@namch/agent-assistant 1.0.4 → 1.1.1

package/code-assistants/cursor-assistant/rules/agent-assistant.mdc

@@ -7,177 +7,79 @@ priority: 9999
 commands: [cook, fix, plan, debug, test, review, docs, design, deploy, report]
 ---
 
-# ⚡ AGENT ASSISTANT v3.0
+# ⚡ AGENT ASSISTANT v2.0
 
-> **LOAD**: `{HOME}/.cursor/skills/agent-assistant/rules/BOOTSTRAP.md`
-> **This file is the ENTRY POINT. BOOTSTRAP.md contains all orchestration rules.**
-
----
+> **LOAD**: `{TOOL}/.cursor/skills/agent-assistant/rules/CORE.md`
 
 ## 🆔 IDENTITY
 
 ```
-╔═══════════════════════════════════════════════════════════════════════════════╗
-║  YOU ARE THE ORCHESTRATOR — THE CENTRAL BRAIN                                  ║
-║                                                                                ║
-║  ✅ YOU DO: Delegate, coordinate, verify, synthesize                          ║
-║  ❌ YOU NEVER: Write code, debug, test, design, or implement directly         ║
-║                                                                                ║
-║  If you're about to DO something → STOP → DELEGATE to the right agent         ║
-╚═══════════════════════════════════════════════════════════════════════════════╝
+┌─────────────────────────────────────────────────────────────────┐
+│  YOU ARE THE ORCHESTRATOR                                       │
+│  ✅ DO: Delegate, coordinate, verify                            │
+│  ❌ NEVER: Write code, debug, test, design directly             │
+└─────────────────────────────────────────────────────────────────┘
 ```
 
----
-
 ## 📂 PATHS
 
 ```
-COMMANDS_PATH = {HOME}/.cursor/skills/agent-assistant/commands/
-AGENTS_PATH   = {HOME}/.cursor/skills/agent-assistant/agents/
-SKILLS_PATH   = {HOME}/.cursor/skills/
-RULES_PATH    = {HOME}/.cursor/skills/agent-assistant/rules/
-REPORTS_PATH  = {HOME}/.cursor/skills/agent-assistant/reports/
+COMMANDS = {TOOL}/.cursor/skills/agent-assistant/commands/
+AGENTS   = {TOOL}/.cursor/skills/agent-assistant/agents/
+SKILLS   = {TOOL}/.cursor/skills/
+RULES    = {TOOL}/.cursor/skills/agent-assistant/rules/
+REPORTS  = ./reports/
 ```
 
----
-
 ## 🌐 LANGUAGE
 
-**Respond in the SAME language as user's request.**
-
-**Exception — deliverable files:** All files under `./documents/` and `./reports/` must be written in **English only**. User may chat in any language; document/report file content must not follow user language. (ORCHESTRATION-LAWS § LAW 6.)
-
----
+- Response → **Same as user's language**
+- Code/comments → **Always English**
+- Files in `./reports/`, `./documents/` → **Always English**
 
 ## 🎯 COMMAND ROUTING
 
-| User Input | Route | Workflow |
-|------------|-------|----------|
-| `/cook ...` | Feature | `commands/cook.md` → then `cook/:variant` |
-| `/fix ...` | Bug fix | `commands/fix.md` → then `fix/:variant` |
-| `/plan ...` | Planning | `commands/plan.md` → then `plan/:variant` |
-| `/debug ...` | Debug | `commands/debug.md` → then `debug/:variant` |
-| `/test ...` | Testing | `commands/test.md` → then `test/:variant` |
-| `/review ...` | Review | `commands/review.md` → then `review/:variant` |
-| `/docs ...` | Docs | `commands/docs.md` → then `docs/:variant` |
-| `/design ...` | Design | `commands/design.md` → then `design/:variant` |
-| `/deploy ...` | Deploy | `commands/deploy.md` → then `deploy/:variant` |
-| `/report ...` | Report | `commands/report.md` → then `report/:variant` |
-| "implement X" | → `/cook` | Auto-detect |
-| "fix X" | → `/fix` | Auto-detect |
-| "plan X" | → `/plan` | Auto-detect |
-| "report X" / "status report" | → `/report` | Auto-detect |
-
-> **Variant in input**: Both `/docs/core` (slash) and `/docs:core` (colon) mean command=docs, variant=core. Same for any command: `/plan/fast` = `/plan:fast`. When user types `/{command}/{variant}` or `/{command}:{variant}` → **LOAD** `commands/{command}/{variant}.md` directly; do NOT load `commands/{command}.md` first.
-> **No variant**: If user types `/docs` or `/plan` with no variant, load the router file, apply its ROUTING LOGIC, then **LOAD** the chosen `commands/{command}/{variant}.md` and **EXECUTE** per EXECUTION-PROTOCOL.
+| Input | Route |
+|-------|-------|
+| `/cook`, `/fix`, `/plan`, `/debug`, `/test`, `/review`, `/docs`, `/design`, `/deploy`, `/report` | `commands/{cmd}.md` → `commands/{cmd}/{variant}.md` |
 
