Fixed-income / credit analysis skills for DeepSeek Harness (dsh) — rating-migration early warning and rate-scenario analysis, ported from the LLMQuant financial Agent Skills library.
固收 / 信用分析技能包：发行人评级迁移预警 + 利率情景分析，为 DeepSeek Harness 生态补齐固定收益方向（当前生态仅有股票技术分析插件）。
| Skill | What it does | 对应分析场景 |
|---|---|---|
rating-migration-watch |
Classify an issuer as upgrade candidate / stable / downgrade watch / crossover (fallen angel, rising star) by triangulating fundamentals trajectory, market-implied spread signals, and migration catalysts. | 信用评级迁移预警、坠落天使/新星识别 |
rate-scenario-analysis |
Build 3–5 historically anchored rate scenarios (parallel shift, steepener/flattener, policy & inflation repricing), trace them across tenors, and estimate duration-based exposure impact. | 利率情景分析、债券持仓久期影响测算 |
Each skill encodes: when to use it, the data contract (required inputs, freshness reporting, missing-data fallback), a 5-step workflow, a fixed output format, and anti-hallucination guardrails (e.g. never present a migration view as an agency action, never claim a market-implied gap without peer-cohort spreads).
Clone into your dsh user skill root:
mkdir -p ~/.dsh/skills
git clone https://github.com/liyc-sys/dsh-fixed-income-skills.git /tmp/dsh-fi
cp -r /tmp/dsh-fi/skills/rating-migration-watch /tmp/dsh-fi/skills/rate-scenario-analysis ~/.dsh/skills/Or point dsh at this repo's skills/ directory via customSkillDirs:
git clone https://github.com/liyc-sys/dsh-fixed-income-skills.git ~/dsh-fixed-income-skills
# in your dsh config:
# skill-filesystem: { customSkillDirs: ["~/dsh-fixed-income-skills/skills"] }Then invoke by name in a dsh session, e.g. skill(rating-migration-watch) — or just ask "is XYZ Corp likely to be downgraded" and the model will pick it up from the catalog.
The workflow content originates from the LLMQuant/skills library (llmquant-credit and llmquant-rates-fx categories), to which these two workflows were contributed upstream. This repo repackages them in dsh's native skill format (frontmatter + bundle layout) so they are discoverable by the dsh skill registry without any conversion layer.
MIT