gold-news-sentiment

---

name: gold-news-sentiment

description: Use this skill when users want to pull recent global gold-related news, assess short-term or medium-term market sentiment for gold, connect macro drivers like Fed policy, US yields, USD, inflation, geopolitics, and ETF flows to gold, and produce a structured conclusion such as 看涨, 看跌, or 观望 with confidence and risks.

---

# gold-news-sentiment

Use this skill to turn recent gold-related news into a structured market sentiment read.

The default job is not to produce a trading call from vibes. The default job is to:

When To Use

Use this skill when the user wants to:

Core Rules

1. Treat the output as analysis support, not investment advice.

2. Prefer high-credibility and recent sources over volume.

3. Do not let repeated headlines count as multiple independent signals.

4. Separate `news sentiment` from `price trend`. They often diverge.

5. Always state uncertainty, key assumptions, and what could invalidate the conclusion.

Workflow

1. Check for a fresh cached snapshot

Before doing a fresh pull, check whether these files already exist:

If `data/latest_sentiment.md` is fresh enough for the user's need, use it first so the user gets an immediate answer. Default freshness window:

If the cache is stale, missing, or the user explicitly asks for a refresh, continue with a fresh pull.

For the recurring workflow, read [references/automation-template.md](references/automation-template.md).

2. Pull recent news

Run the bundled script from the skill directory:

python3 scripts/fetch_news.py --hours 48 --limit 40

Use `--query` when the user wants a narrower theme such as:

The script outputs normalized JSON with:

For source coverage and caveats, read [references/source-list.md](references/source-list.md).

For cached fast-path updates, prefer:

python3 scripts/update_snapshot.py --hours 48 --limit 50

This refreshes:

The automation can then write:

3. Remove weak signals

Before analysis:

If the fetch step returns zero usable items or includes request failures, do not fabricate a sentiment conclusion. Report that retrieval failed or that the evidence is insufficient, then stop or ask for a narrower rerun.

If more detail is needed, read [references/scoring-rules.md](references/scoring-rules.md).

4. Classify each item

Classify each article into one of:

Then note the driver category:

Do not force a directional label when the causal chain is weak.

5. Aggregate into a market read

Build the conclusion in this order:

1. short-term news balance

2. macro driver alignment or conflict

3. whether price-sensitive drivers are pointing the same way

4. remaining uncertainty

Default horizon:

If the user does not specify a horizon, provide both.

6. Produce the final output

Use this structure unless the user asks for another format:

结论：观望

情绪方向：
- 短线：偏多
- 中线：中性

置信度：中

核心驱动：
- <driver 1>
- <driver 2>
- <driver 3>

主要新闻：
1. <headline summary + why it matters>
2. <headline summary + why it matters>
3. <headline summary + why it matters>

为什么不是明确看涨/看跌：
- <reason>

风险提示：
- <risk 1>
- <risk 2>

Output Discipline

Validation Checks

Before answering, verify:

Automation Guidance

For recurring runs, the automation prompt should ask for:

When `data/latest_sentiment.md` exists and is fresh, answer from it first unless the user asks for a real-time refresh.

Avoid claiming certainty. If the signal is mixed, say so.