SireJeff · Feb 23, 2026 · Feb 23, 2026 · Feb 23, 2026
diff --git a/skills/implement/SKILL.md b/skills/implement/SKILL.md
@@ -33,22 +33,22 @@ ONE TODO → ONE CHANGE → ONE TEST → ONE COMMIT
 FOR each CHUNK-Pn in dependency_order:
   IF dependencies_complete:
     1. Load CHUNK-Pn todolist
-    
+
     FOR each TODO in todolist:
       a. Make atomic change
       b. Run todo-specific test
       c. If PASS: commit, mark TODO ✅
       d. If FAIL: STOP, investigate, fix
     END TODO LOOP
-    
+
     2. Update chunk documentation
     3. Mark CHUNK-Pn as IMPLEMENTED
     4. Update research CHUNK-Rn to IMPLEMENTED
-    
+
     IF chunks_processed % 3 == 0 OR context > 35%:
       Context reset (save progress, reload plan)
     END IF
-    
+
     5. Proceed to next ready chunk
   END IF
 END CHUNK LOOP
@@ -68,20 +68,20 @@ END CHUNK LOOP
    - Final chunks (e.g., test additions)
 
 3. **For Each Chunk (in dependency order):**
-   
+
    a. **Check dependencies complete**
-   
+
    b. **Execute each todo atomically:**
       - Make single change
       - Run specified test
       - If pass: commit with message
       - If fail: STOP, investigate, fix
-   
+
    c. **After all todos complete:**
       - Mark CHUNK-Pn as IMPLEMENTED
       - Update CHUNK-Rn in research to IMPLEMENTED
       - Commit chunk documentation updates
-   
+
    d. **Context management:**
       - After every 3 chunks: reload plan
       - If >35% utilization: save, compact, continue

diff --git a/skills/plan/SKILL.md b/skills/plan/SKILL.md
@@ -43,26 +43,26 @@ END LOOP
    - Extract per-chunk files and line numbers
 
 2. **For Each Research Chunk (CHUNK-Rn):**
-   
+
    a. **Analyze chunk content:**
       - Files explored with line numbers
       - Code flow analysis
       - Dependencies identified
-   
+
    b. **Create CHUNK-Pn todolist:**
       ```markdown
       | # | Action | File | Lines | Risk | Test | Status |
       |---|--------|------|-------|------|------|--------|
       | 1 | [Action] | file.ext | XXX | LOW | test_x | ⏳ |
       ```
-   
+
    c. **Define per-todo details:**
       - Current code snippet
       - Proposed change
       - Test to run after
-   
+
    d. **Mark research chunk as PLANNED**
-   
+
    e. **Document chunk dependencies**
 
 3. **Create Chunk Dependency Graph**

diff --git a/skills/research/SKILL.md b/skills/research/SKILL.md
@@ -1,5 +1,5 @@
 ---
-description: RPI Research Phase - Systematic codebase exploration with parallel agents and chunked output
+description: RPI Research Phase: Manifest-Driven Parallel Execution with 5 Search Agents
 ---
 
 # Context Engineering: Research Phase (Enhanced)
@@ -16,20 +16,22 @@ This research phase **KNOWS** how RPI-Plan will consume its output:
 
 ## Process
 
+1. **Parallel Search:** 5 agents simultaneously search their domains.\n2. **Manifest Creation:** Results are aggregated into a table.\n3. **Sequential Deep Dive:** Sub-agents iterate through the manifest to populate details.
+
 1. **Initialize Research Document**
    - Create `.claude/research/active/[feature]_research.md`
    - Use template from `.claude/research/RESEARCH_TEMPLATE.md`
 
