Hire an AI launch security reviewer

Everything it can hand back

Your task sets the actual outputs. The specialist picks the methods that cover them.

• A tool permission table that says allow, deny or ask a person, with the reason
• A policy file, the fields to log for each tool call and a test list
• A pass or fail line for each of the 10 OWASP agent risks, with evidence and a fix
• A file hash record of your tools and packages, with a verified or tampered result when you recheck
• A verdict on each finding's evidence
• The open risks ranked High, Medium or Low, in the order to fix them

How it does the job well

• Name the asset, the threat and the impact before recommending controls.
• Rank findings by real risk (likelihood and impact), with evidence for each.
• Prefer fixes that remove a problem over warnings that rely on people.
• Cite the exact law, standard or clause for compliance and legal points, and recommend qualified review.

Mistakes it avoids

• Long lists of generic issues with no evidence they apply.
• Testing or scanning systems without permission.
• Presenting a legal opinion as final.

What a good result looks like

An example, not your task: You are about to launch an agent that can call tools, such as APIs, files or a database.

• Step 1: Agent control review. A tool permission table (allow, deny or ask a human), a policy file, audit log fields and a test list.
• Step 2: OWASP findings. A pass or fail line for each of the 10 OWASP agent risks, with evidence and a fix for each gap.
• Step 3: Plugin integrity record. A file hash manifest for your tools and packages, and a verified or tampered result when you recheck.
• Step 4: Evidence review. A verdict on each finding from steps 2 and 3, and any audit you supply, marked verified, unsupported or not checked.
• Unresolved risks. Open findings ranked High, Medium or Low in fix-first order, every missing fact and one next action.
• Hand-back summary. Source coverage and each step's status in one list.

How the work is checked

• Every fact traces back to a source you were given, or is marked MISSING:.
• Missing inputs are clear, and no gap was filled with a guess.
• The result matches the stated goal.
• Proposed actions stay apart from completed actions. Nothing was run, sent or changed.
• States which checks passed, failed or could not run, and names missing inputs.

These checks have not been run against a model yet. Your AI uses them to check its own work.

Capabilities

• AI agent control designer. Lists every tool the agent can call and what it can do with it. Sets allow, deny or ask a person for each tool, and writes the policy, log fields and tests. Hands the plan to step 2.
• Agent OWASP risk checker. Checks your agent code against the 10 OWASP agent risks, using the plan from step 1. Gives a pass or fail line for each risk, with evidence and a fix.
• Agent plugin integrity checker. Works on your code and package files, not on step 2. Records a file hash for your tools and packages, checks how packages are pinned and flags files that changed.
• Security audit evidence reviewer. Checks the findings from steps 2 and 3, and any audit you supply, against the files they cite. Marks each claim verified, unsupported or not checked.

How the role works

1. Checks your task, context and available tools.
2. Picks the methods needed and prepares a work plan.
3. Carries out supported tasks, with each method's limits and review gates.
4. Checks the real deliverables and corrects supported faults.
5. Hands back files, evidence and unresolved gaps.

Your specialist's working setup

Download the base specialist without a task form or account. README.md and SETUP.md explain how to use the folder. SOUL.md defines its scope. INTAKE.md asks missing questions inside your AI. PLAYBOOK.md, SKILLS.md, BEST_PRACTICES.md, CHECKS.md and HANDOFF.md guide the work and its checked deliverables.

Complete associated skills, supporting files and licences are included. Choose only the methods needed for the actual task. Keep context, working files and accepted feedback in your AI project.

Use this specialist through MCP or the API

Call prepare_specialist with role_id review-an-agent-before-launch only. No task is required to load it. Your AI asks what job you need done, what you want back and which relevant context it can use.

For task-led hiring, call find_specialist, explain the suggested specialists and let the user accept or change the team.

Complete ZIP: Download base specialist. API: GET /api/v1/specialists/review-an-agent-before-launch/pack. The host AI supplies its own model, tools and execution; a pack does not connect accounts or run a task.

Inputs and helpful context

The specialist starts from the task brief. Only active methods require their inputs.

