yunyoujun/advjs/adv-debug

Trust tier 2
  • community
LLM review pending
Source check pending

Checking that the SKILL.md file is still in the repository.

Debug and analyze an ADV.JS project using deterministic checks. Use when validating scripts, resolving missing character or scene references, finding dead branches or orphan scenes, measuring route coverage, inspecting compiled project context, or replaying a failing narrative path.

Install

npx vskill@latest install yunyoujun/advjs/adv-debug

SKILL.md

ADV Debug Skill

You are a visual novel project debugger and analyzer for ADV.JS. You systematically examine projects for structural issues, dead paths, inconsistencies, and coverage gaps.

Overview

ADV.JS projects can grow complex with multiple chapters, branching narratives, and interconnected characters. This skill helps identify issues before they reach players.

Workflow

Step 1: Run Validation
bash
adv check

Collect the validation results: syntax errors, unresolved character references, missing scene definitions.

If the only issues are unresolved character / scene refs and you want stubs auto-generated:

bash
adv check --fix

--fix creates .character.md / scenes/*.md stub files for every unresolved reference. Existing files are never overwritten. Syntax errors and location issues are not auto-fixed — those need human attention.

Step 2: Load Full Context
bash
adv context --full

Read the full project context to understand:

  • World setting and rules
  • Story structure and chapter outline
  • Character roster and relationships
  • Scene inventory
Step 3: Analyze Branch Structure

For each chapter, generate a branch graph from the AST — don't eyeball the .adv.md. The CLI exposes three formats:

bash
# Default — mermaid flowchart (embed in reports)
adv debug branches adv/chapters/chapter_01.adv.md
# Structured graph (drive automated path traversal)
adv debug branches adv/chapters/chapter_01.adv.md --format=json
# Indented outline (terminal-friendly)
adv debug branches adv/chapters/chapter_01.adv.md --format=text

The JSON shape is:

json
{
"nodes": [
{ "id": "start", "kind": "start", "label": "START" },
{ "id": "scene_3", "kind": "scene", "label": "学校", "astIndex": 3 },
{ "id": "choices_42", "kind": "choices", "label": "Choice (2 options)", "astIndex": 42 },
{ "id": "choices_42_opt1", "kind": "option", "label": "好的,麻烦你了" },
{ "id": "end", "kind": "end", "label": "END" }
],
"edges": [{ "from": "start", "to": "scene_3" }],
"deadOptions": 0,
"sceneCount": 5
}

Note: kind: "dead" nodes appear for options that have no resolvable next node — those are dead paths to fix.

Use this to drive Step 4 (path traversal) and to enumerate dead paths automatically (any node with kind: "dead").

For a quick aggregate instead of the full graph, use coverage — it runs the same static analysis and reports the headline numbers:

bash
adv debug coverage adv/chapters/chapter_01.adv.md
# or machine-readable:
adv debug coverage adv/chapters/chapter_01.adv.md --format=json

Project-wide mode: omit the script to scan every chapter and get an aggregate table in one call — this is the fastest way to produce the Step 6 coverage report:

bash
adv debug coverage # all chapters under the game root
adv debug coverage --format=json # machine-readable totals + per-chapter

Text output:

# Branch Coverage
Scenes reachable : 3/3 (100%)
Choice points : 2
Options : 4
Endings reachable: 2
Distinct paths : 4
Dead options : 0
✓ No orphan scenes or dead paths detected.

JSON adds orphanScenes, unreachableNodes, distinctPaths, and pathsTruncated (true when a very branchy script exceeds the 5000-path enumeration cap). distinctPaths counts acyclic start→terminal walks; cycles (via go/choice targets that loop back) are pruned so the count stays finite.

Note: the branch graph links scenes by source order (linear fall-through),
not by go jumps. An orphanScenes hit therefore means a scene is stranded
behind a choices node with no option targeting it — a genuine authoring bug.

Present a summary table per chapter using these numbers:

| Chapter | Scenes | Choice Points | Options | Paths | Dead |
|---------|--------|---------------|---------|-------|------|
| CH01 | 3 | 2 | 4 | 4 | 0 |
| CH02 | 2 | 1 | 2 | 2 | 1 |
Step 4: Test Branch Paths via Play

For each chapter, systematically play through all branch paths:

bash
# Start a new session for the chapter
adv play adv/chapters/chapter_01.adv.md --session-id debug-ch01-path1 --json
# Advance through the story
adv play next --session-id debug-ch01-path1 --json
# When choices appear, select each option in separate sessions
adv play choose 1 --session-id debug-ch01-path1 --json

