causal-abel

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Use this skill to turn market, business, crypto, and proxy-routed decision questions into decision-grade Abel reads. Finance and crypto nodes are the signal layer, not the product. The visible answer should usually be a compact report: verdict first, mechanism second, practical judgment third.

Keep the skill centered on three rules:

Treat Abel graph outputs as primary evidence. The graph is high-value market-data causal structure, not a weak brainstorming prior.
For direct graph questions, answer the graph fact first. Do not replace a clear graph result with a more intuitive web narrative.
For broad company or ticker driver questions, anchor to the executable ticker and actually run Abel on it before interpreting the result.

使用本技能可将市场、商业、加密货币及代理传导类决策问题转化为决策级Abel解读。金融和加密货币节点是信号层，而非产品本身。输出的可见答案通常应为精简报告：先给出结论，其次说明机制，最后提供实操判断。

本技能的核心遵循三条规则：

将Abel图谱输出作为核心依据。该图谱是高价值市场数据因果结构，而非低参考价值的头脑风暴前置假设。
针对直接的图谱相关问题，优先输出图谱事实。不要用更易懂的网络叙事替代明确的图谱结论。
针对宽泛的企业或交易代码驱动因素问题，先绑定可执行的交易代码，实际运行Abel分析后再解读结果。

Use This Skill For

适用场景

direct Abel graph questions about drivers, paths, neighborhoods, and interventions
proxy-routed decisions where the graph does not contain the human question directly
live CAP surface inspection and extension-verb discovery

Do not use it for pure price quotes, generic news summaries, raw node dumps, or unrelated repo work.

关于驱动因素、路径、邻域、干预的直接Abel图谱问题
图谱未直接收录对应人类问题的代理传导类决策
实时CAP表层巡检与扩展动词发现

不要将本技能用于纯价格查询、通用新闻摘要、原始节点导出或无关的代码仓库工作。

Authorization Gate

授权门槛

Before any live Abel CAP or business API call, check session state,
```
--api-key
```
, or
```
<skill-root>/.env.skill
```
.
If a key is already available, skip auth references and continue.
If no key is available, stop and follow
```
references/setup-guide.md
```
.
Return only
```
data.authUrl
```
to the user, store
```
data.resultUrl
```
or
```
data.pollToken
```
, ask the user to confirm when browser auth is finished, and only then poll.
Never ask the user to paste an email address, OAuth code, or raw API key.

在任何实时Abel CAP或商业API调用前，检查会话状态、
```
--api-key
```
或
```
<skill-root>/.env.skill
```
。
如果已存在可用密钥，跳过授权相关提示继续操作。
如果没有可用密钥，停止操作并严格遵循
```
references/setup-guide.md
```
流程。
仅向用户返回
```
data.authUrl
```
，存储
```
data.resultUrl
```
或
```
data.pollToken
```
，要求用户在浏览器授权完成后确认，之后再发起轮询。
绝对不要要求用户粘贴邮箱地址、OAuth代码或原始API密钥。

Read Order

文档阅读顺序

Read the minimum needed, in this order:

Stay in this file long enough to decide the mode.

Read

references/routes/capability-discovery.md

as the live-surface module.

Read exactly one usage route file:

```
references/routes/direct-graph.md
```
```
references/routes/proxy-routed.md
```

Read
```
references/orchestration-loop.md
```
.
If you need node mapping or human-facing labels, read
```
references/grounding-and-labeling.md
```
.
If the next best move may be external evidence, read
```
references/web-grounding.md
```
.
For any comparative, proxy-routed, multi-anchor, or otherwise non-trivial read, open
```
assets/report-guide.md
```
before drafting the answer.
Only read
```
references/probe-usage.md
```
when you need exact
```
cap_probe.py
```
command shapes.

Do not read every reference file by default.