 2. **Spawn Parallel Agents (3-5 agents)**
-   
+
    ```
    Agent 1: API/Route Entry Points       → CHUNK-R1
    Agent 2: Business Logic & Models      → CHUNK-R2
    Agent 3: Database/Storage Layer       → CHUNK-R3
    Agent 4: External Integrations        → CHUNK-R4
    Agent 5: Test Coverage Analysis       → CHUNK-R5
    ```
-   
+
    Each agent receives:
    - Feature name and objective
    - Assigned domain
@@ -43,13 +45,13 @@ This research phase **KNOWS** how RPI-Plan will consume its output:
    **Status:** COMPLETE
    **Parallel Agent:** Agent N
    **Ready for Planning:** Yes
-   
+
    ### Files Explored
    | File | Lines | Key Findings |
-   
+
    ### Code Flow Analysis
    [call chain with file:line refs]
-   
+
    ### Dependencies (This Chunk)
    - External: [APIs]
    - Internal: [services]

diff --git a/src/cli/repl/index.ts b/src/cli/repl/index.ts
@@ -4,8 +4,6 @@
  * Interactive shell for managing k0ntext context
  */
 
-import fs from 'fs';
-import path from 'path';
 import readline from 'readline';
 import fs from 'fs';
 import path from 'path';

diff --git a/templates/base/RPI_WORKFLOW_PLAN.md b/templates/base/RPI_WORKFLOW_PLAN.md
@@ -37,7 +37,7 @@ Understand the system, locate relevant components, prevent context pollution
 
 ### Process
 1. Load WORKFLOW_INDEX.md first (saves 100k+ tokens)
-2. **Spawn 3-5 parallel Explore agents** (one per research domain)
+2. **Spawn 5 parallel Explore agents (API, Logic, DB, External, Tests)** (one per research domain)
 3. Each agent produces a **Research Chunk** with:
    - Chunk ID (e.g., `CHUNK-R1`, `CHUNK-R2`)
    - Explored files with line numbers

diff --git a/templates/base/agents/rpi-engineer.md b/templates/base/agents/rpi-engineer.md
@@ -0,0 +1,70 @@
+---
+name: rpi-engineer
+version: "1.0.0"
+description: "RPI Orchestrator: Quality, Feature, and Debug Engineer using the RPI Workflow"
+category: "core-agent"
+---
+
+# Agent Profile: RPI Engineer
+
+**Role:** RPI Orchestrator / Quality Engineer
+
+**Mission:** You are the expert orchestrator of the RPI (Research, Plan, Implement) workflow. Your goal is to solve bugs, implement features, and write tests by strictly adhering to the 3-phase RPI process.
+
+**Directives:**
+-   **Debugs:** Do NOT fix bugs directly. Start with `/rpi-research` to understand the root cause.
+-   **Features:** Do NOT code immediately. Start with `/rpi-research` to map the requirements.
+-   **Tests:** Do NOT guess tests. Start with `/rpi-research` to identify coverage gaps.
+
+---
+
+## Capabilities & Skills
+
+You have full mastery of the following skills and **MUST** use them in this order:
+
+1.  **Research (`/rpi-research`):**
+    -   You know this command spawns **5 parallel agents** (API, Logic, DB, External, Tests).
+    -   You expect a **Research Manifest** as output.
+    -   You verify that all 5 domains have been explored before proceeding.
+
+2.  **Plan (`/rpi-plan`):**
+    -   You execute this only AFTER research is complete.
+    -   You expect a **Plan Manifest** with specific chunks linked to research chunks.
+    -   You verify that atomic todolists (with file:line precision) are generated.
+
+3.  **Implement (`/rpi-implement`):**
+    -   You execute this only AFTER the plan is approved.
+    -   You expect this to run manifest chunks sequentially.
+    -   You verify that statuses in both Plan and Research manifests are updated to `IMPLEMENTED`.
+
+---
+
+## Workflow Triggers
+
+### Trigger: "Fix this bug"
+1.  **Analyze:** "I need to understand the bug first."
+2.  **Action:** Run `/rpi-research [bug-description]`.
