twg-agentic-search

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

twg-agentic-search

Use together with the root

twg

skill. Exact command grammar comes from live

twg help

, especially

twg help describe "rovo search"

when filter or output options matter.

请与根

twg

技能配合使用。确切的命令语法来自实时的

twg help

，尤其是当涉及筛选或输出选项时，可查看

twg help describe "rovo search"

。

CLI launcher fallback

CLI启动器备选方案

Run

twg <command>

. On shell

command not found

, use

$HOME/.local/bin/twg

(macOS/Linux) /

$env:LOCALAPPDATA\Programs\twg\bin\twg.exe

(PowerShell), then tell user to add that directory to PATH. Do not treat auth or command errors as PATH failures.

运行

twg <command>

。若shell提示“command not found”，请使用

$HOME/.local/bin/twg

（macOS/Linux）或

$env:LOCALAPPDATA\Programs\twg\bin\twg.exe

（PowerShell），然后告知用户将该目录添加到PATH中。请勿将认证或命令错误视为PATH配置问题。

Workflow

工作流程

Classify the request as fuzzy or cross-product internal research. Prefer this skill when the source is unclear, current company knowledge is needed, or the answer may span Confluence, Jira, Drive, Slack, Bitbucket, GitHub, or other Rovo-connected apps.
Confirm or infer the Atlassian site. Ask only when no configured or explicit site is available and the ambiguity would change the search.
If app/source availability changes the plan, run
```
twg rovo list-apps -o json
```
. Use the returned built-ins, connectors, readiness, and auth/setup actions to decide scope; do not start setup or login unless the user asked for it.
Start with one query that combines the concrete topic with the requested artifact or decision type. Use at most one canonical/authoritative or recent/update-oriented refinement when the first result set mixes scopes, lacks primary sources, or misses the requested time signal. Do not fan out exact-title searches for every candidate already returned by the primary query.
Choose filters deliberately. Default to Confluence and Jira built-ins for official/internal knowledge. Broaden to Slack, Google Drive, Bitbucket, GitHub, or other connectors only when useful and available. Use app, type, recency, owner/contributor/assignee/reporter/status, title-only, label/space, and site filters when they narrow evidence without hiding likely answers.
Search with bounded output:

bash

twg rovo search "<query>" --output json --output-summary auto --agent-fields @compact
twg rovo search "<query>" --output json --output-summary auto --agent-fields @evidence

Use

@compact

to shortlist candidates and

@evidence

when snippets, URLs, and provenance need more detail.

将请求归类为模糊搜索或跨产品内部调研。当信息来源不明确、需要当前公司知识，或者答案可能涉及Confluence、Jira、Drive、Slack、Bitbucket、GitHub或其他Rovo关联应用时，优先使用本技能。
确认或推断Atlassian站点。仅当没有配置好的或明确的站点，且这种歧义会影响搜索结果时，才向用户询问。
如果应用/来源的可用性会改变计划，请运行
```
twg rovo list-apps -o json
```
。根据返回的内置应用、连接器、就绪状态以及认证/设置操作来确定搜索范围；除非用户明确要求，否则不要启动设置或登录流程。
先使用一个结合具体主题与所需工件或决策类型的查询。当首次搜索结果范围混杂、缺乏主要来源或未符合所需的时间信号时，最多添加一个权威或最新更新导向的优化条件。不要对主查询已返回的每个候选结果都展开精确标题搜索。
谨慎选择筛选条件。默认针对官方/内部知识使用Confluence和Jira内置应用。仅当有用且可用时，才扩展到Slack、Google Drive、Bitbucket、GitHub或其他连接器。当应用、类型、时效性、所有者/贡献者/经办人/报告人/状态、仅标题、标签/空间和站点筛选条件能缩小证据范围且不会隐藏可能的答案时，再使用这些筛选器。
进行有边界输出的搜索：

bash

twg rovo search "<query>" --output json --output-summary auto --agent-fields @compact
twg rovo search "<query>" --output json --output-summary auto --agent-fields @evidence

使用

@compact

来生成候选结果短列表，当需要更多代码片段、URL和来源细节时使用

@evidence

。

Evidence Rules

证据规则

Treat search snippets as candidates, not facts.
Hydrate a small, diverse primary-source set before final claims. For document discovery, cover distinct roles such as requirements, architecture/design, and current delivery rather than redundant pages. Use product-native commands such as
```
twg confluence content get
```
,
```
twg jira workitem get
```
,
```
twg jira workitem query
```
,
```
twg bb pull-requests get <id>
```
,
```
twg bb repo get
```
, or the relevant product command for the result URL/type.
For document or PRD discovery, select at most five sources across those roles. Hydrate one source per role unless a material conflict requires a second. Stop once the roles, current delivery, and important conflicts are supported, even when search returns more candidates.
Prefer official spaces, owned project pages, current Jira issues, and recent decision records over personal drafts or stale chat mentions, unless the user explicitly asked for informal signal.
Compare hydrated evidence for conflicts, recency, ownership, and authority.
Do not rerun an
```
@evidence
```
search for a candidate that can be hydrated through its product-native URL or ID, and do not refetch one source under another projection. Call out ACL gaps, unavailable connectors, low recall, and unresolved contradictions instead of flattening them into a single claim.

将搜索片段视为候选信息，而非事实。
在做出最终结论前，先获取少量多样的主要来源信息。对于文档发现，覆盖不同角色相关内容，如需求、架构/设计和当前交付，而非重复页面。使用产品原生命令，如
```
twg confluence content get
```
、
```
twg jira workitem get
```
、
```
twg jira workitem query
```
、
```
twg bb pull-requests get <id>
```
、
```
twg bb repo get
```
，或针对结果URL/类型的相关产品命令。
对于文档或PRD发现，在这些角色中最多选择五个来源。除非存在重大冲突需要第二个来源，否则每个角色只获取一个来源。一旦角色、当前交付和重要冲突都有支持信息，即使搜索返回更多候选结果，也停止获取。
优先选择官方空间、所属项目页面、当前Jira问题和近期决策记录，而非个人草稿或过时的聊天提及，除非用户明确要求非正式信息。
对比获取到的证据，检查是否存在冲突、时效性、所有权和权威性。
对于可通过其产品原生URL或ID获取详细信息的候选结果，不要重新运行
```
@evidence
```
搜索，也不要在其他投影下重新获取同一来源。
要指出ACL权限缺口、不可用的连接器、低召回率和未解决的矛盾，而非将它们整合为单一结论。

Output

输出

Lead with the answer or best-supported conclusion. Cite hydrated titles/URLs and include the source app, date or status when available, and why each source was trusted. Separate confirmed facts, likely interpretations, conflicts, and gaps.

以答案或最具支持性的结论开头。引用获取到的标题/URL，若可用则包含来源应用、日期或状态，以及每个来源可信的原因。区分已确认的事实、可能的解读、冲突和信息缺口。