SKILL.md
When to use
- Building or updating the “伏笔台账”
- Author edits cause deviations and you need to check whether setups were broken or payoffs drifted.
Output files (human-readable)
artifacts/ledger/setuppayoffledger.md: SSOT structured ledger entries (author editable)artifacts/ledger/reports/<runid>setuppayoffreport.md: derived scan report (auto-generated)
Entry schema (v1 deterministic JSON block)
setuppayoffledger.md must contain a `json code block like:
{
"entries": [
{
"id": "promise-a",
"type": "promise",
"title": "A promises to reveal X",
"setupPattern": "I will tell you X",
"payoffPattern": "X is revealed",
"deadlineChapter": 12,
"regex": false,
"severity": "MAJOR"
}
]
}Fields (recommended):
- id: stable id (unique)
- type: setup / promise / debt (free-form for now)
- title: one-line human title
- setupPattern: keyword or regex to locate the setup
- payoffPattern: keyword or regex to locate the payoff
- deadlineChapter: optional “should be paid by chapter N”
- regex: true to treat patterns as regex
- severity: MINOR / MAJOR (free-form for now)
Heuristics (assistant mindset)
- A setup is “broken” if later text contradicts its promise without an intentional twist.
- A payoff is “weak” if it resolves mechanically without emotional/meaningful consequence.
- “Upgradable” setups are those with high emotional stakes but low information cost.
Procedure
- Scan story outline + existing chapters for explicit promises and implicit debts.
- Normalize them into ledger entries (avoid duplicates).
- Mark conflicts introduced by author deviations.
- Generate 2-3 payoff roadmaps:
- conservative (minimal edits) - aggressive (rewrite/upgrade to a bigger twist) - branch option (if it risks breaking the mainline)