+3.  **Outcome:** Research Manifest identifying the buggy component.
+4.  **Next:** Run `/rpi-plan` -> `/rpi-implement`.
+
+### Trigger: "Add this feature"
+1.  **Analyze:** "I need to map the feature requirements."
+2.  **Action:** Run `/rpi-research [feature-name]`.
+3.  **Outcome:** Research Manifest covering all 5 domains.
+4.  **Next:** Run `/rpi-plan` -> `/rpi-implement`.
+
+### Trigger: "Add tests" (Dests)
+1.  **Analyze:** "I need to find where tests are missing."
+2.  **Action:** Run `/rpi-research [test-scope]` (Agent 5 will be key here).
+3.  **Outcome:** Research Manifest highlighting coverage gaps.
+4.  **Next:** Run `/rpi-plan` -> `/rpi-implement`.
+
+---
+
+## Inter-Phase Context Awareness
+You are the guardian of the context. You ensure:
+-   **Research** outputs a manifest readable by **Plan**.
+-   **Plan** outputs a manifest readable by **Implement**.
+-   **Implement** updates the status of the entire chain.
+
+If any phase fails to produce the correct manifest format, you **STOP** and request a correction before proceeding.
diff --git a/templates/base/commands/rpi-implement.md b/templates/base/commands/rpi-implement.md
@@ -78,7 +78,7 @@ ONE TODO → ONE CHANGE → ONE TEST → ONE COMMIT
 │ RPI-IMPLEMENT CHUNK PROCESSING LOOP                     │
 ├─────────────────────────────────────────────────────────┤
 │ FOR each CHUNK-Pn in dependency_order:                  │
-│   1. Load CHUNK-Pn todolist                             │
+│   1. Load CHUNK-Pn todolist (from Plan Manifest)                             │
 │   2. FOR each TODO in CHUNK-Pn:                         │
 │      a. Make atomic change                              │
 │      b. Run specified test                              │

diff --git a/templates/base/commands/rpi-plan.md b/templates/base/commands/rpi-plan.md
@@ -65,7 +65,7 @@ RPI-Plan **KNOWS**:
 │ RPI-PLAN CHUNK PROCESSING LOOP                          │
 ├─────────────────────────────────────────────────────────┤
 │ FOR each research_chunk (CHUNK-R1 to CHUNK-RN):         │
-│   1. Read research_chunk content                        │
+│   1. Read research_chunk content (Sequential Sub-Agent)                        │
 │   2. Create corresponding CHUNK-Pn todolist:            │
 │      - Define atomic action items                       │
 │      - Specify file:line for each action                │

diff --git a/templates/base/commands/rpi-research.md b/templates/base/commands/rpi-research.md
@@ -6,7 +6,7 @@ category: "rpi-orchestration"
 rpi_phase: "research"
 context_budget_estimate: "50K tokens"
 typical_context_usage: "25%"
-parallel_agents: "3-5"
+parallel_agents: "5"
 chunk_output: true
 inter_phase_aware: true
 prerequisites: []
@@ -82,7 +82,7 @@ Spawn 3-5 parallel Explore agents, each focused on a specific domain:
 ### Step 1: Initialize Research Document
 Create `.ai-context/research/active/[feature]_research.md` from RESEARCH_TEMPLATE.md
 
-### Step 2: Spawn Parallel Agents (3-5 agents)
+### Step 2: Spawn 5 Parallel Search Agents
 
 Each agent receives:
 - Feature name and objective
@@ -92,6 +92,13 @@ Each agent receives:
 
 ### Step 3: Aggregate Chunk Results
 
+### Step 4: Sequential Deep Dive (Sub-Agents Loop)
+**For each CHUNK-Rn in Manifest:**
+- Start a sub-agent to "explore and append" deep details.
+- Trace call chains (File:Line).
+- Identify dependencies.
+- Mark status as `COMPLETE` in manifest.
+
 Collect outputs from all agents and structure into:
 - Chunk Manifest (table of all chunks with status)
 - Individual chunk sections with full details