Store Fixer

Store Fixer

Implement Shopify store changes safely and reversibly via the Admin GraphQL API.

Bundled Resources

Read progressively — only load what the current task needs:

• references/access-setup.md — credential paths, token fetch, scope reference
• references/graphql-mutations.md — exact mutation syntax and gotchas
• references/theme-fixes.md — Tier 2 Liquid code and theme backup
• references/safety-and-rollback.md — canonical rollback manifest schema
• assets/fix-plan-template.md — approval step format
• assets/fix-summary-template.md — delivery format

What It Can Fix

Tier 1 — Data fixes (safe, API-only)

Tier 2 — Theme fixes (requires write_themes, higher risk)

Tier 2 always requires backup before editing. See references/theme-fixes.md.

Critical Rules

1. Never write without explicit user approval. Show exact changes, wait for "yes." If scope changes mid-execution, stop and re-ask.
2. Rollback is mandatory. Save current values before every write. Format: references/safety-and-rollback.md. If rollback cannot be prepared, stay in Plan mode.
3. Use GraphQL, not REST — except theme asset reads where REST is acceptable as fallback. See references/graphql-mutations.md for rate limits and mutation costs.
4. Do NOT generate content by default. Present the fix plan first, then ask: "Should I generate content, or do you want to provide it?"
5. Tier 2 requires explicit opt-in. Never modify theme files unless specifically asked.
6. One batch at a time. Complete one fix category, verify, then proceed.
7. Report every failure. Show userErrors immediately. On error, stop the current batch and ask how to proceed.
8. Execute only approved scope. Do not modify fields or resources outside the approved fix plan.

Workflow

1. Confirm task mode

• Execute: Changes will be made (after approval).
• Plan: Implementation plan only, no writes.
• Review: User already changed the store, verify the work.

2. Classify the fix channel

• Data (Tier 1): Admin GraphQL API mutations.
• Theme (Tier 2): Liquid template edits. Read references/theme-fixes.md.
• Mixed: Both. Confirm both access paths before starting.
• Manual: Requires merchant-only settings or third-party app config. Provide exact manual steps.

3. Confirm access

Read references/access-setup.md. Verify with a shop query (see graphql-mutations.md). If access is missing, switch to Plan mode and tell the user exactly what they need.

4. Present fix plan

Use assets/fix-plan-template.md. Include exact counts, mutations, and scopes. Wait for approval.

5. Save rollback manifest

Query and save current state for every resource you'll modify. Use the canonical format in references/safety-and-rollback.md. Save to local file: store-fixer-rollback-{domain}-{timestamp}.json.

6. Execute in batches

Process 10-20 mutations per batch. For 50+ items, use bulkOperationRunMutation (see graphql-mutations.md). Check userErrors on every response. On error, stop and report.

7. Verify

Data: Follow-up read on 2-3 resources to confirm values changed. Theme: Fetch the modified page HTML and check the target element is present/changed.

8. Deliver summary

Use assets/fix-summary-template.md. Include changes, failures, rollback file path, and any remaining manual steps.

Rollback

1. Read the rollback manifest
2. Show what will be reverted
3. Wait for approval
4. Execute reverse mutations using old_value from each entry
5. For theme files: restore original_content or delete created files via themeFileDelete
6. Verify

Content Generation Guidelines

When the user approves AI-generated content:

Product descriptions: 150-300 words. Write for shoppers. Include materials, sizing, care if derivable from product data. HTML with <p> and <ul> tags.

SEO titles: Format {Product Name} - {Key Attribute} | {Brand}. Under 60 chars. Note: character counts are generation constraints, not audit findings — the store-analyzer correctly ignores character-count violations because Google rewrites freely.

SEO descriptions: 120-155 chars. Include a benefit and product category.

Image alt text: Describe what's in the image. Include product name and distinguishing feature. If you can't see the image, use: "{Product Title} - Image {n}".

Examples

Example 1: Fix the issues from this audit report → Parse report, classify findings, present fix plan, execute with approval.

Example 2: Tell me what access the client needs before we can fix product SEO and theme schema → Plan-only response with exact access requirements and scopes.