Illustrator Preflight Check

Comprehensive pre-press quality check workflow for Illustrator documents.

Workflow

Execute these 3 tool calls in parallel, then analyze results together:

Step 1: Parallel Data Collection

Run simultaneously:

1. preflight_check — core checks (RGB, links, resolution, text, overprint, transparency, spot colors)
2. get_overprint_info — detailed overprint analysis with intent classification
3. check_text_consistency — dummy text, notation variation detection, and full text dump for LLM analysis

If user specifies a target PDF profile, pass target_pdf_profile: "x1a" or "x4" to preflight_check. If user specifies DPI threshold, pass min_dpi to preflight_check (default: 300).

Step 1b: Conditional Follow-up

If preflight_check reports spot_color warnings, run get_separation_info to get detailed plate information and usage counts. This helps determine whether spot colors are actively used or just leftover swatches.

Step 2: Analyze and Classify

Read references/preflight-rules.md for severity levels and judgment criteria.

Merge results from all 3 tools into a unified report, grouped by severity:

1. Critical — Must fix. Blocks submission.
2. Warning — Review recommended. May or may not need fixing depending on context.
3. Info — Awareness items. No action required unless relevant.

For overprint results, cross-reference get_overprint_info intent classification:

• intentional_k100: Suppress from report (standard practice)
• rich_black_overprint: Include as warning if ink coverage exceeds the limit for the paper type (uncoated: 300%, coated: 350%, newspaper: 240%). If paper type is unknown, ask user or use 300% as default.
• likely_accidental: Escalate to critical

For text consistency:

• Dummy text hits → critical (must replace before submission)
• Notation variations (katakana, fullwidth/halfwidth) → warning
• Use allTexts from check_text_consistency for LLM-driven deeper analysis: look for typos, version mismatches, inconsistent terminology, and any other anomalies that regex patterns would miss.

Step 3: Report

Present results as a structured summary in this order:

## Preflight Results: [document name]

### Critical Issues (X items)
[List with object UUID, description, and recommended action]

### Warnings (X items)
[List with context-dependent guidance]

### Info (X items)
[Brief notes]

### Summary
- Total issues: X critical, X warnings, X info
- Submission ready: Yes/No

For each critical issue that is auto-fixable (e.g., white overprint), offer to fix it immediately.

Context-Dependent Decisions

Some checks require asking the user before acting:

• Non-outlined text: Ask whether the print shop accepts font-embedded PDFs before recommending outlining
• Spot colors: Ask whether spot colors are intentional (special ink printing or should be converted to CMYK)
• Transparency with PDF/X-1a target: Flag as critical; with X-4 or no target, flag as warning
• DPI threshold: If user hasn't specified, use 300 for print, 72 for web/screen

Auto-Fix Capabilities

When the user agrees to fix issues, use these tools:

• Text outlining → convert_to_outlines (irreversible — confirm first)
• Text content replacement → modify_object with contents property (for dummy text fixes)

Not currently auto-fixable (require manual fix in Illustrator):

• White overprint — modify_object does not support fillOverprint/strokeOverprint. Instruct user: select the object and disable overprint in the Attributes panel.
• RGB to CMYK color conversion — requires designer decision on color appearance
• Broken links — requires user to locate and relink original files
• Low resolution images — requires higher resolution source

AI Limitation Awareness

Do NOT add a disclaimer to every report. Instead, apply these rules based on the situation:

Rule 1: Distinguish mechanical checks from AI analysis

When reporting check_text_consistency results, clearly separate the two reliability tiers:

• Mechanical checks (_reliability: deterministic): pattern-matched results (dummy text, fullwidth/halfwidth, katakana long vowel). These are reliable.
• AI analysis (_reliability: ai-assisted): typos, contextual inconsistencies, terminology drift found by LLM analysis of allTexts. These may contain false positives or miss real errors. Label them as AI-based findings.

Rule 2: When no issues are detected

Never say "no issues found — ready to submit." Instead:

• Say "no issues were detected by these automated checks"
• Note that items outside scope (design intent, contextual spelling, regulatory requirements, print-shop-specific rules) still require human review
• The preflight_check tool includes a _note field when all checks pass — relay its message

Rule 3: When the user treats this as a final verification

This rule applies when:

• The user explicitly asks for a go/no-go decision ("Is this ready for submission?", "Can I send this to print?")
• The user's wording implies this is the last step before submission ("最終チェック", "入稿前チェック", "final check", "last check", "これで最後")

In these cases:

• Remind them that automated checks are not exhaustive
• A human must perform the final review
• This does not replace a professional preflight check

Language

Always respond in the user's language.