Asosiy tarkibga o'tish

​
7. System Prompt

  1. OpenClaw har bir agent ishga tushirilishi uchun maxsus tizim promptini yaratadi. Prompt OpenClaw’ga tegishli va pi-coding-agent standart promptidan foydalanmaydi.
  2. Prompt OpenClaw tomonidan yig‘iladi va har bir agent ishga tushirilishiga kiritiladi.

​
11. Tuzilma

  1. Prompt ataylab ixcham qilingan va qat’iy bo‘limlardan foydalanadi:
    1. Tooling: joriy asboblar ro‘yxati + qisqa tavsiflar.
    1. Safety: kuchga intilish xatti-harakatlari yoki nazoratni chetlab o‘tishdan qochish uchun qisqa qo‘riqlovchi eslatma.
    1. Skills (mavjud bo‘lganda): modelga ko‘nikma ko‘rsatmalarini talab bo‘yicha qanday yuklashni aytadi.
    1. OpenClaw Self-Update: config.apply va update.run ni qanday ishga tushirish.
    1. Workspace: ishchi katalog (agents.defaults.workspace).
    1. Documentation: OpenClaw hujjatlarining lokal yo‘li (repo yoki npm paketi) va ularni qachon o‘qish kerakligi.
    1. Workspace Files (injected): bootstrap fayllari quyida kiritilganini bildiradi.
    1. Sandbox (yoqilganda): sandboxlangan runtime, sandbox yo‘llari va yuqori huquqli exec mavjud yoki yo‘qligini bildiradi.
    1. Current Date & Time: foydalanuvchi lokal vaqti, vaqt zonasi va vaqt formati.
    1. Reply Tags: qo‘llab-quvvatlanadigan provayderlar uchun ixtiyoriy javob teglari sintaksisi.
    1. Heartbeats: heartbeat prompti va tasdiqlash (ack) xatti-harakati.
    1. Runtime: xost, OS, node, model, repo ildizi (aniqlanganda), fikrlash darajasi (bir qatorda).
    1. Reasoning: joriy ko‘rinish darajasi + /reasoning almashtirish bo‘yicha ishora.
  1. Tizim promptidagi xavfsizlik qo‘riqlovchilari maslahat xarakteriga ega. 27. Ular model xatti-harakatini yo‘naltiradi, ammo siyosatni majburan ijro etmaydi. 28. Qattiq ijro uchun asbob siyosati, exec tasdiqlashlari, sandboxing va kanal ruxsat ro‘yxatlaridan foydalaning; operatorlar bularni dizayn bo‘yicha o‘chirib qo‘yishi mumkin.

​
29. Prompt rejimlari

  1. OpenClaw sub-agentlar uchun kichikroq tizim promptlarini yaratishi mumkin. 31. Runtime har bir ishga tushirish uchun promptMode ni o‘rnatadi (foydalanuvchi ko‘radigan sozlama emas):
  • promptMode=minimal bo‘lganda, qo‘shimcha kiritilgan promptlar Group Chat Context o‘rniga Subagent Context deb belgilanadi.
    1. minimal: sub-agentlar uchun ishlatiladi; Skills, Memory Recall, OpenClaw Self-Update, Model Aliases, User Identity, Reply Tags, Messaging, Silent Replies va Heartbeats bo‘limlarini chiqarib tashlaydi. 34. Tooling, Safety, Workspace, Sandbox, Current Date & Time (ma’lum bo‘lganda), Runtime va kiritilgan kontekst mavjud bo‘lib qoladi.
    1. none: faqat asosiy identifikatsiya qatorini qaytaradi.
  1. promptMode=minimal bo‘lganda, qo‘shimcha kiritilgan promptlar Group Chat Context o‘rniga Subagent Context deb belgilanadi.

​
37. Workspace bootstrap kiritilishi

  1. Bootstrap fayllari Project Context ostida qisqartiriladi va qo‘shiladi, shunda model aniq o‘qishlarsiz identifikatsiya va profil kontekstini ko‘ra oladi:
  • AGENTS.md
  • SOUL.md
  • TOOLS.md
  • IDENTITY.md
  • USER.md
  • HEARTBEAT.md
    1. BOOTSTRAP.md (faqat mutlaqo yangi workspace-larda)
  • MEMORY.md va/yoki memory.md (workspace’da mavjud bo‘lsa; bittasi yoki ikkalasi ham kiritilishi mumkin)
Ushbu fayllarning barchasi har bir turn’da context window’ga kiritiladi, bu esa tokenlar sarflashini anglatadi. Ularni ixcham saqlang — ayniqsa vaqt o‘tishi bilan kattalashishi mumkin bo‘lgan MEMORY.md, bu kutilmaganda yuqori context ishlatilishiga va tez-tez compaction’ga olib kelishi mumkin.
Eslatma: memory/*.md kundalik fayllar avtomatik ravishda kiritilmaydi. Ular memory_search va memory_get vositalari orqali talab bo‘yicha olinadi, shuning uchun model ularni aniq o‘qimaguncha context window hisobiga kirmaydi.
  1. Katta fayllar belgi (marker) bilan qisqartiriladi. 47. Har bir fayl uchun maksimal hajm agents.defaults.bootstrapMaxChars (standart: 20000) tomonidan boshqariladi. Fayllar bo‘yicha jami kiritiladigan bootstrap kontent agents.defaults.bootstrapTotalMaxChars (standart: 24000) bilan cheklangan. Mavjud bo‘lmagan fayllar uchun qisqa missing-file belgisi kiritiladi.
Sub-agent session’lari faqat AGENTS.md va TOOLS.md fayllarini kiritadi (boshqa bootstrap fayllar sub-agent context’ini kichik saqlash uchun filtrlanadi).
  1. Ichki hooklar agent:bootstrap orqali bu bosqichni ushlab qolib, kiritilgan bootstrap fayllarini o‘zgartirishi yoki almashtirishi mumkin (masalan, SOUL.md ni muqobil persona bilan almashtirish).
  2. Har bir kiritilgan fayl qancha hissa qo‘shishini (xom vs kiritilgan, qisqartirish va asbob sxemasi ustama xarajatlari bilan) tekshirish uchun /context list yoki /context detail dan foydalaning. See Context.

​
Time handling

The system prompt includes a dedicated Current Date & Time section when the user timezone is known. To keep the prompt cache-stable, it now only includes the time zone (no dynamic clock or time format). Use session_status when the agent needs the current time; the status card includes a timestamp line. Configure with:
  • agents.defaults.userTimezone
  • agents.defaults.timeFormat (auto | 12 | 24)
See Date & Time for full behavior details.

​
Skills

When eligible skills exist, OpenClaw injects a compact available skills list (formatSkillsForPrompt) that includes the file path for each skill. The prompt instructs the model to use read to load the SKILL.md at the listed location (workspace, managed, or bundled). If no skills are eligible, the Skills section is omitted. 
<available_skills>
  <skill>
    <name>...</name>
    <description>...</description>
    <location>...</location>
  </skill>
</available_skills>
This keeps the base prompt small while still enabling targeted skill usage.

​
Documentation

When available, the system prompt includes a Documentation section that points to the local OpenClaw docs directory (either docs/ in the repo workspace or the bundled npm package docs) and also notes the public mirror, source repo, community Discord, and ClawHub (https://clawhub.com) for skills discovery. The prompt instructs the model to consult local docs first for OpenClaw behavior, commands, configuration, or architecture, and to run openclaw status itself when possible (asking the user only when it lacks access).