For the primary method, when used

• What the agent does, in one or two sentences
• Each tool the agent can call, and whether it reads or changes things
• The risk level (internal test, general production, sensitive such as money, health or legal, or compliance-critical) and the data it can touch
• Your agent code, tool list and package files
• A security audit or findings list you already have (optional)

Tools and access

• Any good chat AI: Claude, ChatGPT or Gemini can turn a pasted tool list into the table and policy file.
• Your code editor with an AI helper: It can place the code patterns in your project once the policy is agreed.
• Claude Code: Opens your repository and checks each cited file and line itself.
• Any good chat AI: Works if you paste the report and the code lines it cites.

Functions and deliverables

• AI agent launch security reviewer: Share agent code, design, tools and packages. Get permission, OWASP and dependency findings with an evidence check and ranked fixes.
  • A tool permission table that says allow, deny or ask a person, with the reason
  • A policy file, the fields to log for each tool call and a test list
  • A pass or fail line for each of the 10 OWASP agent risks, with evidence and a fix
  • A file hash record of your tools and packages, with a verified or tampered result when you recheck
  • A verdict on each finding's evidence
  • The open risks ranked High, Medium or Low, in the order to fix them
• AI agent control designer: Set tool permissions, approval rules and an audit log for your AI agent. Get a policy file and a test list before it goes live.
  • A tool table with allow, block or ask-a-human for each tool
  • A policy file in YAML with call limits and blocked text patterns
  • A list of risky requests to flag, with their limits stated
  • The fields to record in your audit log
  • A test list with one test per control
  Source: GitHub Copilot community.
• Agent OWASP risk checker: Check your AI agent code against the OWASP agent security risks. Get a pass, fail or not confirmed for each of 10 risks, with evidence and fixes.
  • A count of how many of the 10 risks look covered
  • A pass, fail or not confirmed for each risk with the code that shows why
  • A short list of the biggest gaps
  • A fix to try for each gap
  Source: GitHub Copilot community.
• Agent plugin integrity checker: Check your agent plugin folder for changed, missing or new files. Get a hash manifest and a list of loose dependency versions.
  • An INTEGRITY.json file with a hash for every file in the plugin
  • A VERIFIED or FAILED result that lists changed, missing and new files
  • A list of loose package versions, with how to pin each one
  • A ready or not ready answer for going live
  • A GitHub Actions step you can paste into your build
  Source: GitHub Copilot community.
• Security audit evidence reviewer: Check a security audit before you trust it. Find findings with no proof, dropped findings and missing coverage, then get a pass or rework verdict.
  • A verdict of pass, rework or deliver with a flagged gap
  • A table that marks each finding as verified, unsupported or wrong location
  • A list of findings the report dropped or softened, with whether the reason holds
  • A list of areas the audit did not cover
  • A 1 to 10 score in each quality category
  • A short list of fixes before you share the report
  Source: GitHub Copilot community.
• me.md builder: Answer a few questions, or let your AI use what it remembers. Get one private file that tells any AI who you are.
  • One me.md file with up to ten dated sections
  • A list of facts taken from memory, for you to check
  • A short note of what it removed for privacy
  • The two next steps, to save it and set it up

Tasks you can use it for

• You are about to launch an agent that can call tools, such as APIs, files or a database.
• You need to decide which tools the agent may use, which are blocked and which need a person to approve.
• You want the agent's code and packages checked before real users reach it.
• You were given a security audit of the agent and want to know which claims have proof.

Limits to check before starting

• Your agent has no tools and only writes text: there is nothing to lock down, so skip this.
• You want to attack the agent and see if it breaks: use the AI agent attack tester.
• You want to watch the agent once it is live: use the AI agent monitoring planner.
• Your concern is the API behind the agent: use API security repair reviewer.

Best way to use this role

Give it one clear outcome, the actual files and a definition of done. Keep the work in one AI project so it can reuse checked context and earlier files.

• Download the pack. It holds HANDOFF.md and four skill files, in the order they run: AI agent control designer, Agent OWASP risk checker, Agent plugin integrity checker, then Security audit evidence reviewer.
• Share the agent's job, its tool list, and its code and package files.
• Ask your AI to follow HANDOFF.md and mark any missing fact.
• Review the ranked risks yourself before launch. Nothing is deployed or changed for you.

Read it as data

Read the complete role setup as JSON. Connect your AI to the AI Blocks MCP library and read aiblocks://roles/review-an-agent-before-launch for its role instructions and pinned sources.

This is a prepared AI role. No task has run. Your AI carries out authorized work using its available tools and reports the actual deliverables and checks.