fix: pull writes canonical model files and detects changes reliably

[angeloashmore]

Resolves:

Description

Model comparison now sorts JSON keys at every depth before it compares. Field and tab order stays the same. This corrects three faults:

1. pull rewrote files that had no changes

• Setup: Your model files came from an earlier prismic pull. The API returns nested keys (inside a field's config, for example) in a different order.
• Before: The comparison sorted only the top-level keys. Pull saw a change and rewrote the files. git diff showed only key-order movement.
• After: Pull prints "Already up to date." and does not touch the files.

2. pull did not rewrite files with a non-canonical key order

• Setup: Slice Machine wrote your model files. Its key order is different from the format that the CLI writes.
• Before: The comparison sorted the keys on both sides. Pull saw no difference and kept the old order. A later content change then mixed key-order noise into its diff.
• After: Pull rewrites these files one time, in the canonical format. The next pull prints "Already up to date."

3. sync --watch reported changes that did not exist

• Setup: A poll returns the same models with a different key order.
• Before: The loop generated types and printed "Changes detected in " with an empty list.
• After: Sync hashes the canonical models. It generates types and reports changes only on the initial sync or when a model changed.

push and status continue to ignore key order. They push nothing and report nothing when only the key order is different.

Checklist

• A comprehensive Linear ticket, providing sufficient context and details to facilitate the review of the PR, is linked to the PR.
• If my changes require tests, I added them.
• If my changes affect backward compatibility, it has been discussed.
• If my changes require an update to the CONTRIBUTING.md guide, I updated it.

Preview

How to QA ¹

1. Open a project with committed model files that Slice Machine wrote.
2. Run prismic pull. The command rewrites the files.
3. Run prismic pull again. The command prints "Already up to date." and leaves no dirty files.
4. Run prismic sync --watch. The command does not report changes when the remote models do not change.

🤖 Generated with Claude Code

---

Note

Medium Risk
Touches core model diffing for pull, push, status, and sync; incorrect canonicalization could hide real model changes or skip needed type generation, but behavior is heavily covered by new integration tests.

Overview
Model comparison and on-disk format now treat nested JSON key order as insignificant while preserving field/tab order and slice-zone choice order. canonicalizeCustomType / canonicalizeSlice recurse through configs (Groups, Slices zones, legacy Slice choices) and sortKeys walks arrays and nested objects.

pull compares remote models to already-canonical local files (canonicalize(remote) vs local), so a first pull normalizes Slice Machine–style key order and repeat pulls report “Already up to date.” without noisy rewrites.

sync --watch hashes canonicalized remote payloads, uses the same equality for diffs, runs generateTypes only on initial sync or when models actually change, and logs which resource types changed (e.g. slices only) instead of empty “Changes detected” spam.

Tests cover canonical pull idempotency, status/push staying quiet on key reorder, and sync reporting slice-only updates.

^{Reviewed by Cursor Bugbot for commit 6f8185e. Bugbot is set up for automated code reviews on this repo. Configure here.}

Footnotes

1. Please use these labels when submitting a review:
   ❓ #ask: Ask a question.
   💡 #idea: Suggest an idea.
   ⚠️ #issue: Strongly suggest a change.
   🎉 #nice: Share a compliment. ↩

[cursor]

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

Want reviews to match your repository better? Bugbot Learning can learn team-specific rules from PR activity. A team admin can enable Learning in the Cursor dashboard.

^{Reviewed by Cursor Bugbot for commit d72140d. Configure here.}

[lihbr]

LGTM but I think there's an issue to check with select fields, see bellow :)