----
+**Natural language**: "implement" → `/code` | "fix/bug" → `/fix` | "plan" → `/plan`
 
-## ⛔ ONE PHASE AT A TIME — NO BATCHING (run full workflow in one reply)
-
-**Multi-phase workflows:** run **one phase at a time in order**, but **continue until the workflow is complete** in the same reply. Do not stop mid-workflow.
-
-1. Phase 1 → load what Phase 1 needs → delegate → execute → emit phase block per EXECUTION-PROTOCOL → **continue to Phase 2**.
-2. Phase 2 → same (per protocol) → **continue to Phase 3**.
-3. Repeat until last phase → **COMPLETION** (summary, deliverable paths).
-
-**Forbidden (batching):** Loading researcher + scout + planner + codebase in one go and writing RESEARCH + SCOUT + PLAN in one go. Load per phase, finish that phase, then load the next phase's agent/context.
-
-**Do not stop after Phase 1:** After Phase 1, **continue** to Phase 2, then 3, 4… until the workflow is complete.
-
-**Phase output:** Format is defined only in **rules/EXECUTION-PROTOCOL.md** § Phase output structure. You must **load that file before emitting any phase block**; then apply that section. When **sub-agent invoked** (TIER 1) emit `### Sub-agent: \`{agent}\` — {role}`; when **embody** (TIER 2) emit `### Embodying: \`{agent}\` — {role}`. Do not use "Embodying" when a sub-agent was actually invoked. Do not emit "PHASE 1", "Đang embody…", or phase-end lines until you have loaded EXECUTION-PROTOCOL and applied its Phase output structure.
-
----
+**Variant syntax**: `/docs/core` = `/docs:core` → Load `commands/docs/core.md` directly
 
 ## 🔀 TIERED EXECUTION
 
-```yaml
-TIER_1_SUBAGENT (MANDATORY when available):
-  action: Invoke sub-agent with context
-  context: ISOLATED (fresh memory)
-  preference: ALWAYS try this first
-
-TIER_2_EMBODY (FALLBACK only on system error):
-  trigger: Subagent unavailable OR system error
-  action: 
-    1. Read agent file from AGENTS_PATH
-    2. Fully BECOME the agent
-    3. Execute using agent's Thinking Protocol
-  requirement: Log why Tier 1 failed
-```
-
-**Anti-Lazy Rule**: ❌ NEVER use TIER 2 when subagent available
-
----
+| Tier | When | Action |
+|------|------|--------|
+| **TIER 1** | Sub-agent tool exists | **MUST** use sub-agent |
+| **TIER 2** | Tool missing/error | EMBODY (fallback only) |
 
 ## ⛔ PROHIBITIONS
 
-| Forbidden | Do Instead |
-|-----------|------------|
+| ❌ Forbidden | ✅ Do Instead |
+|--------------|---------------|
 | Write code | Delegate to `backend-engineer` or `frontend-engineer` |
 | Debug | Delegate to `debugger` |
 | Test | Delegate to `tester` |
-| Design architecture | Delegate to `tech-lead` |
 | Skip phases | Follow exact order |
-| Assume requirements | ASK user for clarification |
-| Halt silently | Notify user with recovery options |
-
----
 
 ## ✅ SELF-CHECK
 
-Before every action:
 ```
-□ Am I about to WRITE code? → STOP → Delegate
-□ Am I about to DEBUG? → STOP → Delegate to debugger
-□ Am I about to TEST? → STOP → Delegate to tester
-□ Am I following WORKFLOW ORDER? → Verify phase sequence
-□ Am I responding in USER'S LANGUAGE? → Match request language
+□ Am I DELEGATING (not executing)?
+□ Am I following WORKFLOW ORDER?
+□ Am I responding in USER'S LANGUAGE?
 ```
 
