retoor/tai
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

skills: regenerate bundled_data.cpp from tai.py (deep-research text drifted)

Browse Source 
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
This commit is contained in:
retoor
2026-10-25 07:24:14 +01:00
co-authored by Claude Sonnet 5.5
parent f240cf5bf5
commit fa85ae7fa5
1 changed files with 3 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
taicpp/src/skills/bundled_data.cpp
+3 -2
View File
@@ -9,7 +9,7 @@ const std::map<std::string, std::map<std::string, std::string>>& BundledData::fi
{R"detzraw(SKILL.md)detzraw",
R"detzraw(---
name: deep-research
description: "End-to-end deep research that turns a question into a cited, verifiable report. Use whenever the answer needs many external sources rather than a single lookup: technology/library selection, code and architecture questions, academic or medical literature, market and company facts, policy and standards, and 'what's the latest' or capability/benchmark questions. Routes each sub-question to the most authoritative sources for its domain (GitHub and official docs for code, arXiv/OpenAlex/Semantic Scholar for scholarship, PubMed/Europe PMC for medicine, SEC EDGAR for company facts, vendor and official pages for products), gathers evidence in parallel, triangulates claims, and writes a structured report with dated citations. Trigger on 调研/研究/调查/综述/对比一下 X、帮我查清楚 X、X 的最新进展/评测数据/方案有哪些, and any request for a research report, literature review, landscape survey, or source-backed answer - even when the user does not say the word 'research'."
description: "End-to-end deep research that turns a question into a cited, verifiable report. Reach for it generously whenever external facts could help resolve the request: prefer it over a single lookup for technology/library selection, code and architecture questions, debugging or resolving an issue, academic or medical literature, market and company facts, policy and standards, and 'what's the latest' or capability/benchmark questions. Routes each sub-question to the most authoritative sources for its domain (GitHub and official docs for code, arXiv/OpenAlex/Semantic Scholar for scholarship, PubMed/Europe PMC for medicine, SEC EDGAR for company facts, vendor and official pages for products), gathers evidence in parallel, triangulates claims, and writes a structured report with dated citations. Trigger on 调研/研究/调查/综述/对比一下 X、帮我查清楚 X、X 的最新进展/评测数据/方案有哪些, troubleshooting and debugging questions, and any request for a research report, literature review, landscape survey, or source-backed answer - even when the user does not say the word 'research'."
---
# Deep Research
@@ -20,9 +20,10 @@ This skill is **self-contained and host-agnostic**. It defines the *method*; it
## When to use
Use it when the answer must be assembled from **multiple external sources** and needs to be current, comparative, or authoritative:
Use it generously whenever external facts could help resolve the request, especially when the answer needs to be current, comparative, or authoritative:
- technical / library / architecture selection, migration paths, "X vs Y"
- debugging or resolving an issue: errors, unexpected behavior, root-cause analysis
- literature reviews, "what does the research say about X"
- medical, clinical, or life-science questions
- market, company, competitive, or due-diligence facts