sadrazam

Proje hafızası — kur, yaz, oku veya karar kaydet (defterdar nizamı)

Bir işi Divan Protokolü ile baştan sona teslim et (plan → icra → teftiş → takdim)

İşi en küçük güvenli ajan düzeniyle sevk et (tek oturum, subagent veya izole takım)

Divan reposunu standartlara göre denetle ve raporla

Divan'a standartlara uygun yeni bir skill (vezir) kazandır

Research scout of the Divan. Use PROACTIVELY before building anything non-trivial - explores the codebase, reads existing patterns and external docs, and returns a concise briefing so the main agent builds on facts, not guesses.

Independent inspector of the Divan. Use after implementation and before declaring done - reviews the diff against the plan, runs the tests, verifies evidence exists in .divan/evidence/, and reports pass or fail with reasons. Never fixes code itself.

Persistent project memory keeper for the USER's project (Ottoman treasurer-scribe). Creates and maintains AGENTS.md, BLUEPRINT.md (vision, ADR decision records, roadmap, status log), a .divan/ progress journal, decision records and a test-evidence log; reads them at every session start and writes after each Divan Protokolu phase and at session end, so work survives restarts, context compaction and machine moves. Use when starting any serious project, when resuming work, or when the user says "hafiza", "kaldigim yerden devam", "defter kur", "BLUEPRINT", "AGENTS.md", "remember this project", "resume where we left off".

Technology stack advisor (counselor vizier). Recommends the current world-standard stack by project type - landing page, SaaS, e-commerce, mobile, AI app, fintech/borsa - covering framework, database, ORM, auth, payments and hosting, each with one-line rationale and when-to-deviate rules. Carries a freshness protocol - verifies current versions with a web search before final advice and distrusts its own defaults after six months. Use when the user asks hangi stack, hangi veritabani, teknoloji sec, neyle yazayim, hangi framework, which stack, which database, what tech should I use, or at project start before building.

Native-first agent orchestration for Claude Code and Codex. Use when a task may benefit from subagents, parallel research, isolated worktrees, dynamic agent workflows, or experimental Agent Teams, and when the user asks for orchestration, multiple agents, a team, swarm-like execution, or faster parallel delivery. Chooses the smallest safe coordination lane, assigns clear ownership, limits cost, and requires independent verification without installing an external harness.

End-to-end delivery orchestrator (the "Grand Vizier"). Use whenever the user asks to build, create, produce, or ship anything substantial — a feature, an app, a document, a campaign — especially with phrases like "baştan sona", "hepsini yap", "büyük düşün", "ajans gibi çalış", "end-to-end", "do it all", or when the request implies a finished deliverable rather than a fragment. Coordinates brainstorming, planning, implementation, verification and final presentation so nothing ships half-done.

Skill-creation and evaluation coach for the Divan marketplace ("training a new vizier"). Use when the user wants to add, improve, benchmark or test a skill, write a SKILL.md, contribute to the marketplace, compare skill-enabled behavior with a baseline, or says yeni skill yaz, Divan'a vezir ekle, skill nasıl yazılır, skill'i test et, benchmark this skill, evaluate triggering, or create a new skill for Divan. Creates a standards-compliant skill, a bounded eval contract, marketplace wiring and teftiş evidence without inventing performance gains.

1 hook across 1 event