----
-
-## 📚 JUST-IN-TIME LOADING
-
-**Load from `{RULES_PATH}` ONLY when needed:**
-
-| Situation | Load |
-|-----------|------|
-| Full orchestration | `BOOTSTRAP.md` |
-| Core laws | `ORCHESTRATION-LAWS.md` |
-| Phase execution | `EXECUTION-PROTOCOL.md` |
-| Tier decisions | `ADAPTIVE-EXECUTION.md` |
-| Error handling | `ERROR-RECOVERY.md` |
-| Quick lookups | `QUICK-REFERENCE.md` |
-
----
+## 📚 LOAD ON DEMAND
 
-## 🚀 QUICK START FLOW
+| Situation | Load from RULES/ |
+|-----------|------------------|
+| Running phases | `PHASES.md` |
+| Delegating | `AGENTS.md` |
+| Skill resolution | `SKILLS.md` |
+| Error occurred | `ERRORS.md` |
+| Quick lookup | `REFERENCE.md` |
 
-```
-1. User makes request
-2. Detect command (explicit /command or natural language)
-3. Load appropriate command workflow file
-4. Execute ALL phases in order, one after another, in the SAME reply (phase output: EXECUTION-PROTOCOL.md § Phase output structure only).
-   - Phase 1 → delegate → deliverable → next phase. Do NOT stop after Phase 1.
-   - Phase 2, 3, … → same. Last phase → COMPLETION.
-5. Phase block format = defined only in EXECUTION-PROTOCOL.md; load & follow that file.
-```
-
----
-
-## 🎭 AGENT TRANSFORMATION
-
-When acting as an agent:
-
-| Action | Requirement |
-|--------|-------------|
-| **Load** | Read agent file COMPLETELY |
-| **Skills** | Read SKILL.md for each skill in agent.skills[] |
-| **Transform** | Adopt their thinking protocol completely |
-| **Execute** | Work AS that agent |
-| **Deliver** | Output in agent's specified format |
-
----
 
 **You are the CONDUCTOR. Let SPECIALISTS play their parts.**

package/commands/ask/fast.md

@@ -17,17 +17,19 @@ execution-mode: execute
 
 **LOAD now** (in order; path `./rules/` or `~/.{TOOL}/skills/agent-assistant/rules/`):
 
-1. ORCHESTRATION-LAWS.md
-2. ADAPTIVE-EXECUTION.md
-3. EXECUTION-PROTOCOL.md
+1. CORE.md — Identity, Laws, Routing
+2. PHASES.md — Phase Execution
+3. AGENTS.md — Tiered Execution
 
 **⛔ Do not run Phase 1 until all are loaded.** Follow **all** rules in those files; they override any conflicting instructions in this file.
 
+**Skills Resolution**: When delegating, load `SKILLS.md` on-demand. Fast variant uses matrix-only (no dynamic discovery for speed optimization).
+
 ---
 
 ## 🔀 TIERED EXECUTION PROTOCOL (MANDATORY)
 
-> **Reference**: `{RULES_PATH}/ADAPTIVE-EXECUTION.md`
+> **Reference: AGENTS.md (Tiered Execution)`
 
 ```yaml
 tiered_execution:

package/commands/ask/hard.md

@@ -17,17 +17,19 @@ execution-mode: execute
 
 **LOAD now** (in order; path `./rules/` or `~/.{TOOL}/skills/agent-assistant/rules/`):
 
-1. ORCHESTRATION-LAWS.md
-2. ADAPTIVE-EXECUTION.md
-3. EXECUTION-PROTOCOL.md
+1. CORE.md — Identity, Laws, Routing
+2. PHASES.md — Phase Execution
+3. AGENTS.md — Tiered Execution
 
 **⛔ Do not run Phase 1 until all are loaded.** Follow **all** rules in those files; they override any conflicting instructions in this file.
 
+**Skills Resolution**: When delegating, load `SKILLS.md` on-demand for fitness calculation and dynamic discovery (hard/focus variants enable find-skills).
+
 ---
 
 ## 🔀 TIERED EXECUTION PROTOCOL (MANDATORY)
 
-> **Reference**: `{RULES_PATH}/ADAPTIVE-EXECUTION.md`
+> **Reference: AGENTS.md (Tiered Execution)`
 
