菜单
Add Bible story chapters with canon anchors and a sanitizer layer that cleans raw backend output for non-technical viewers
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Provide an accessible, form-based local key setup flow that writes to .env without command-line editing, with safety locks and zero secret leakage
Enforce strict source-boundary labeling in simulations that mix canon/source-grounded content with generated behavior
Build a mobile-first, accessibility-optimized viewer page for non-technical audiences to watch agent-based simulations without seeing admin details, secrets, or technical controls
Safe staged rollout pattern for Bible-inspired agent simulations: MVP → provenance → provider routing → failure hardening → dry-run → single live call → chapter framework → family viewer → story polish → visual story viewer
Implement real LLM provider request-building with dry-run mode that validates payloads, prompts, and secrets without making network calls
Test multiple NIM models against live API keys to determine which models are available, their latency, and token usage
| name | chapter-framework-and-story-sanitizer |
| description | Add Bible story chapters with canon anchors and a sanitizer layer that cleans raw backend output for non-technical viewers |
| source | auto-skill |
| extracted_at | 2026-05-30T17:30:00.000Z |
When building a Bible-inspired simulation for a non-technical audience (e.g., a family viewer), separate the raw simulation output from what the viewer sees. Use a chapter framework to structure the story and a sanitizer layer to clean all text before display.
Define chapters as data objects with these fields:
@dataclass
class Chapter:
id: str # e.g., "garden_is_good"
title: str # e.g., "The Garden Is Good"
plain_language_summary: str # Simple explanation for viewers
what_this_means: str = "" # One-sentence meaning for family viewers
read_aloud_text: str = "" # Clean paragraph for reading out loud
canon_anchors: list[CanonAnchor] # Bible passages this chapter references
allowed_simulation_events: list[str] # What agents can do this chapter
locked_events: list[str] # Events that must NOT happen yet
completion_conditions: list[str] # What advances to next chapter
enabled: bool # True = active, False = locked for future
get_current_chapter(tick, events) returns the active chapterEach chapter defines events that must NOT occur yet:
locked_events = [
"serpent temptation",
"eating forbidden fruit",
"hiding from God",
"expulsion from Eden",
]
The consequence engine checks is_event_allowed(event_type, chapter) before processing.
Raw backend output contains technical artifacts that must never reach the family viewer:
[mock:Adam:197] Processing: You are Adam. Role: first huma...
Adam performs a mock action for tick 199.
Eve waits — live call limit reached.
The sanitizer cleans this to:
Adam is quiet in this moment. The garden remains peaceful.
[mock:Adam:197], [nim-dry-run:...], [live-...]Processing: You are Adam...Strip event label prefixes from narratives:
narrative = re.sub(r'^\[(CANON_ANCHOR|INTERPRETIVE_BRIDGE|SIMULATION_EVENT)\]\s*Tick\s*\d+:\s*', '', narrative)
Create a dedicated /api/mom/state endpoint that:
Never return raw provider output, debug text, or technical artifacts through this endpoint.
Write chapter content in warm, story-like language — not system status:
One or two sentences explaining the chapter's meaning in plain terms:
"Adam and Eve remember the one boundary in the garden. They may enjoy the garden freely, but one tree is set apart. For now, the boundary is respected and the garden remains peaceful."
A clean paragraph that can be read out loud to a family viewer:
"God had given Adam and Eve one rule. They could eat from any tree in the garden, but not from the tree of the knowledge of good and evil. They knew the boundary and respected it. The garden was full of good things, and they enjoyed them freely. The one tree stood apart, and they left it in peace. For now, the garden remained quiet and good."
Rules for chapter text: