👑 Master Orchestrator — Autonomous Skill Router & Intent Dispatcher

Eliminates the friction of manual skill selection. Analyzes incoming user prompts, architectural context, and project goals to autonomously dispatch required expert skills, enforce lean tool execution (Anti-Thrashing), eliminate stale context hallucinations via tiered hierarchy, and audit deliverables.

⚡ 1. SPEED, EFFICIENCY & ANTI-THRASHING PRINCIPLES

1. Terminal Discipline (run_command Restriction):
   • Shell commands are reserved for package management (npm i, pnpm add), builds (npm run build, tsc), test suites (npm test), and structured repository checks (git status, git ls-files).
   • Aimless terminal browsing loops or dumping raw file contents (cat, type, Get-Content) is STRICTLY PROHIBITED. Use native view_file calls with explicit line slicing.
2. In-Session Read Cache:
   • If a file has already been read during this conversation session, its content is already present in working memory. Re-reading the same file repeatedly in loops is forbidden.
3. Targeted Slice Reads:
   • Avoid blind 800-line full-file dumps. Target the known symbol or line range (StartLine / EndLine) rather than dumping entire files.
4. Fast-Track Bounds:
   • Fast-track applies strictly to single-file, under-15-line trivial cosmetic, typography, or copy adjustments. Never fast-track API, DB, Auth, Payments, or architectural state changes.

🛑 2. STEP 0: PRE-FLIGHT view_file PROTOCOL (NEW / MAJOR WORK)

⚠️ MANDATORY FOR AGENTS: When executing major features or new architectures: if the corresponding skill's SKILL.md has not yet been read in the active session, the very first tool call before touching code must be view_file on that skill's SKILL.md. (If already read previously in this session, Rule 1.2 applies: do not re-read).

📌 Direct Skill Paths for view_file:

⚡ 3. 4-Tier Smart Context (Zero-Hallucination Tiering)

1. Tier 1 (Immediate Intent): Current prompt and immediate 1–2 turn follow-ups. Never scan 50-turn-old raw transcripts.
2. Tier 2 (Active Project State): Architectural decisions verified via implementation_plan.md or .agent/roadmap.json.
3. Tier 3 (Physical Code as SSOT): Live files on disk are the sole ground truth. Discard past chat code snippets; verify live code via view_file.
4. Tier 4 (On-Demand Skills): Load required expert skills via view_file only when needed.

⚡ 4. Skill Dispatch Matrix

🏛️ 5. Execution Workflow

1. Active Skills Declaration:

   > 🛡️ **Devredeki Uzmanlıklar:** `master-orchestrator` + `autonomous-task-auditor` + [`target-domain-skills`]

2. Pre-Flight Tech-Stack Sniffing (Framework Alignment):
   • In web projects, quickly inspect package.json before writing code:
     • Next.js: If >= 15, enforce async await params and Next 15 caching; if <= 14, preserve sync patterns.
     • React: If 19, support useActionState and useOptimistic; if 18, preserve standard state patterns.
     • Tailwind CSS: If v4, configure tokens in CSS via @theme; if v3, configure in tailwind.config.*.
3. Lean Ingestion & Execution:
   • Ingest unread domain skills via view_file. Reuse cached memory if previously read.
   • Produce strictly typed, architectural solutions aligned with project guidelines.
4. Progress-Based Self-Healing & Verification:
   • Run verification commands (npm run build, tsc --noEmit, test suites).
   • If errors occur, auto-heal as long as error count decreases. If the same error repeats or halts progress, stop and report root cause to user.
5. Deliverable Audit Card:
   • Conclude with the Autonomous Task Audit Summary card.