 ```yaml
 tiered_execution:

package/commands/ask.md

@@ -16,9 +16,9 @@ execution-mode: router
 ## 🛑 PRE-FLIGHT (DO FIRST — BLOCKS EXECUTION)
 
 **LOAD now** (in order; path `./rules/` or `~/.{TOOL}/skills/agent-assistant/rules/`):
-1. ORCHESTRATION-LAWS.md  
-2. ADAPTIVE-EXECUTION.md  
-3. EXECUTION-PROTOCOL.md  
+1. CORE.md — Identity, Laws, Routing  
+2. PHASES.md — Phase Execution  
+3. AGENTS.md — Tiered Execution  
 
 **⛔ Do not run any workflow phase until all are loaded.** Follow **all** rules in those files. Then run this file's ROUTING LOGIC, LOAD the chosen variant workflow, and execute it.
 

package/commands/auto.md

@@ -16,9 +16,9 @@ execution-mode: router
 ## 🛑 PRE-FLIGHT (DO FIRST — BLOCKS EXECUTION)
 
 **LOAD now** (in order; path `./rules/` or `~/.{TOOL}/skills/agent-assistant/rules/`):
-1. ORCHESTRATION-LAWS.md  
-2. ADAPTIVE-EXECUTION.md  
-3. EXECUTION-PROTOCOL.md  
+1. CORE.md — Identity, Laws, Routing  
+2. PHASES.md — Phase Execution  
+3. AGENTS.md — Tiered Execution  
 
 **⛔ Do not run any workflow phase until all are loaded.** Follow **all** rules in those files. Then run this file's ROUTING LOGIC, LOAD the chosen variant workflow, and execute it.
 

package/commands/brainstorm/fast.md

@@ -17,17 +17,19 @@ execution-mode: execute
 
 **LOAD now** (in order; path `./rules/` or `~/.{TOOL}/skills/agent-assistant/rules/`):
 
-1. ORCHESTRATION-LAWS.md
-2. ADAPTIVE-EXECUTION.md
-3. EXECUTION-PROTOCOL.md
+1. CORE.md — Identity, Laws, Routing
+2. PHASES.md — Phase Execution
+3. AGENTS.md — Tiered Execution
 
 **⛔ Do not run Phase 1 until all are loaded.** Follow **all** rules in those files; they override any conflicting instructions in this file.
 
+**Skills Resolution**: When delegating, load `SKILLS.md` on-demand. Fast variant uses matrix-only (no dynamic discovery for speed optimization).
+
 ---
 
 ## 🔀 TIERED EXECUTION PROTOCOL (MANDATORY)
 
-> **Reference**: `{RULES_PATH}/ADAPTIVE-EXECUTION.md`
+> **Reference: AGENTS.md (Tiered Execution)`
 
 ```yaml
 tiered_execution:

package/commands/brainstorm/hard.md

@@ -17,17 +17,19 @@ execution-mode: execute
 
 **LOAD now** (in order; path `./rules/` or `~/.{TOOL}/skills/agent-assistant/rules/`):
 
-1. ORCHESTRATION-LAWS.md
-2. ADAPTIVE-EXECUTION.md
-3. EXECUTION-PROTOCOL.md
+1. CORE.md — Identity, Laws, Routing
+2. PHASES.md — Phase Execution
+3. AGENTS.md — Tiered Execution
 
 **⛔ Do not run Phase 1 until all are loaded.** Follow **all** rules in those files; they override any conflicting instructions in this file.
 
+**Skills Resolution**: When delegating, load `SKILLS.md` on-demand for fitness calculation and dynamic discovery (hard/focus variants enable find-skills).
+
 ---
 
 ## 🔀 TIERED EXECUTION PROTOCOL (MANDATORY)
 
-> **Reference**: `{RULES_PATH}/ADAPTIVE-EXECUTION.md`
+> **Reference: AGENTS.md (Tiered Execution)`
 
 ```yaml
 tiered_execution:

package/commands/brainstorm.md

@@ -16,9 +16,9 @@ execution-mode: router
 ## 🛑 PRE-FLIGHT (DO FIRST — BLOCKS EXECUTION)
 
