glyphs-mcp-italic-first-pass

Glyphs MCP italic first pass

Use this skill for a guarded Roman-to-italic first pass. Its goal is to help a designer begin an emphasis companion to the Roman. It accelerates construction; it does not create, replace, or claim optical completion of a designed italic.

Core rules

• Use the typed candidate workflow for every italic first pass. Preview and session review do not alter the font; materialization and acceptance require dry run, exact targets, and explicit approval before confirmation.
• Read current font, master, and selection before mutation.
• Resolve an explicit Roman-to-italic master pair for every requested master. Do not interpret "all masters" as one source/target pair.
• Default angle is 12.0 degrees unless the user specifies another value.
• Interpret angle as a Glyphs source/Transformations angle: positive values lean Latin outlines to the right. In exported OpenType/UFO metadata, the corresponding post.italicAngle / slnt value is negative (+12 in Glyphs source convention maps to about -12 in exported font convention).
• Before outline work, verify the target italic master's italicAngle from get_font_masters equals +angle within 0.01; do not use slantAngle, which reports the separate postscriptSlantAngle custom parameter.
• If the target master italicAngle differs, run set_master_italic_angle with dry_run=true, show the before/after, and only set it with confirm=true after explicit approval.
• Candidate acceptance preserves target topology. Before preview, every target master layer must therefore be an untouched Roman-compatible bootstrap copy, not an empty layer. The candidate may change approved coordinates, anchors, component placement, and metrics; it must not create topology.
• Copy components as live components, but do not skew component transforms or component outlines; component shapes should resolve from their own italic master layers.
• Default compatibility_mode is preserve_if_possible; path compatibility is useful but not required.
• Before slant_mode="cursivy", run review_master_stem_metrics for the target italic master.
• Default to slant_mode="balanced". Use Cursivy only when the designer explicitly requests the legacy Transformations-backed construction or needs comparison with an existing Cursivy workflow. Balanced builds Raw, applies the pure-Python conservative correction at partial strength, interpolates using curve_strength, and then applies the independent final stem_compensation. It never calls the Glyphs Transformations filter.
• Balanced defaults to curve_strength=0.75 and stem_compensation=1.0. Pass slant_mode="balanced" explicitly because the tool's omitted-argument compatibility default remains Cursivy.
• The Inter/Noto Sans/IBM Plex Sans validation preserved topology and source layers, accepted 135/102/102 stem pairs, and passed because every non-commuting component construction was blocked before application.
• For reflected, rotated, or non-uniformly scaled component constructions, cycles, unreadable transforms, or component-master mismatches, stop after review and ask the designer to inspect or rebuild the construction. Do not silently omit it or claim that the generated direction is safe.
• Do not exclude glyphs by name or Unicode category. Use the reported outline and recursive component analysis. If the user explicitly approves a smaller safe scope, rerun with skip_glyphs.
• Treat the symbol slant tiers below as Unicode-aware review prompts, never as automatic exclusions or transformations. Report affected glyphs and ask the designer whether to include, exclude, or defer them.
• If Cursivy stems are missing, ask whether to set stems, measure suggestions, use raw slant, or stop.
• Call preview_italic_first_pass_candidate to generate detached proposals.
• Diagnostic preview partitioning is read-only and does not require approval. Approval is required only before mutation or before explicitly omitting a diagnosed glyph from the final pass.
• Call review_outline_candidate_session, then accept_outline_candidate_session with dry_run=true and the exact review token before any mutating call.
• Only accept after explicit approval with confirm=true and the unexpired, fingerprint-bound token.
• Never auto-save the font.
• Materialize only when the user wants to edit the proposed layers manually. Re-review exact edits and accept with the newly issued token. Never recreate or delete candidate layers through arbitrary code or LLM memory.