docs(cursor): enhance documentation for rule configuration on Cursor

[chrux]

🔗 Linked Issue

Closes #106

---

🏷️ PR Type

• type:bug — Bug fix
• type:feature — New feature
• type:docs — Documentation only
• type:refactor — Code refactoring (no behavior change)
• type:chore — Maintenance, dependencies, tooling
• type:breaking-change — Breaking change

---

📝 Summary

• Refined and expanded guidelines on managing persistent memory protocols within Cursor
• Emphasizing best practices for configuring .cursor/rules/engram.mdc
• Improved examples and clarified file paths, including Windows considerations

📂 Changes

File	Change
docs/AGENT-SETUP.md	Replace legacy .cursorrules guidance with .cursor/rules/ documentation

🧪 Test Plan

• Manually tested the affected functionality

Manual verification:

• Confirmed the docs now reflect the current Cursor rule layout
• Verified the updated instructions locally in Cursor

---

✅ Contributor Checklist

• I linked an approved issue above (Closes #N)
• I added exactly one type:* label to this PR
• Docs updated to reflect the current behavior

💬 Notes for Reviewers

This is a docs-only update to bring the Cursor guidance in line with the current .cursor/rules/ approach.

[Alan-TheGentleman]

Thanks for the docs work. The content itself looks useful, but I am not going to merge a PR that skips the contribution contract. Open or link a real issue, get it approved, and add the correct type label. After that, push an updated PR body and we can review it on the merits instead of on preventable process failures.

[Alan-TheGentleman]

Maintainer Review — Approved with Direct Edits

Thanks @chrux — the direction here is correct: moving away from the deprecated .cursorrules approach to the proper .cursor/rules/ MDC format. The content intent is good.

I've pushed a fixup commit directly onto your branch (8536926) to address the issues found during review before merging. Here's the full verdict:

---

What Was Wrong / Risky in the Original PR

Issue 1 — Run-on callout in the Cursor section
The Memory Protocol callout was a single run-on sentence that buried three pieces of information: the filename, the two placement options (project vs global), and the Windows path. Hard to scan, easy to miss the global option entirely.

Issue 2 — Missing version caveat
The .cursor/rules/ MDC format was introduced in Cursor 0.43. Without calling that out, users on older Cursor versions will be confused when the rules directory doesn't behave as documented.

Issue 3 — Missing deprecation note
The PR replaced the .cursorrules reference without any note that the legacy file is still recognized by Cursor. Existing users with .cursorrules might think their setup is broken. Fixed by adding: "The legacy .cursorrules file at the project root is still recognized by Cursor but is deprecated."

Issue 4 — Implied filename constraint
The PR said engram.mdc as if that specific name was required. In reality, Cursor picks up any .mdc file in .cursor/rules/. The fix clarifies that engram.mdc is a recommendation, not a requirement.

Issue 5 — alwaysApply: true frontmatter unexplained
The Surviving Compaction Cursor block showed the frontmatter without explaining what it does. A user copying this would wonder why there's YAML at the top. Added a one-line explanation before the code block.

Issue 6 — No language hint on the code fence
The Surviving Compaction block used a plain fence. Changed totext so markdown renderers don't try to interpret the --- frontmatter as a page break.

---

Files Changed

File	Action
docs/AGENT-SETUP.md	Modified — maintainer fixup commit 8536926 on contributor branch

---

Status

✅ Mergeable after CI refreshes. All process requirements are satisfied: Closes #106 is in the body, issue #106 has status:approved, and the PR has the type:docs label. The new head commit will trigger a fresh CI run. Once the PR Validation checks pass, this is ready to merge.

[Alan-TheGentleman]

Thanks for the contribution. The original direction was correct, and I pushed a maintainer docs fixup on top to tighten the Cursor guidance so it is accurate, explicit about deprecation, and easier to follow. The process issues are now resolved and the checks are green, so this is good to merge.