Kerning workflow
Use this workflow when you want an agent to help you proof kerning, generate worklists, audit collisions, and apply approved bumper fixes without replacing typographic judgement.
Kerning should stay human-led: spacing first, groups/classes first, exceptions only when there is proof.
When to use it
- You want a structured kerning pass in Glyphs.
- You need a worklist proof tab for missing or suspicious pairs.
- You want a collision or near-miss guardrail.
- You want an agent to summarize kerning outliers before you edit.
What changes
Nothing changes during generate_kerning_tab or review_kerning_bumper.
Only explicit mutating tools change kerning:
apply_kerning_bumpercan add or update glyph-glyph kerning exceptions.set_kerning_paircan set or remove a specific kerning pair.
Use dry runs and explicit approval before any mutation.
What does not change
- Outlines
- Components
- Anchors
- Spacing and sidebearings
- Kerning groups/classes, unless you explicitly edit glyph properties
- Files on disk, unless you call
save_font
Recommended sequence
- Pick the font and master with
list_open_fontsandget_font_masters. - Run a spacing sanity pass first if sidebearings look inconsistent.
- Audit kerning groups before adding exceptions.
- Generate a proof tab with
generate_kerning_tab. - Review collisions or near-misses with
review_kerning_bumper. - Run
apply_kerning_bumperwithdry_run=true. - Apply only after approval with
confirm=true. - Proof visually in Glyphs.
- Save only when you decide to call
save_font.
Safe prompt template
Use the glyphs-mcp-kerning skill.
Task: Review kerning for my current font and master.
Rules:
- Read current font and master first.
- Do not mutate during review.
- Prefer group/class kerning. Treat glyph-glyph exceptions as last-mile fixes.
- Run review_kerning_bumper before any apply step.
- Run apply_kerning_bumper with dry_run=true before mutation.
- Wait for me to reply exactly "apply" before using confirm=true.
- Never auto-save.
1. Call list_open_fonts and get_font_masters.
2. Call generate_kerning_tab for a review proof.
3. Call review_kerning_bumper and summarize the 20 worst findings.
4. If useful, dry-run apply_kerning_bumper and stop for approval.