Skip to content
alyn
gamedev-skills

Visual Novel skill

Build a visual novel: a branching script, character and background display, a text box with choices, save/load, backlog, and skip/auto. Use for a VN, dating sim, or branching story game.

$npx skills add gamedev-skills/awesome-gamedev-agent-skills --skill visual-novel
GitHub stars
8,175
Pack installs
748k
Alyn Score
77/99
Updated
9 h ago
License
Apache-2.0

Install Visual Novel

Pick your agent. Every command is ready to paste.

$npx skills add gamedev-skills/awesome-gamedev-agent-skills --skill visual-novel

The open skills CLI detects your agents (Claude Code, Cursor, Codex and 70+ more) and installs to the right folder. Add -g to install globally.

01

Install it

Run the command above, or upload the folder in Claude.

02

Ask naturally

Describe the task. Your agent loads visual-novel when the request matches its description.

03

Name it if needed

If it does not trigger, say “use the visual-novel skill”. In Codex, type $visual-novel.

The skill itself

SKILL.md

GitHub

Visual Novel

A playbook for visual novels — the branching script, the presentation (text box, characters, backgrounds), choices, and the quality-of-life systems players expect (save anywhere, backlog, skip, auto). This is a compositional skill: it drives a dialogue engine and a UI layer. It does not re-teach the dialogue engine or UI nodes; it defines the script model and the player conveniences that make a VN pleasant to read.

When to use

  • Use when the game is mostly reading branching text with character art and backgrounds: visual novel, dating sim, branching interactive fiction, story-choice game.
  • Use when designing a choice/route structure, story flags, or VN conveniences (backlog, skip, auto-advance, save-anywhere).

When not to use: dialogue as one feature inside a larger game → rpg consuming dialogue-systems. Card/board play → other genres. For the branching-script engine itself, use dialogue-systems (Ink / Yarn Spinner).

If the project is Ren'Py (current stable 8.5.3): the engine already ships save-anywhere, rollback, a history (backlog) screen, skip that stops at unseen text by default (preferences.skip_unseen = False), auto-forward and a text-speed preference. Use and restyle those through its screens and preferences; apply this skill's design guidance (routes, flags, reactivity) rather than rebuilding the systems in Patterns 2–3.

Core loop

Read a line → advance → (at a branch) make a choice → the story branches on flags/choices → read on → reach an ending. The "game" is the shape of the branching and whether choices feel consequential; everything else is presentation and convenience.

Must-have systems

  1. Branching script — ordered lines + choices + jumps, with conditions and variables (Ink/Yarn).
  2. Text box — speaker name, body text, typewriter reveal, advance on click/key.
  3. Characters — sprites with expressions/poses, positions, show/hide transitions.
  4. Backgrounds + transitions — scene images, fades/dissolves.
  5. Choices — present options, gate some on flags, record the pick.
  6. Story state — flags/variables that branch the script and unlock content.
  7. Save/load (save-anywhere) — full script position + state; multiple slots; quick save.
  8. VN conveniences — backlog/history, skip (read text), auto-advance, text-speed setting.
  9. Audio — music per scene, SFX, optional voice clips.

Design knobs

Knob Effect Notes
Text speed / instant reading comfort Always allow instant + a skip.
Auto-advance delay hands-free reading Tunable; pause on choices.
Skip scope re-reading Skip read text only by default.
Branch breadth/depth replay value vs. cost Branches multiply writing/art work.
Flag-gated content reactivity Lines/choices that check past decisions.
Route structure story shape Branch-and-merge vs. distinct routes (refs).
Choice visibility fairness Show locked choices vs. hide them.
Backlog length convenience Keep enough to re-read recent context.

Patterns

1. Script as data the engine walks

# Pseudocode. Lines, choices, and jumps as data — usually authored in Ink/Yarn and stepped
# through by that runtime. The engine asks the script for "the next thing to show".
node = script.current()
if node.kind == "line":
    show_text(node.speaker, node.text)        # wait for advance input
elif node.kind == "choice":
    options = [o for o in node.options if condition_met(o.condition, flags)]  # gate by flags
    show_choices(options)                      # wait for selection
elif node.kind == "set":
    flags[node.var] = eval_expr(node.expr, flags)
script.advance(selected_option_or_none)

2. Typewriter reveal + advance (skippable)

# Pseudocode. Lay out the whole line once, then reveal it; a press first completes the line.
def show_text(speaker, text):
    name_label.text = speaker
    body_label.text = text                     # full text set up front: wrapping is decided now
    shown = 0.0
    while shown < visible_length(body_label):  # count displayed characters, not markup tags
        if advance_pressed():                   # first press: reveal the whole line instantly
            break
        shown += chars_per_second * frame_dt()
        body_label.visible_characters = int(shown)
        wait_one_frame()
    show_all_characters(body_label)
    backlog.append((speaker, text))             # once, when the line is complete
    wait_for_advance()                          # second press: go to the next line