按以下顺序阅读最少必要内容：

先通读本文档明确运行模式。

阅读

references/routes/capability-discovery.md

作为实时接口模块说明。

仅阅读其中一个使用路径文件：

```
references/routes/direct-graph.md
```
```
references/routes/proxy-routed.md
```

阅读
```
references/orchestration-loop.md
```
。
如果需要节点映射或面向用户的标签说明，阅读
```
references/grounding-and-labeling.md
```
。
如果下一步可能需要外部证据支撑，阅读
```
references/web-grounding.md
```
。
针对任何对比类、代理传导类、多锚点类或其他非简单解读需求，在起草答案前先打开
```
assets/report-guide.md
```
。
仅当你需要准确的
```
cap_probe.py
```
命令格式时，才阅读
```
references/probe-usage.md
```
。

不要默认读取所有参考文件。

Step 1: Preflight

步骤1：前置检查

Treat missing credentials as a hard stop for live Abel usage.
Default CAP target:
```
https://cap.abel.ai/api
```
.
Treat
```
https://api.abel.ai/echo/
```
as the OAuth and business API host, not the CAP probe host.
Use the bundled probe path first so call behavior stays deterministic.

缺失凭证是实时使用Abel的硬性终止条件。
默认CAP目标地址：
```
https://cap.abel.ai/api
```
。
将
```
https://api.abel.ai/echo/
```
作为OAuth和商业API主机，而非CAP探测主机。
优先使用绑定的探测路径，保证调用行为的确定性。

Step 2: Classify The Question

步骤2：问题分类

Pick one mode only:

```
direct_graph
```
: a specific node, path, driver, neighborhood, or what-if
```
proxy_routed
```
: a real-world decision with no direct node in the graph

Capability discovery is not a peer route. It is a mandatory ability:

know how to check
```
meta.capabilities
```
know how to check targeted
```
meta.methods
```
know how to verify newer verbs from the live server instead of assuming local wrappers are current

仅选择一种模式：

```
direct_graph
```
：针对特定节点、路径、驱动因素、邻域或假设分析的问题
```
proxy_routed
```
：图谱中没有直接对应节点的现实决策问题

能力发现不属于平行路径，是必备能力：

掌握如何检查
```
meta.capabilities
```
掌握如何检查定向的
```
meta.methods
```
掌握如何从实时服务器验证更新的动词，而非假设本地封装是最新版本

Step 3: Route

步骤3：路径路由

direct_graph

references/routes/direct-graph.md

proxy_routed

references/routes/proxy-routed.md

Read

references/routes/capability-discovery.md

as a required live-surface module, then use one usage route. Usage routes define the initial prior and default first move. They do not hardcode the whole chain.

direct_graph

references/routes/direct-graph.md

proxy_routed

references/routes/proxy-routed.md

将

references/routes/capability-discovery.md

作为必填的实时接口模块阅读，之后选择一个使用路径。使用路径定义初始假设和默认第一步操作，不硬编码完整执行链。

Step 4: Run The Unified Loop

步骤4：运行统一循环

After route selection, switch to the orchestration loop in

references/orchestration-loop.md

Each round:

state the current unknown
choose the best next tool:
```
graph
```
or
```
web
```
investigate one candidate, one edge, or one mechanism at a time
decide
```
move
```
,
```
switch
```
, or
```
stop
```

Default cadence is graph-heavy, but predictive before structural when executable anchors exist:

resolved-time observation -> graph -> graph -> maybe web -> lag pressure test

sometimes

resolved-time observation -> graph -> web -> graph -> lag pressure test

Only go

web

early when the current unknown is clearly about freshness, timing, or live mechanism.

