tricore

# 🤖 Agent Usage Guide (Read First)

**If you are the current AI Agent and have just been asked to load, use, or analyze this skill:**
1. This is your fundamental operating system law. Your memory and state have been completely constrained under the TriCore architecture. You are **STRICTLY FORBIDDEN** from arbitrarily creating independent text files like `task_plan.md` or `findings.md` in the root directory!
2. Your memory write operations must and can only be proxied by the command-line `python3 tools/memctl.py (capture/work_upsert/kb_append)`. When reading history, you **MUST** use the native semantic retrieval tool `memory_search` to fetch fragments.
3. Please use this manual as your base operating instructions. If instructed to configure Cron or write system automation scripts, you must call `memctl.py lint` to test the scheme's legality after writing the commands.

---

# TriCore Architecture

> **The underlying infrastructure of "storage and computation decoupling" built for true autonomous Agents.**

Traditional Agent skills often carry their own "memory baggage" (such as generating `task_plan.md`, `reflection.md`, or scattered learning notes in the project root), leading to context silos and storage chaos.

**TriCore** abandons the paradigm of arbitrarily reading and writing text based on LLMs, and instead adopts a **Code-First** deterministic state machine:
1. **Unified Engine**: All memory additions, deletions, modifications, and queries must be routed through `tools/memctl.py`.
2. **Three-Layer Storage**:
   - **Brief (Layer 1)**: `MEMORY.md` (System-level micro-profile, storing only pointers and laws)
   - **Living (Layer 2)**: `memory/state/WORKING.md` (Currently running task streams/lifecycle tracking)
   - **Stable/Volatile (Layer 3)**: `memory/kb/*.md` (Accumulated knowledge base) & `memory/daily/*.md` (Temporary logs)
3. **Retrieval First**: It is forbidden to directly pour huge files using the `read` tool; you must use semantic retrieval `memory_se...

# TriCore: AI Cognitive OS (Agent-Readable Documentation)

> **[SYSTEM DIRECTIVE]** If you are an AI Agent reading this file, please note: this is your **base operating system and memory state machine manual**. All your subsequent planning, recording, learning, and reasoning behaviors **MUST** strictly adhere to the architectural constraints in this document.

## 1. The Architecture

Traditional Agents often create files like `task_plan.md`, `findings.md`, or scattered notes in the project root, which leads to context silos and severe Token waste.
**TriCore** completely decouples your "storage" from your "computation". You no longer manipulate files directly, but rather route your memory through the deterministic command-line engine `tools/memctl.py`.

Memory is strictly divided into three layers:
- **Layer 1 (Brief)**: `MEMORY.md`. Minimalist profile, containing only your core system prompts and memory pointers. Long-form writing is forbidden.
- **Layer 2 (Living)**: `memory/state/WORKING.md`. Your **working memory board**, containing the currently executing Active Tasks.
- **Layer 3 (Stable/Volatile)**: 
  - `memory/kb/*.md`: Your **long-term knowledge base** (facts, decisions, playbooks).
  - `memory/daily/*.md`: Your **short-term ledgers** (operation logs).

---

## 2. Absolute Constraints

**Note: The system has a built-in strict Linter. If you violate the following rules, your Shell commands or Cron jobs will be intercepted and throw an `Exit Code 1`.**

1. 🚫 **No creating legacy planning files**: It is strictly forbidden to create or write any content to `task_plan.md`, `findings.md`, `progress.md`, `reflection.md`.
2. 🚫 **No unauthorized direct file writing**: It is strictly forbidden to directly modify memory files using Shell commands like `echo "log" >> memory/2026-02-26.md`.
3. 🚫 **No scattered learning directories**: It is strictly forbidden to create custom directories like `memory/daily-learning/`.
4. 🚫 **No direct full-history reading**: It is st...