How do I put a Vindex ready check into my agent's standing instructions?

Put the ready check in the file your agent reads on every task: CLAUDE.md for Claude Code, AGENTS.md for Codex, a rule in .cursor/rules for Cursor, and .github/copilot-instructions.md for GitHub Copilot. Three lines are enough: score before building, proceed on Healthy, report and ask otherwise. Once the rule is in the file, nobody has to remember to say it.

This guide assumes the agent is already connected to the Vindex MCP server. If it is not, the setup pages take about three minutes per client, and the ready check itself explains what the agent sends and what comes back.

Where does each agent read its standing instructions?

• Claude Code reads CLAUDE.md from the repository root at the start of a session, and also picks up AGENTS.md. A user-level ~/.claude/CLAUDE.md applies to every project on the machine.
• Codex reads AGENTS.md from the repository root and its parent directories, and merges a global one from ~/.codex/.
• Cursor reads rule files under .cursor/rules/. A file with alwaysApply: true in its front matter is attached to every Agent request in the project.
• GitHub Copilot reads .github/copilot-instructions.md in Copilot Chat and in the coding agent, so one file covers both.

Put the rule in the repository file, not only the personal one. The point is that every agent working this codebase checks, including the one your colleague runs.

What does the rule say?

One block, the same in every file. The heading makes it easy to find; the numbering makes the order unambiguous to the agent.

## Before implementing a story
1. Score it with the Vindex `score_story` tool. Pass the tracker key as `story_id`.
2. Healthy: proceed. Draft the acceptance checks from the Testable reasoning before writing code.
3. Needs review or At risk: do not write code. Report the finding and the top
   recommendations, and ask the product owner the open questions. Wait for an
   edit or an answer, then score again under the same story_id.

For Cursor, save it as .cursor/rules/vindex.mdc with a short front matter:

---
description: Vindex ready check before implementing any story
alwaysApply: true
---
## Before implementing a story
1. Score it with the Vindex `score_story` tool. Pass the tracker key as `story_id`.
2. Healthy: proceed. Draft the acceptance checks from the Testable reasoning before writing code.
3. Needs review or At risk: do not write code. Report the finding and the top
   recommendations, and ask the product owner the open questions. Wait for an
   edit or an answer, then score again under the same story_id.

Two details matter. “Pass the tracker key as story_id” is what makes every later check free and keeps the agent and the tracker plugin talking about the same story. “Wait for an edit or an answer” stops the agent from resolving the questions itself, which is the failure the rule exists to prevent.

How do I make the threshold the team’s?

The band is Vindex’s suggestion; the rule is yours, and it is plain text. The variations teams reach for:

• Tighten a dimension.“Proceed only when health is healthy and the Testable dimension is 80 or better.” An agent that cannot verify its work will still declare it done.
• Require a tracker key.“If the story has no tracker key, ask for one before scoring.” Nothing gets built from a story that exists only in a chat.
• Exempt what is not a story.“Skip the check for spikes, chores, and bug fixes with reproduction steps.” INVEST is for user stories; scoring a dependency bump wastes a call.
• Decide what “ask” means.In Claude Code and Codex the agent asks in the terminal; in Copilot’s coding agent it can leave the questions as a comment on the issue. Say which.

How do I confirm the rule fires?

Hand the agent a story you know is weak and watch what it does. The Northwind sample story “Warehouse scanning should be faster and better” scores 38%, At risk, through the live server; an agent with the rule in place should score it, report that it has no measurable acceptance criteria and no baseline, list the questions, and write no code. If it starts building, the file is not being read: check the path, the front matter for Cursor, and that the MCP server is connected in that session.

Then hand it a Healthy story and confirm it proceeds, and that the first thing it drafts is the acceptance checks. A rule that only stops is half a rule.

How do I keep the checks free?

A story counts against the company’s monthly allowance once. Every later score of the same story_idis free, which is why the rule insists on the tracker key: the agent’s re-check after the product owner edits the story, and the check the next agent runs on the same story next week, cost nothing. Ask the agent to run get_usage when you want to see where the team stands for the month.