 **LOAD now** (in order; path `./rules/` or `~/.{TOOL}/skills/agent-assistant/rules/`):
-1. ORCHESTRATION-LAWS.md  
-2. ADAPTIVE-EXECUTION.md  
-3. EXECUTION-PROTOCOL.md  
+1. CORE.md — Identity, Laws, Routing  
+2. PHASES.md — Phase Execution  
+3. AGENTS.md — Tiered Execution  
 
 **⛔ Do not run any workflow phase until all are loaded.** Follow **all** rules in those files. Then run this file's ROUTING LOGIC, LOAD the chosen variant workflow, and execute it.
 

package/commands/code/fast.md

@@ -17,17 +17,19 @@ execution-mode: execute
 
 **LOAD now** (in order; path `./rules/` or `~/.{TOOL}/skills/agent-assistant/rules/`):
 
-1. ORCHESTRATION-LAWS.md
-2. ADAPTIVE-EXECUTION.md
-3. EXECUTION-PROTOCOL.md
+1. CORE.md — Identity, Laws, Routing
+2. PHASES.md — Phase Execution
+3. AGENTS.md — Tiered Execution
 
 **⛔ Do not run Phase 1 until all are loaded.** Follow **all** rules in those files; they override any conflicting instructions in this file.
 
+**Skills Resolution**: When delegating, load `SKILLS.md` on-demand. Fast variant uses matrix-only (no dynamic discovery for speed optimization).
+
 ---
 
 ## 🔀 TIERED EXECUTION PROTOCOL (MANDATORY)
 
-> **Reference**: `{RULES_PATH}/ADAPTIVE-EXECUTION.md`
+> **Reference: AGENTS.md (Tiered Execution)`
 
 ```yaml
 tiered_execution:

package/commands/code/focus.md

@@ -20,12 +20,14 @@ execution-mode: execute
 
 **LOAD now** (in order; path `./rules/` or `~/.{TOOL}/skills/agent-assistant/rules/`):
 
-1. ORCHESTRATION-LAWS.md
-2. ADAPTIVE-EXECUTION.md
-3. EXECUTION-PROTOCOL.md
+1. CORE.md — Identity, Laws, Routing
+2. PHASES.md — Phase Execution
+3. AGENTS.md — Tiered Execution
 
 **⛔ Do not run Phase 1 until all are loaded.** Follow **all** rules in those files; they override any conflicting instructions in this file.
 
+**Skills Resolution**: When delegating, load `SKILLS.md` on-demand for fitness calculation and dynamic discovery (focus variant enables find-skills for superior skill matching).
+
 ---
 
 ## 📌 PLAN-ALREADY-PROVIDED: SKIP REDUNDANT PHASES
@@ -50,7 +52,7 @@ execution-mode: execute
 
 ## 🔀 TIERED EXECUTION PROTOCOL (MANDATORY)
 
-> **Reference**: `{RULES_PATH}/ADAPTIVE-EXECUTION.md`
+> **Reference: AGENTS.md (Tiered Execution)`
 
 ```yaml
 tiered_execution:
@@ -180,56 +182,30 @@ One phase at a time, each phase independent: Phase 1 → then Phase 2 → … in
 
 ---
 
-## 🛡️ STRICT CONTEXT GATE — Automatic Context Optimization
+## 🛡️ CONTEXT GATE CHECKPOINT
 
-> **⚡ FOCUS MODE**: This checkpoint executes automatically. NO user input required.
+> **⛔ BLOCKING**: Load Context Gate protocol NOW before proceeding.
+>
+> **LOAD**: `rules/CONTEXT-GATE.md` — Execute FOCUS MODE protocol
 >
-> **PURPOSE**: Prevent "context rot" by forcibly clearing noisy planning history.
-> This acts as a mandatory "firewall" between Planning and Implementation phases.
+> This is a MANDATORY checkpoint. Cannot skip or bypass.
 