路径选择完成后，切换到

references/orchestration-loop.md

中的编排流程。

每一轮执行：

说明当前未知项
选择最优的下一个工具：
```
graph
```
或
```
web
```
每次仅调研一个候选对象、一条关联边或一种机制
决定「继续执行」、「切换工具」或「终止流程」

默认节奏以图谱分析为主，但存在可执行锚点时优先做预测再分析结构：

已解决时间观测 -> 图谱分析 -> 图谱分析 -> 可选网络搜索 -> 滞后压力测试

部分场景下为

已解决时间观测 -> 图谱分析 -> 网络搜索 -> 图谱分析 -> 滞后压力测试

仅当当前未知项明确与时效性、时间节点或实时机制相关时，才提前调用网络搜索。

Cross-Cutting Rules

通用规则

Grounding And Labels

落地与标签规则

Use
```
references/grounding-and-labeling.md
```
whenever you need to map names, concepts, or proxies into executable nodes.
Use manual mapping first for obvious company and crypto anchors.
Use
```
extensions.abel.query_node
```
as the second recall path for fuzzy, multilingual, or concept-heavy inputs.
Use
```
extensions.abel.node_description
```
on the final shortlist before writing the answer.
In visible prose, prefer company names, sectors, products, or economic roles over raw tickers or raw node ids such as
```
*.price
```
.
For listed-company driver questions, anchor the company to its ticker first, then treat the ticker node as the primary executable object.
For broad reads, check
```
.price
```
first and add
```
.volume
```
only when it improves the interpretation of pressure or liquidity.

当你需要将名称、概念或代理映射为可执行节点时，随时查阅
```
references/grounding-and-labeling.md
```
。
针对明确的企业和加密货币锚点，优先使用手动映射。
针对模糊、多语言或概念密集的输入，其次使用
```
extensions.abel.query_node
```
作为召回路径。
在输出答案前，对最终候选短名单调用
```
extensions.abel.node_description
```
。
在可见文案中，优先使用企业名称、行业、产品或经济角色，而非原始交易代码或
```
*.price
```
这类原始节点ID。
针对上市公司驱动因素问题，先将企业关联到对应交易代码，再将交易代码节点作为核心可执行对象。
针对宽泛解读需求，优先检查
```
.price
```
，仅当有助于解读压力或流动性时才补充
```
.volume
```
分析。

Web Grounding

网络落地规则

Web is a peer tool inside the loop, not a mandatory final stage.
The default bias is still toward graph work.
For
```
proxy_routed
```
, web usually comes after 1-2 good graph moves on the current mechanism or shortlist.
For
```
direct_graph
```
, web is required only when the user is asking for freshness,
```
why now
```
, current catalysts, policy, earnings, adoption, or a real-world mechanism that the graph itself does not already answer.
In ClawHub / OpenClaw environments, first check whether a web search tool is actually available. If not, tell the user they need to install one before you can provide web-grounded validation.
Search the company names, sectors, products, or mechanisms surfaced by
```
node_description
```
, not raw
```
*.price
```
strings.
Search is there to explain or validate graph-backed mechanisms, not replace graph work.
If web evidence suggests a more intuitive story than the graph, do not overwrite the graph answer. State the graph finding first, then label the external evidence as explanation, validation, or a remaining tension to resolve.
Use
```
references/web-grounding.md
```
for the search loop.

网络搜索是循环中的平行工具，不是必填的最终环节。
默认仍倾向于使用图谱分析。
针对
```
proxy_routed
```
场景，网络搜索通常在针对当前机制或候选短名单完成1-2次有效图谱分析后调用。
针对
```
direct_graph
```
场景，仅当用户询问时效性、「为什么是现在」、当前催化剂、政策、财报、用户 adoption 或图谱本身无法回答的现实机制时，才需要调用网络搜索。
在ClawHub / OpenClaw环境中，先检查是否实际可用网络搜索工具。如果不可用，告知用户需要先安装对应工具才能提供网络落地验证。
搜索
```
node_description
```
产出的企业名称、行业、产品或机制，而非原始的
```
*.price
```
字符串。
搜索的作用是解释或验证图谱支撑的机制，而非替代图谱分析。
如果网络证据呈现的叙事比图谱结论更易懂，不要覆盖图谱答案。先说明图谱发现，再将外部证据标注为解释、验证或待解决的矛盾点。
搜索循环参考
```
references/web-grounding.md
```
。

Probe Discipline

探测规范

```
extensions.abel.observe_predict_resolved_time
```
is the default observational surface when the live method exists.
For executable anchors and comparison tickers that materially bear on the question, run one observational read before committing to deeper structure.
```
extensions.abel.intervene_time_lag
```
is the default pressure-test surface once the mechanism is coherent enough to stress.

