jas0nh/eva-title-skill

EVA Title Cards

Render locally with the vendored original layout engine. Do not call an image-generation model or recreate layout geometry.

Slash-command contract

Treat the instruction appended to a /eva invocation as command arguments.

• Run node scripts/eva-command.mjs --command '<arguments>' --output-dir /tmp/eva-output.
• For --help or a bare /eva, return the JSON text and its exact MEDIA: value. This generates the 16-layout help image locally.
• Default to e1 when no layout is supplied.
• Split manual | segments before removing punctuation and whitespace. For e1, users specify [top, vertical, horizontal]; the command wrapper converts it to vendor order.
• Without |, choose natural semantic segments when the layout needs multiple inputs. Pass them with --segments '["segment 1","segment 2"]'. Preserve every cleaned title character exactly once and in order. If no semantic split is supplied, the wrapper uses a balanced fallback.
• Return the exact MEDIA: value from successful JSON so Hermes sends the PNG as a native attachment.

Examples:

node scripts/eva-command.mjs --command '--e24 测试' --output-dir /tmp/eva-output
node scripts/eva-command.mjs --command '--e1 顶部|竖排|横排' --output-dir /tmp/eva-output
node scripts/eva-command.mjs --command '--e1 领导喜欢安排泡汤局' \
  --segments '["领导","喜欢安排","泡汤局"]' --output-dir /tmp/eva-output
node scripts/eva-command.mjs --command '--help' --output-dir /tmp/eva-output

Rendering rules

Confirm that FOT-Matisse Pro EB and open-source Source Han Serif SC Heavy are installed in the current user's macOS Fonts folder. Convert simplified Chinese with the upstream mapping, then use Matisse per glyph and Source Han Serif Heavy as fallback.

air and e24 require one semantic segment. e25, e12, e3, e25-2, e26, anno-kandoku, e15, and do-you-love-me require two. All other layouts require three.

Use scripts/render-eva-title.mjs directly only when the caller already provides vendor-order --texts. Verify title PNGs are 640×480 and nonempty before sharing them.