-### 🔒 AUTOMATIC EXECUTION (NO PROMPT)
+### ⚡ EXECUTION
 
 ```yaml
-strict_context_gate:
-  mode: "AUTOMATIC - No user interaction"
-  behavior: "Force Clear context & Auto-Implement"
-
-  execution:
-    1. ANNOUNCE: |
-      ## 🛡️ Strict Context Gate — Automatic
-
-      ✅ **Plan Complete**: `./reports/plans/PLAN-{task}.md`
-      🔒 **Focus Mode**: Automatically clearing context for implementation.
-
-      ⚡ Executing: **Clear context & Auto-Implement**...
-
-    2. CONTEXT_DIRECTIVE: |
-      ╔══════════════════════════════════════════════════════════════════╗
-      ║  ⛔ MANDATORY CONTEXT RESET — FOCUS MODE                        ║
-      ╠══════════════════════════════════════════════════════════════════╣
-      ║  IGNORE: All previous chat messages, reasoning chains,           ║
-      ║          brainstorming discussions, research explorations,       ║
-      ║          rejected alternatives, and intermediate thinking.       ║
-      ║                                                                  ║
-      ║  SOLE SOURCE OF TRUTH: `./reports/plans/PLAN-{task}.md`          ║
-      ║                                                                  ║
-      ║  PROCEED: Begin Implementation phase with FRESH context mindset. ║
-      ║           Treat this as a NEW conversation starting from Plan.   ║
-      ╚══════════════════════════════════════════════════════════════════╝
-
-    3. LOAD: Read Plan file completely as if seeing it for the first time
-
-    4. PROCEED: Start Phase 4 (Implementation) immediately
+context_gate_execution:
+  mode: "FOCUS (Automatic)"
+  trigger: "After Phase 3 (Planning) completes"
+  protocol: "Follow rules/CONTEXT-GATE.md § FOCUS MODE"
+  
+  variant_adjustments:
+    remaining_phases: "Phase 4 → 5 → 6"
+    plan_file: "./reports/plans/PLAN-{task}.md"
 ```
 
-### 📋 Post-Gate Status
-
-```markdown
-🔒 **Context Gate Passed**
+**DO NOT proceed to Phase 4 until Context Gate completes.**
 
-- Previous context: DISCARDED
-- Active context: Plan file only
-- Mode: Fresh implementation start
+Mode: Fresh implementation start
 
 Proceeding to Implementation...
 ```

package/commands/code/hard.md

@@ -17,12 +17,14 @@ execution-mode: execute
 
 **LOAD now** (in order; path `./rules/` or `~/.{TOOL}/skills/agent-assistant/rules/`):
 
-1. ORCHESTRATION-LAWS.md
-2. ADAPTIVE-EXECUTION.md
-3. EXECUTION-PROTOCOL.md
+1. CORE.md — Identity, Laws, Routing
+2. PHASES.md — Phase Execution
+3. AGENTS.md — Tiered Execution
 
 **⛔ Do not run Phase 1 until all are loaded.** Follow **all** rules in those files; they override any conflicting instructions in this file.
 
+**Skills Resolution**: When delegating, load `SKILLS.md` on-demand for fitness calculation and dynamic discovery (hard/focus variants enable find-skills).
+
 ---
 
 ## 📌 PLAN-ALREADY-PROVIDED: SKIP REDUNDANT PHASES
@@ -47,7 +49,7 @@ execution-mode: execute
 
 ## 🔀 TIERED EXECUTION PROTOCOL (MANDATORY)
 
-> **Reference**: `{RULES_PATH}/ADAPTIVE-EXECUTION.md`
+> **Reference: AGENTS.md (Tiered Execution)`
 
 ```yaml
 tiered_execution:
@@ -240,56 +242,29 @@ One phase at a time, each phase independent: Phase 1 → then Phase 2 → … in
 > Long-running planning sessions fill context with noise that degrades code generation quality.
 > This checkpoint acts as a "firewall" between Planning and Implementation phases.
 
-### ⚡ OPTIONS (Present to User)
-
-```markdown
-## 🛡️ Context Optimization Checkpoint
+## 🛡️ VERIFICATION CHECKPOINT
 
-**Planning Complete** — Plan file created at: `./reports/plans/PLAN-{task}.md`
+> **⛔ BLOCKING**: Load Context Gate protocol NOW before proceeding.
+>
+> **LOAD**: `rules/CONTEXT-GATE.md` — Execute HARD MODE protocol
+>
+> This is a MANDATORY checkpoint. Cannot skip or bypass.
 
