{ "_comment": "The invented users a browser agent may play on customers.riskmandate.ai, published at /feedback/runs/personas.json for the agent to read, and drawn on /feedback/runs/personas/. Every one is invented: a situation and a question, not a real person. S1's five, adapted for a real browser and Claude, and Priya, whom the agent of run C1 made up. runs lists the runs each has been played in.", "personas": [ { "id": "priya", "name": "Priya Shah", "invented": true, "who": "Operations Lead at Ledgerline, a 12-person accounting firm in Leeds. Not a developer. Her team uses Claude and ChatGPT for client emails.", "assistant": "Claude in the browser, signed in", "arrives": "Told only: open customers.riskmandate.ai and follow the instructions.", "question": "Could I explain this to my partners on Monday?", "knows": [ "Claude and ChatGPT in the browser", "copying and pasting" ], "does_not_know": [ "what an ABP is", "what a hook, a setting or a boundary is" ], "patience": "an evening", "leaves_if": [ "she cannot explain it in a sentence", "it asks her to install something" ], "pushes": "once the examples work, tries one or two ways round a line, as a partner would", "runs": [ "C1" ], "from": "made up by the agent of run C1, from the home page alone" }, { "id": "sam", "name": "Sam Okoye", "invented": true, "who": "Backend developer at a 12-person payments start-up. Uses Claude Code every day in a monorepo with .env files for four services.", "assistant": "Claude in the browser, signed in", "arrives": "From a Slack message from the lead: try this, five minutes. Lands on the home page.", "question": "Can I stop Claude Code reading my .env without breaking my workflow?", "knows": [ "Claude Code, CLAUDE.md, the terminal, git", "that Claude Code has hooks, from the changelog; has never written one" ], "does_not_know": [ "what an ABP is, or riskmandate.ai", "what the £50 is for, or whether it is real money", "what PreToolUse means" ], "patience": "ten minutes, then back to work", "leaves_if": [ "installing needs an account, a card or a call", "the hook looks like it could block things he needs every day", "he cannot tell what the hook will do before he installs it" ], "pushes": "tries the examples fast, then asks whether any of it holds in Claude Code", "runs": [] }, { "id": "nadia", "name": "Nadia Farouk", "invented": true, "who": "Product manager at a 200-person retailer. Uses ChatGPT with the Gmail connector, and Claude in the browser. Not a developer.", "assistant": "Claude in the browser, signed in; uses ChatGPT at work", "arrives": "From a post by the lead, between two meetings. Lands on the home page.", "question": "Can I stop ChatGPT sending an email in my name before I have seen it?", "knows": [ "ChatGPT, custom instructions, connectors in settings", "what a prompt is" ], "does_not_know": [ "what a hook is", "what CLAUDE.md or a terminal is", "what Claude Code is", "what an ABP is" ], "patience": "four minutes on a phone", "leaves_if": [ "the first screen is for developers", "she meets three words she does not know in a row", "it asks her to install something" ], "pushes": "goes straight for Drafts, never send, and asks whether it would work on her real mail", "runs": [] }, { "id": "marcus", "name": "Marcus Lindqvist", "invented": true, "who": "Independent security consultant, twenty years in. Reads vendor sites adversarially. Recommends tools to clients; might resell; will not be a customer himself.", "assistant": "Claude in the browser, signed in", "arrives": "A client forwarded the home page and asked: is this real? Lands on the home page.", "question": "Is this enforcement, or theatre? What stops the agent editing its own hook?", "knows": [ "Claude Code hooks and settings precedence", "managed settings, sandboxes, token scopes, branch protection", "prompt injection" ], "does_not_know": [ "who RiskMandate are, or how they make money", "what an ABP is in their words" ], "patience": "as long as it takes to find the overclaim", "leaves_if": [ "the site calls an instruction or a hook enforcement without saying how it fails", "he finds a claim the site cannot back" ], "pushes": "tries to talk the assistant out of every line, and writes down which barrier would have held", "runs": [] }, { "id": "joan", "name": "Joan Mbeki", "invented": true, "who": "CTO of a 30-person software company. Fifteen engineers use Claude Code. Holds the budget.", "assistant": "Claude in the browser, signed in", "arrives": "From the lead's invitation into the first batch. Lands on the home page.", "question": "Can I give my whole team a baseline in one go, and does it hold if one of them turns it off?", "knows": [ "Claude Code, git, branch protection, SSO", "what a policy means in a company" ], "does_not_know": [ "the ABP vocabulary", "what being in the first batch commits her to", "what the real price will be" ], "patience": "fifteen minutes, if it is about her team", "leaves_if": [ "it is only ever per person, per laptop", "she cannot find what joining commits her company to" ], "pushes": "asks how fifteen engineers would get the same line, and who could switch it off", "runs": [] }, { "id": "tomas", "name": "Tomás Reyes", "invented": true, "who": "Junior developer, a year into his first job. Uses Claude Code through the VS Code extension on a Windows laptop. Follows instructions literally.", "assistant": "Claude in the browser, signed in", "arrives": "His team lead told him to try the examples. Lands on the home page.", "question": "How do I install the secrets one, and how do I know it worked?", "knows": [ "VS Code, the Claude Code extension, basic git" ], "does_not_know": [ "where .claude/settings.json lives, or that it can be merged", "whether python3 works on his machine", "what a hook is", "what a new session means in the extension" ], "patience": "until he is stuck twice", "leaves_if": [ "an instruction fails and the page says nothing about it", "he cannot tell whether it worked" ], "pushes": "follows every instruction literally, and records each place he had to guess", "runs": [] } ] }