Reveal with the label's visible-character count, not by slicing the string: slicing cuts markup tags in half and re-wraps the text every frame, so a word starts on one line and jumps to the next. Godot 4.7: RichTextLabel.visible_characters (or visible_ratio) with visible_characters_behavior = TextServer.VC_CHARS_AFTER_SHAPING; the default VC_CHARS_BEFORE_SHAPING leaves hidden text out of line breaking, so words still jump. Unity 6.3: TMP_Text.maxVisibleCharacters (TextMeshPro, part of the uGUI package).

3. Choice sets a flag that branches later content

# Pseudocode. Choices write flags; later conditions read them — that is "reactivity".
def on_choice(option):
    if option.set: flags[option.set] = True     # e.g. flags["helped_npc"] = True
    script.jump(option.target)                   # follow the branch

# Elsewhere, a line/choice/ending checks the flag:
if flags.get("helped_npc"): play_route("good_ending") else: play_route("neutral_ending")

Pitfalls / failure modes

  • Save that only stores a checkpoint → VNs need save-anywhere. Persist the exact script position and all flags/variables (and seen-text data) so a load resumes the same line. Check what your runtime can restore: Ink saves the exact position, Yarn Spinner only the node (refs §4).
  • Typewriter text that jumps between lines or shows raw tags → the reveal slices the string. Set the full text and raise the visible-character count (Pattern 2).
  • Presentation logic baked into the script → unmaintainable. Keep content (text, choices) in the script and how it looks (sprites, transitions) in the engine layer.
  • No skip/auto/backlog → readers feel trapped, especially on replays. These are expected baseline features, not extras.
  • Skipping unread text → players miss content. Skip should fast-forward read text only.
  • Choices with no consequence → branches that reconverge instantly feel fake. Set flags that visibly change later lines, choices, or endings.
  • Combinatorial branch explosion → unshippable. Prefer branch-and-merge with a few flagged variations over fully distinct trees (refs).
  • Lost reading context → no backlog to re-read the last lines. Keep a history buffer.
  • Hardcoded language → no localization path. Keep text in data keyed for translation.

Composition (build it from these skills)

  • Script engine: dialogue-systems (Ink / Yarn Spinner) — branching, conditions, variables, localization hooks.
  • Presentation: game-ui-ux for text-box/choice-menu layout, scaling, and safe areas; godot-ui-control for the concrete text box, choice menu, name plate, and backlog UI.
  • Persistence: save-systems for save-anywhere slots, seen-text/skip data, and settings.
  • Audio: audio-design for per-scene music, SFX, and voice playback.
  • Visuals: the engine animation/Tween skill for sprite/background transitions; shader-programming for dissolves.
  • Process: prototype-fast to test the branch structure in plain text before adding art.

References

  • For the branching data model, route structures (branch-and-merge vs. routes), flags/variables, save-anywhere + backlog/skip data, and the content/presentation split, read references/script-and-flow.md.

Tips for Visual Novel

  • • Start a fresh session after installing so your agent picks up the new skill.
  • • It bundles reference files. The agent reads them only when needed, so they cost no context until then.
  • • It is part of a pack with 40 skills. Install only the ones you need with --skill to keep triggers precise.
  • • If two skills overlap, mention the one you want by name.

Questions about Visual Novel

What does the Visual Novel skill do?

Build a visual novel: a branching script, character and background display, a text box with choices, save/load, backlog, and skip/auto. Use for a VN, dating sim, or branching story game.

How do I install Visual Novel?

Run "npx skills add gamedev-skills/awesome-gamedev-agent-skills --skill visual-novel" in your project. The skills CLI installs it for Claude Code, Cursor, Codex and 70+ other agents. In Claude on the web, upload the skill folder as a ZIP under Customize > Skills.

Does Visual Novel work in Cursor and Codex?

Yes. Visual Novel uses the open SKILL.md format, so it works in Claude Code, Claude, Cursor, OpenAI Codex, Gemini CLI, GitHub Copilot and other agents that support agent skills.

Is Visual Novel free?

Visual Novel is open source on GitHub under the Apache-2.0 license. It is free to install; check the license before commercial use.

Who made Visual Novel?

Visual Novel is published by gamedev-skills in the gamedev-skills/awesome-gamedev-agent-skills repository, together with 39 other skills.

Similar skills. From other creators.