Credit Lgd Model

构建并训练 LGD（违约损失率）机器学习模型，支持基于历史违约数据的信用风险量化评估与预测。

Security Scan

The skill claims to build LGD (credit loss) models but its instructions and reference files mix in trading/backtest tooling (ZVT, MACD, trading execution) and filesystem checks, creating a mismatch between stated purpose and what it will actually prompt the agent to do.

Purpose & Capability

The name/description state 'LGD (违约损失率) machine learning model' for credit risk, but SKILL.md, human_summary, seed.yaml and many reference files repeatedly reference ZVT, backtesting, trading pipelines (MACD, trading_execution) and Sphinx documentation. Semantic locks and preconditions include trading rules (execute sell before buy, next-bar execution, MACD param locks) that are unrelated to a pure LGD credit model. This mixed purpose is incoherent: a credit-LGD modelling skill normally would not ask about markets/strategies or enforce trading semantics unless it's intentionally a dual-purpose blueprint.

Instruction Scope

Although instruction-only, the SKILL.md and referenced files direct the agent to run precondition checks that execute Python commands (e.g. 'python3 -c "import zvt..."'), to install packages via pip if checks fail, and to create/write files under a ZVT_HOME (~/.zvt). It also instructs the agent to reload seed.yaml before decisions and to follow an execution protocol. These runtime actions access the host filesystem and package manager and go beyond merely describing modelling steps; they also include trading/backtest question prompts. The instructions are not strictly scoped to LGD estimation and give broad discretion to run installs and filesystem touches.

Install Mechanism

There is no formal install spec (lowest static install risk). However, SKILL.md and seed.yaml explicitly instruct running pip installs and zvt initialization as preconditions if checks fail. Those commands would pull packages from external registries at runtime even though no install recipe is declared in the registry metadata — this is an implicit install path the agent may perform.

Credentials

The skill declares no required environment variables or credentials (good). Still, runtime preconditions reference ZVT_HOME and instruct touching files under that directory (~/.zvt) and checking writability. That implies the agent will read/write local configuration directories. No secrets are requested, but the skill expects filesystem access which is proportionally reasonable for a ZVT/backtest setup but unexpected for a pure LGD model.

Persistence & Privilege

The skill does not request always:true or other elevated persistent privileges. It is user-invocable and allows model invocation (normal). It does not declare modifications to other skills or system-wide agent settings in the metadata. Seed.yaml indicates the host should reload it during execution, but that is a documentation/execution protocol instruction, not a registry-level persistence flag.

What to consider before installing

This package mixes credit LGD model artifacts with trading/backtesting (ZVT/MACD/Sphinx) and directs the agent to run Python checks, pip installs, and create files under ~/.zvt. Before installing or invoking: 1) Confirm whether you actually want a trading/backtest workflow or a pure LGD credit model — the skill appears to attempt both. 2) If you proceed, run it in an isolated environment (VM/container) because the skill may install packages and write to your home directory. 3) Inspect seed.yaml and references/LOCKS.md yourself to verify the enforced semantic locks (trading rules) are acceptable. 4) Ask the author/owner to clarify scope (remove trading/backtest prompts if only LGD is desired) or provide a trimmed skill that only contains credit-modeling instructions. 5) If you lack trust or need least privilege, avoid allowing the agent to run the precondition install commands automatically; perform those steps manually after review.

信用违约损失模型 (credit-lgd-model)

构建并训练 LGD（违约损失率）机器学习模型，支持基于历史违约数据的信用风险量化评估与预测。

Pipeline

data_collection -> data_storage -> factor_computation -> target_selection -> trading_execution -> visualization

Top Use Cases (1 total)

Sphinx Documentation Configuration (UC-101)

This file configures the Sphinx documentation builder for the openLGD project, setting up project metadata, version information, and path configuratio Triggers: documentation, sphinx, configuration

Execute trigger: When user intent matches intent_router.uc_entries[].positive_terms AND user uses action verb (run/execute/跑/执行/backtest/fetch/collect)

What I'll Ask You

• Target market: A-share (default), HK, or crypto? (US stocks in ZVT are half-baked — stockus_nasdaq_AAPL exists but coverage is thin)
• Data source / provider: eastmoney (free, no account), joinquant (account+paid), baostock (free, good history), akshare, or qmt (broker)?
• Strategy type: MACD golden-cross, MA crossover, volume breakout, fundamental screen, or custom factor?
• Time range: start_timestamp and end_timestamp for backtest period
• Target entity IDs: specific stocks (stock_sh_600000) or index components (SZ1000)?

Semantic Locks (Fatal)

ID	Rule	On Violation
SL-01	Execute sell orders before buy orders in every trading cycle	halt
SL-02	Trading signals MUST use next-bar execution (no look-ahead)	halt
SL-03	Entity IDs MUST follow format entity_type_exchange_code	halt
SL-04	DataFrame index MUST be MultiIndex (entity_id, timestamp)	halt
SL-05	TradingSignal MUST have EXACTLY ONE of: position_pct, order_money, order_amount	halt
SL-06	filter_result column semantics: True=BUY, False=SELL, None/NaN=NO ACTION	halt
SL-07	Transformer MUST run BEFORE Accumulator in factor pipeline	halt
SL-08	MACD parameters locked: fast=12, slow=26, signal=9	halt

Full lock definitions: references/LOCKS.md

Top Anti-Patterns (14 total)

• AP-CREDIT-RISK-001: Empty DataFrame passed to bucketing pipeline
• AP-CREDIT-RISK-002: Multi-dimensional target array causing WoE shape mismatch
• AP-CREDIT-RISK-003: OptimalBucketer receiving high-cardinality numerical features

All 14 anti-patterns: references/ANTI_PATTERNS.md

Evidence Quality Notice

[QUALITY NOTICE] This crystal was compiled from blueprint finance-bp-112. Evidence verify ratio = 21.0% and audit fail total = 23. Generated results may have uncaptured requirement gaps. Verify critical decisions against source files (LATEST.yaml / LATEST.jsonl).

Reference Files

File	Contents	When to Load
references/seed.yaml	V6+ 全量权威 (source-of-truth)	有行为/决策争议时必读
references/ANTI_PATTERNS.md	14 条跨项目反模式	开始实现前
references/WISDOM.md	跨项目精华借鉴	架构决策时
references/CONSTRAINTS.md	domain + fatal 约束	规则冲突时
references/USE_CASES.md	全量 KUC-* 业务场景	需要完整示例时
references/LOCKS.md	SL-* + preconditions + hints	生成回测/交易代码前
references/COMPONENTS.md	AST 组件地图（按 module 拆分）	查 API 时

---

Compiled by Doramagic crystal-compilation-v6.1 from finance-bp-112 blueprint at 2026-04-22T13:00:54.441302+00:00. See human_summary.md for non-technical overview.