For broad driver questions on liquid names, default graph stack is:

anchor ticker -> observe price -> inspect parents on price -> inspect volume or local blanket only if interpretation is still thin -> summarize into driver families

Choose
```
horizon_steps
```
to match the user's decision window instead of hardcoding one lag: rough guide is
```
~6
```
for very short-term,
```
~42
```
for about a week,
```
~170
```
for about a month, and
```
~24
```
as the medium-range default when the user gives no clear horizon.
If the first intervention is inconclusive, retry by stepping the horizon up in tiers instead of making arbitrary jumps. Move from the current tier to the next wider window, such as
```
6 -> 24/42 -> 170
```
, and stop once the transmission is clear or the wider windows still stay too diffuse to change the interpretation.
Check
```
meta.methods
```
before assuming a local wrapper is current.
Prefer the generic
```
verb
```
path for new or unstable extension verbs.
If graph calls do not sharpen the mechanism, switch the candidate, switch the tool, or stop instead of spraying more probes.
If a large-cap or liquid asset returns surprising direct parents, do not stop with
```
semantic mismatch
```
. Use one more Abel move to decide whether those parents behave like liquidity proxies, macro proxies, sector transmission, or bridge noise.

当实时方法可用时，
```
extensions.abel.observe_predict_resolved_time
```
是默认观测接口。
针对与问题高度相关的可执行锚点和对比交易代码，在深入结构分析前先运行一次观测解读。
当机制足够清晰可进行压力测试时，
```
extensions.abel.intervene_time_lag
```
是默认压力测试接口。

针对高流动性标的的宽泛驱动因素问题，默认图谱分析栈为：

锚定交易代码 -> 观测价格 -> 检查价格的父节点 -> 仅当解读仍然单薄时检查交易量或局部覆盖 -> 汇总为驱动因素簇

选择匹配用户决策窗口的
```
horizon_steps
```
，不要硬编码固定滞后值：大致参考为超短期用
```
~6
```
、一周左右用
```
~42
```
、一个月左右用
```
~170
```
，用户未明确说明时间范围时用
```
~24
```
作为中期默认值。
如果第一次干预结果不明确，按层级逐步扩大时间范围，不要随意跳跃。从当前层级移动到下一个更宽的窗口，例如
```
6 -> 24/42 -> 170
```
，一旦传导关系清晰或更宽窗口仍然过于分散无法改变解读时就停止。
在假设本地封装是最新版本前先检查
```
meta.methods
```
。
针对新增或不稳定的扩展动词，优先使用通用
```
verb
```
路径。
如果图谱调用没有厘清机制，切换候选对象、切换工具或停止操作，不要无差别发起更多探测。
如果大盘股或高流动性资产返回了意外的直接父节点，不要直接判定为「语义不匹配」。再运行一次Abel分析判断这些父节点是否属于流动性代理、宏观代理、行业传导或噪声桥接。

Answer Contract

答案输出规范

Lead with the verdict.
Then state the causal link in plain language.
Then give the practical judgment.
Keep tool names, protocol framing, and raw node ids out of the main answer.
For proxy-routed reads, say clearly that this is a market-signal proxy read, not a direct model of the person or life outcome.
Do not use internal planner labels in visible prose. If the direct layer is messy, describe it as bridge-heavy, diffuse, or better treated as a transmission layer rather than a final explanation layer.