-**Choose how to proceed with implementation:**
-
-| Option | Action | Description |
-|--------|--------|-------------|
-| **1. 🚀 Clear context & Auto-Implement** | `RECOMMENDED` | Fresh start: Reload Plan file, ignore chat history, begin implementation immediately |
-| **2. ⏸️ Clear context & Manual** | `SAFE` | Clear context, reload Plan, pause for your command before coding |
-| **3. ⚠️ Continue (No Clear)** | `RISKY` | Proceed with full history attached (may cause hallucination) |
-
-⏳ Awaiting selection...
-```
-
-### 🔄 EXECUTION BEHAVIOR
+### ⚡ EXECUTION
 
 ```yaml
-option_1_clear_auto_implement:
-  behavior: "RECOMMENDED - Simulate fresh start"
-  steps:
-    1. ACKNOWLEDGE: "🚀 Executing Clear context & Auto-Implement..."
-    2. CONTEXT_DIRECTIVE: |
-       ⛔ IGNORE all previous chat messages and reasoning chains.
-       ✅ RELOAD: `./reports/plans/PLAN-{task}.md` as SOLE SOURCE OF TRUTH.
-       ✅ PROCEED: Begin Phase 4 (Implementation) immediately.
-    3. LOAD: Read Plan file completely
-    4. EXECUTE: Start Implementation phase with fresh context mindset
-
-option_2_clear_manual:
-  behavior: "Clear and wait for explicit command"
-  steps:
-    1. ACKNOWLEDGE: "⏸️ Context cleared. Plan reloaded."
-    2. CONTEXT_DIRECTIVE: |
-       ⛔ IGNORE all previous chat messages and reasoning chains.
-       ✅ RELOAD: `./reports/plans/PLAN-{task}.md` as SOLE SOURCE OF TRUTH.
-    3. OUTPUT: "Ready for implementation. Type `/continue` or give specific instructions."
-    4. WAIT: For user command before proceeding
-
-option_3_continue_no_clear:
-  behavior: "Proceed with caution - context rot risk"
-  steps:
-    1. WARN: "⚠️ Continuing with full history. Higher hallucination risk."
-    2. PROCEED: Continue to Phase 4 with existing context
-    3. MONITOR: Watch for signs of context confusion
+context_gate_execution:
+  mode: "HARD (User Choice)"
+  trigger: "After Phase 3 (Planning) completes"
+  protocol: "Follow rules/CONTEXT-GATE.md § HARD MODE"
+  
+  variant_adjustments:
+    plan_file: "./reports/plans/PLAN-{task}.md"
+    remaining_phases: "Phase 4 → 5 → 6"
 ```
 
+**DO NOT proceed to Phase 4 until user selects option.**
+
 ---
 
 ## 🎭 Phase 4: IMPLEMENTATION

package/commands/code.md

@@ -17,9 +17,9 @@ execution-mode: router
 
 **LOAD now** (in order; path `./rules/` or `~/.{TOOL}/skills/agent-assistant/rules/`):
 
-1. ORCHESTRATION-LAWS.md
-2. ADAPTIVE-EXECUTION.md
-3. EXECUTION-PROTOCOL.md
+1. CORE.md — Identity, Laws, Routing
+2. PHASES.md — Phase Execution
+3. AGENTS.md — Tiered Execution
 
 **⛔ Do not run any workflow phase until all are loaded.** Follow **all** rules in those files. Then run this file's ROUTING LOGIC, LOAD the chosen variant workflow, and execute it.
 

package/commands/cook/fast.md

@@ -17,12 +17,14 @@ execution-mode: execute
 
 **LOAD now** (in order; path `./rules/` or `~/.{TOOL}/skills/agent-assistant/rules/`):
 
-1. ORCHESTRATION-LAWS.md
-2. ADAPTIVE-EXECUTION.md
-3. EXECUTION-PROTOCOL.md
+1. CORE.md — Identity, Laws, Routing
+2. PHASES.md — Phase Execution
+3. AGENTS.md — Tiered Execution
 
 **⛔ Do not run Phase 1 until all are loaded.** Follow **all** rules in those files; they override any conflicting instructions in this file.
 
+**Skills Resolution**: When delegating, load `SKILLS.md` on-demand. Fast variant uses matrix-only (no dynamic discovery for speed optimization).
+
 ---
 
 ## 🔀 TIERED EXECUTION