4. ☐ Lines aimed at the AI’s obedience? “Ignore previous”, “always”, “regardless of”: walk away.
  5. ☐ Does it reach outside itself? A web address it fetches, a file it reads, somewhere it posts to. Each is a way out for your data.
  6. ☐ Does it ask for more than the job? A formatting tool that wants network access is the clearest warning there is.
  7. ☐ What will it read that strangers write? Emails, web pages, shared files, comments. That is where hidden instructions come in. 4
  8. ☐ Keep secrets out of reach. Passwords, personal records, your main inbox: leave them out of what it can see.

What it may do: pick the lowest level

1 Read only. 2 Write to a scratch place. 3 Write with a change list you review. 4 Write and save. 5 Send, pay, post or delete: a human yes, every time. Set the level by how hard the action is to undo, not by how much you trust the tool.

OWASP, the security group, says it is unclear whether any defence is fool-proof, so its advice is to limit the damage: least access, and a person’s approval for risky actions. 5

The first run

  • ☐ Start at level 1. A dry run catches the wrong scope before anything is written.
  • ☐ Leave any “always allow” switch off. An agent asked only to suggest deletions has deleted email itself. 6
  • ☐ Check what it did, not what it says it did: what it opened, changed and sent.

Afterwards

  • ☐ Take the access back when the job is done. Access for a task ends with the task.
  • ☐ Re-check after updates. Skills have no lockfile or signature, so the next version may not be the one you read.
TAKEAWAY

Ask two questions of anything you connect: what does it read that a stranger can write, and what could it do if it obeyed? Keep the second small.

Want the full version?

Read permission tiers for agents, skills are a supply chain and prompt injection: why there is no fix. New to agents? Start with the AI agents cheat sheet. Building an app with AI? Use the AI app security checklist.

Related on this site: Zapier vs Make vs n8n

SOURCES
  1. 1Model Context Protocol, What is the Model Context Protocol (MCP)?, read at source 1 Oct 2026“MCP (Model Context Protocol) is an open-source standard for connecting AI applications to external systems.” ↑
  2. 2Model Context Protocol, Security Best Practices, read at source 1 Oct 2026“Local MCP servers with inadequate restrictions or from untrusted sources introduce several critical security risks”. ↑
  3. 3Snyk, “ToxicSkills”, 5 Feb 2026, quote re-read 22 Sep 2026“13.4% of all skills, or 534 in total, all contain at least one critical-level security issue”. ↑
  4. 4OWASP GenAI Security Project, LLM01:2025 Prompt Injection, read at source 23 Sep 2026“Indirect prompt injections occur when an LLM accepts input from external sources, such as websites or files.” ↑
  5. 5OWASP GenAI Security Project, LLM01:2025 Prompt Injection, undated page, read at source 12 Oct 2026 (as cited on this site’s prompt injection meaning page): “it is unclear if there are fool-proof methods of prevention for prompt injection.” “Enforce privilege control and least privilege access”; “Require human approval for high-risk actions”↑
  6. 6OWASP GenAI Exploit Round-up Report Q1 2026, read at source 22 Sep 2026 (as cited on this site’s permission tiers page): “A Meta AI security researcher reported that an OpenClaw agent ignored stop commands and rapidly deleted email”; the researcher had asked it to “review an email inbox and suggest what to delete or archive”; mitigation: “Require explicit confirmation for destructive actions”.↑

Model Context Protocol, What is the Model Context Protocol (MCP)? and Security Best Practices, read at source 1 Oct 2026.

OWASP GenAI Security Project, LLM01:2025 Prompt Injection, read at source 23 Sep and 12 Oct 2026; GenAI Exploit Round-up Report Q1 2026, read at source 22 Sep 2026.

The five levels, the checks before you connect and the two questions are this site’s own method, from the pages linked above, argued from what each mistake costs rather than measured; they are not claims from these sources.

Free printable AI cheat sheets, and AI news every week.

Sixteen one-page AI guides and checklists in one free PDF, no sign-up. And if you want to keep up, the newsletter covers what changed in AI each week, in plain English, with the sources.

SUBSCRIBE FREE ↗ GET THE CHEAT SHEET PACK
aipromptgeneer://$ about this siteAI PROMPT GENEERAI explained in plain English. Every claim links to its source, with the date it was checked.NEWSLETTERWhat changed in AI each week, in plain English.SUBSCRIBE FREE ↗AI PROMPT GENEER · MACHINE ASSISTANCE DISCLOSEDEVERY ARTICLE CARRIES ITS OWN CHECKED DATE