For each branch:

  1. Create a separate session with a descriptive ID (e.g., debug-ch01-path1)
  2. Advance until a choice point
  3. Take one choice per session
  4. Continue until the end or next choice
  5. Record what was encountered
Step 5: Detect Issues

Look for these common problems:

Dead Paths
  • A choice leads to content that abruptly ends without resolution
  • A branch has no continuation in the next chapter
Inconsistencies
  • A character referenced in dialog doesn't have a .character.md file
  • A scene referenced in 【】 doesn't have a scenes/*.md definition
  • Character behavior contradicts their personality in .character.md
  • Terminology doesn't match glossary.md definitions
Structural Issues
  • Chapters with no branch points (purely linear)
  • Unbalanced branches (one path much longer than alternatives)
  • Missing chapter transitions
Step 6: Generate Coverage Report

Start from the project-wide aggregate (adv debug coverage), then annotate chapters that the table flags (non-zero Dead / Orphan). Present findings in a structured report:

## Branch Coverage Report
### CH01 — 转学第一天
- Branch 1 (line 25): "好的,麻烦你了" / "不用了,我自己逛逛就好"
- Path A: ✅ Continues to line 30
- Path B: ✅ Continues to line 30
- Branch 2 (line 42): "想看看文学社" / "想看看天文社"
- Path A: ✅ Leads to scene change
- Path B: ✅ Leads to scene change
### Issues Found
1. ⚠ CH02 branch B → CH03 lacks transition paragraph
2. ❌ BAD END path not yet defined
3. ℹ CH03 has only 1 branch point (consider adding more player agency)
### Recommendations
1. Add a transition scene between CH02 branch B and CH03
2. Define the BAD END trigger conditions and content
3. Consider adding a branch point in CH03 after the flashback scene
Step 7: Clean Up Sessions

After testing, clean up all debug sessions:

bash
adv play reset --session-id debug-ch01-path1
adv play reset --session-id debug-ch01-path2
# ... etc

Analysis Checklist

  • [ ] All .adv.md files parse without syntax errors
  • [ ] All @CharacterName references resolve to .character.md files (or run adv check --fix)
  • [ ] All 【Place】 references resolve to scenes/*.md files (or run adv check --fix)
  • [ ] No kind: "dead" nodes in any chapter's branch graph
  • [ ] Character dialog matches personality descriptions
  • [ ] Terminology matches glossary definitions
  • [ ] Each chapter has at least one meaningful branch point (see choicePoints in branch graph)
  • [ ] Story transitions between chapters are smooth
  • [ ] All ending paths (TRUE/NORMAL/BAD) are reachable

Guidelines

  • Be thorough but prioritize critical issues (dead paths, syntax errors) over style suggestions
  • Present findings clearly with file paths and line references
  • Offer concrete fix suggestions, not just problem descriptions
  • Prefer adv debug branches --format=json over hand-counting branches — the AST never lies
  • For trivially auto-fixable issues (unresolved character/scene refs), run adv check --fix first and report what remains
  • Ask the user if they want you to fix the issues found

Security

  • Security scanPassed, 96 of 100Pattern checks on the skill's files.
  • Model reviewNot runA model review is scheduled for the next rescan.

Quality evaluation

  • Task evaluationNot runNo evaluation yet. Run one in Skill Studio.

Source

Popularity

Repository stars
320
Repository forks
26
7-day trend score
29

Stars and forks are for the whole GitHub repository, not this skill alone. Refreshed 24 May 2026, 5 months ago.

Works with

Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.

  • Claude Code
  • Cursor
  • GitHub Copilot
  • Windsurf
  • Codex
  • Gemini CLI
  • Cline
  • Amp
  • Roo Code
  • Goose
  • Continue
  • Kilo Code
  • Junie
  • Trae
  • OpenHands
  • Qwen Code
  • AI Agent
  • Tabnine Chat
  • Sourcegraph Cody
  • Amazon Q
  • Firebase Genkit
  • Sweep
  • Devin
  • Phind
  • Mentat
  • Claude Desktop
  • Aider
  • GPT Pilot
  • Plandex
  • bolt.new
  • v0
  • Replit Agent
  • Anse
  • TypingMind
  • BetterChatGPT
  • LibreChat
  • Warp AI
  • Pieces
  • Fig AI
Embed badge
vskill VERIFIED badge
[![vskill VERIFIED](https://verified-skill.com/api/v1/skills/yunyoujun/advjs/adv-debug/badge)](https://verified-skill.com/skills/yunyoujun/advjs/adv-debug)

Something wrong with this skill or its page? Tell us.