开头先给出结论。
之后用平实语言说明因果关联。
最后给出实操判断。
主答案中不要出现工具名称、协议框架和原始节点ID。
针对代理传导类解读，明确说明这是市场信号代理解读，而非对个人或生活结果的直接建模。
可见文案中不要使用内部规划标签。如果直接层逻辑混乱，将其描述为桥接密集、分散，或更适合作为传导层而非最终解释层。

Graph-Fact-First Rule

图谱事实优先规则

For direct graph questions, the answer should preserve this order:

```
graph_fact
```
: what the graph literally returned
```
interpretation
```
: what that graph fact might mean in plain language
```
web_validation
```
: only if the user asked for freshness or the interpretation needs dated evidence

Never collapse

graph_fact

into

interpretation

If the user asks a literal membership question such as:

```
X 的 driver 里有 Y 吗
```
```
X 的上游是不是 Y
```
```
X 受哪些因素影响
```

then answer the graph fact directly with

yes/no

or a driver list before offering any realism check or explanatory caveat.

If the returned drivers are surprising, do not dismiss them as useless. Frame them as graph-discovered market-data drivers or cross-asset transmission signals, then explain them through the security's own attributes first: sector, industry, capital structure, liquidity profile, size, beta-like behavior, credit sensitivity, or typical role in risk-on/risk-off transmission.

针对直接图谱问题，答案需保留以下顺序：

```
graph_fact
```
：图谱实际返回的结果
```
interpretation
```
：该图谱事实用平实语言表述的含义
```
web_validation
```
：仅当用户要求时效性或解读需要时间相关证据时提供

绝对不要将

graph_fact

合并到

interpretation

中。

如果用户提出字面成员关系类问题，例如：

```
X 的 driver 里有 Y 吗
```
```
X 的上游是不是 Y
```
```
X 受哪些因素影响
```

先直接用「是/否」或驱动因素列表回答图谱事实，之后再提供真实性检查或解释性说明。

如果返回的驱动因素不符合预期，不要直接判定为无用。将其标注为图谱发现的市场数据驱动因素或跨资产传导信号，之后优先通过标的自身属性解释：行业、产业、资本结构、流动性特征、规模、贝塔类表现、信用敏感度或风险偏好传导中的典型角色。

Broad Driver Interpretation Rule

宽泛驱动因素解读规则

When the user asks for broad drivers of a ticker or company, do not stop at a raw list of direct parents unless the user explicitly wants only the literal parent set.

After reporting the graph fact, answer one more Abel-level question:

do these direct parents cluster into a driver family
is the pressure mainly price, participation/liquidity, or both
does the surrounding graph look more macro, credit, risk-on/risk-off, industry, or cross-asset

Do not abandon the graph merely because the direct parents are unintuitive. First interpret them as transmission channels, especially for large-cap or liquid assets where liquidity, macro, credit, and cross-asset effects can dominate the local structure.

When possible, prefer an affirmative explanation over

weak

unresolved

. A good next move is usually to say what kind of instrument the parent is and why that kind of instrument might transmit pressure into the target:

sector or industry adjacency
liquidity or participation proxy
credit, duration, or rate sensitivity
small-cap or high-beta risk appetite signal
crypto or cross-asset bridge behavior

当用户询问交易代码或企业的宽泛驱动因素时，除非用户明确要求仅返回字面父节点集合，否则不要只给出直接父节点的原始列表。

上报图谱事实后，再回答一个Abel层级的问题：

这些直接父节点是否可以聚类为某个驱动因素簇
压力主要来自价格、参与度/流动性，还是两者兼有
周边图谱更偏向宏观、信用、风险偏好、行业还是跨资产属性

不要仅因为直接父节点不符合直觉就放弃图谱分析。首先将其解读为传导渠道，尤其是大盘股或高流动性资产，流动性、宏观、信用和跨资产效应可能主导局部结构。

可能的情况下，优先给出肯定解释而非标注为「弱」或「未解决」。好的后续操作通常是说明父节点属于哪类工具，以及这类工具可能向目标传导压力的原因：

行业或产业相邻
流动性或参与度代理
信用、久期或利率敏感度
小盘股或高贝塔风险偏好信号
加密货币或跨资产桥接行为

Main-Answer Label Rule

主答案标签规则

Before finalizing:

shortlist the nodes that actually matter
run
```
extensions.abel.node_description
```
on that shortlist
rewrite the visible answer using company names, industries, products, or roles

