القائمة
Use when you need web research, documentation lookup, or knowledge gathering.
يُحدَّد Prompt الذي يراجع المصدر أولًا بشكل افتراضي. يمكنك التبديل إلى أمر مباشر أو تنزيل نسخة محلية.
اقرأ SKILL.md وأي ملفات مرافقة يعرضها SkillsMP قبل أن تقرر التثبيت.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
يتجاوز الأمر المباشر Prompt المخصّص للمراجعة. افحص المصدر قبل تشغيله.
npx skills add https://github.com/seokan-jeong/team-shinchan --skill team-shinchan-researchيبقى الأمر في سطر واحد. مرّر أفقيًا لمراجعته كاملًا قبل النسخ.
تفضّل نسخة محلية؟ نزّل الملفات المتاحة حاليًا لدى SkillsMP.
Use when you want to start a new task with the integrated workflow.
Use when you want autonomous completion from requirements to verification without intervention.
Use when you need backend development for APIs, databases, servers, or endpoints.
استنادا إلى تصنيف SOC المهني
عرض SKILL.md
| name | team-shinchan:research |
| description | Use when you need web research, documentation lookup, or knowledge gathering. |
| user-invocable | false |
If args is empty or only whitespace:
Ask user: "What would you like to research?"
STOP and wait for user response
If args length > 2000 characters:
Truncate to 2000 characters
Warn user: "Request was truncated to 2000 characters"
Parse mode from args:
youtube or article or auto : extract mode as first word, remainder is the URL/query.auto for URLs (args starts with http), else mode = search (standard web search).Set target = the URL/query: the remainder after the mode keyword, or the whole args when there is no mode keyword. (target is referenced by the Step 2A prompt for every mode.)
Modes youtube, article, auto require a URL. If mode detected but no URL found, ask: "Please provide a URL for {mode} extraction."
youtube / article / auto (one URL), OR a search for a single specific fact/definition ("what is X", "X's default port", one API detail).search that is comparative or survey-shaped — triggers include "compare", "vs", "survey", "which … should we", "options for", "best practices for", "investigate", "pros and cons", or anything that benefits from several independent sources.Route: narrow → Step 2A (single Masumi). breadth → Step 2B (orchestrator-worker fan-out).
Task(
subagent_type="team-shinchan:masumi",
model="sonnet",
prompt=`/team-shinchan:research has been invoked.
## Research Request
Mode: ${mode} <!-- youtube | article | auto | search -->
URL/Query: ${target}
Conduct thorough research and provide:
| Section | Content |
|---------|---------|
| Key Findings | Main discoveries with sources |
| Documentation | Relevant docs and reference links |
| Best Practices | Recommended approaches |
| Caveats | Potential concerns or limitations |
User request: ${args || '(Please describe what to research)'}
`
)
A single Sonnet pass under-covers a multi-part question. Anthropic's orchestrator-worker Research beat single-agent by 90.2% because each parallel worker spends a full, isolated context on one facet. The SKILL's main loop owns the orchestration (Masumi has no Task tool):
${args} into 3–5 independent sub-questions, sized to complexity (a 2-way comparison → 2–3; a broad survey → 5).Task(subagent_type="team-shinchan:masumi", model="sonnet",
prompt=`/team-shinchan:research worker. Parent question: ${args}
Sub-question (yours only): ${subq}
Use WebSearch + WebFetch. Return a tight cited brief: Key Findings (with source URLs) | Best Practices | Caveats. Flag low-confidence or conflicting claims explicitly.`)
Task(subagent_type="team-shinchan:masumi", model="sonnet",
prompt=`/team-shinchan:research synthesis. Original question: ${args}
Worker briefs:
${worker_briefs}
Produce ONE cited report — | Key Findings | Documentation | Best Practices | Caveats | — reconciling conflicts across workers and marking any single-source claim as low-confidence.`)
Launch workers in parallel (cap ~5, matching the 3–5 subagent guidance).
If the breadth research is claim-critical (security best practices, a hard-to-reverse vendor/architecture choice, anything feeding an irreversible Stage-2 decision), after Step 2B narrate a one-line opt-in: "This looks claim-critical — want me to run the native /deep-research for adversarial per-claim verification across more sources?" Do NOT auto-fire it; the user launches /deep-research. (Don't reimplement adversarial claim verification here — native deep-research already does it.)