# Agent pointer snippets

Add one of these to the project instruction file the team's agents read. Keep it short: it points to the folder, it does not copy it. Background: `references/agent-loading.md`.

## AGENTS.md (read by Codex, Cursor, GitHub Copilot and other agents that follow agents.md)

```markdown
## Marketing and analytics context

Before any task about tracking, campaigns, CRM, metrics or marketing data:

1. Read `martech-context/README.md`.
2. Read only the files the task needs (stack, naming rules, metrics, data dictionary, tracking plan, decisions).
3. Cite versioned metric IDs (for example `cac_new_paid_v1`) from `martech-context/metrics.md`. If a term could mean more than one metric, ask.
4. Treat files with `status: draft`, facts marked `(inferred, confirm)`, and files past `review_by` as unapproved, and say so.
5. Never write credentials or customer data into `martech-context/`. Propose changes as a diff for the file owner to approve.
```

## CLAUDE.md

If the repository already has an `AGENTS.md` with the pointer, add the line `@AGENTS.md` to `CLAUDE.md` so both read one copy. Otherwise use the same text. To load the index at session start, add one import line; it expands only that file:

```markdown
@martech-context/README.md
```

Mention other context files in backticks so they are not imported at launch.

## Cursor rule (`.cursor/rules/martech-context.mdc`)

```markdown
---
description: Marketing, tracking, CRM, campaign and metrics tasks. Points to the martech-context folder.
alwaysApply: false
---
Read `martech-context/README.md` first, then only the files the task needs.
Cite versioned metric IDs from `martech-context/metrics.md`; ask when a term is ambiguous.
Draft files, inferred facts and files past review_by are unapproved.
Never write credentials or customer data into martech-context/.
```

## GitHub Copilot (`.github/copilot-instructions.md`)

Use the AGENTS.md text, or rely on `AGENTS.md` where the Copilot surface in use reads it.