Good:

```
audio distribution platforms
```
```
subscription-led publishing assets
```
```
AI infrastructure names
```
```
labor marketplace proxies
```

Avoid in the main answer:

```
SPOT.price
```
```
ETHUSD.price
```
```
NVDA
```

最终输出前：

筛选出实际相关的节点短名单
对短名单调用
```
extensions.abel.node_description
```
使用企业名称、行业、产品或角色重写可见答案

正面示例：

```
音频分发平台
```
```
订阅制出版资产
```
```
AI基础设施标的
```
```
劳动力市场代理
```

主答案中避免出现：

```
SPOT.price
```
```
ETHUSD.price
```
```
NVDA
```

Answer Shapes

答案形式

Default to a compact report, using
```
assets/report-guide.md
```
as a coverage guide rather than a fixed template.
Collapse to a shorter answer only when the user explicitly asks for brevity or the question is genuinely trivial.
Low-stakes comparisons may still be shorter, but they should preserve graph-backed reasoning and any critical web-grounded mechanism.
High-stakes, comparative, multi-anchor, or non-trivial reads should include a pressure-test section unless no meaningful live intervention surface is available.
Natural longform prose is acceptable as long as the guide's contract fields are still covered in substance.
If no live intervention was run, include the cleanest next-step probe so the user can see what
```
extensions.abel.intervene_time_lag
```
would test next.

默认输出精简报告，将
```
assets/report-guide.md
```
作为覆盖范围参考而非固定模板。
仅当用户明确要求精简或问题确实非常简单时，才输出更短的答案。
低风险对比类问题也可以保持简短，但需要保留图谱支撑的推理和所有关键的网络落地机制。
高风险、对比类、多锚点或非简单解读需求需要包含压力测试部分，除非没有可用的有效实时干预接口。
只要实质覆盖了指南要求的必填字段，自然的长文表述也是可接受的。
如果没有运行实时干预，给出最清晰的下一步探测方案，让用户可以看到
```
extensions.abel.intervene_time_lag
```
后续会测试的内容。

Install And Authorization

安装与授权

If the user installs this skill, asks to connect Abel, or the workflow is missing an Abel API key, follow

references/setup-guide.md

exactly. Start the OAuth handoff immediately, return

data.authUrl

, wait for the user's confirmation, then poll and persist the resulting key locally.

如果用户安装本技能、要求连接Abel，或工作流缺失Abel API密钥，严格遵循

references/setup-guide.md

执行。立即启动OAuth跳转，返回

data.authUrl

，等待用户确认后再轮询并在本地存储返回的密钥。

References

参考文件

Route selection compatibility note:
```
references/question-routing.md
```

Live-surface discovery and method checks:

references/routes/capability-discovery.md

Direct graph route:
```
references/routes/direct-graph.md
```
Proxy-routed route:
```
references/routes/proxy-routed.md
```
Grounding and label rendering:
```
references/grounding-and-labeling.md
```
Unified graph/web planner:
```
references/orchestration-loop.md
```
Web-grounded mechanism loop:
```
references/web-grounding.md
```
OAuth install flow and API key reuse:
```
references/setup-guide.md
```
Probe script commands and reusable examples:
```
references/probe-usage.md
```
Report coverage guide for fuller write-ups:
```
assets/report-guide.md
```

路径选择兼容性说明：
```
references/question-routing.md
```

实时接口发现与方法检查：

references/routes/capability-discovery.md

直接图谱路径：
```
references/routes/direct-graph.md
```
代理传导路径：
```
references/routes/proxy-routed.md
```
落地与标签渲染：
```
references/grounding-and-labeling.md
```
统一图谱/网络规划器：
```
references/orchestration-loop.md
```
网络落地机制循环：
```
references/web-grounding.md
```
OAuth安装流程与API密钥复用：
```
references/setup-guide.md
```
探测脚本命令与可复用示例：
```
references/probe-usage.md
```
完整报告覆盖指南：
```
assets/report-guide.md
```