opencode-orchestrator 0.9.22 → 0.9.24

package/README.md

@@ -18,6 +18,69 @@
 
 ---
 
+## 🧠 Core Philosophy
+
+> **Explore → Adapt → Act**  
+> *Never assume. Always verify. Then execute.*
+
+### The Problem with AI Agents
+
+Most AI agents fail because they:
+- **Assume** instead of checking
+- **Guess** instead of researching  
+- Follow **rigid rules** instead of adapting
+- Trust **memory** instead of verifying
+
+### Our Solution: Adaptive Intelligence
+
+Every agent in this system follows a simple principle:
+
+```
+┌─────────────────────────────────────────────────────────────┐
+│                                                             │
+│   🔍 EXPLORE    →    🔄 ADAPT    →    ⚡ ACT               │
+│                                                             │
+│   Discover         Adjust to         Execute with           │
+│   the reality      what you find     confidence             │
+│                                                             │
+└─────────────────────────────────────────────────────────────┘
+```
+
+### Agent Philosophies
+
+| Agent | Philosophy | What It Means |
+|:-----:|:-----------|:--------------|
+| 🎯 **Commander** | Explore → Adapt → Act | Discover project structure first, then delegate intelligently |
+| 📋 **Planner** | Research → Plan → Document | Never guess syntax. Find official docs, verify, then plan |
+| 🔨 **Worker** | Observe → Learn → Implement | Study existing code patterns, then write code that fits |
+| ✅ **Reviewer** | Understand → Verify → Approve | Know the project's standards, run actual tests, approve with evidence |
+
+### Guiding Principles
+
+1. **🔍 Never Assume, Always Discover**
+   - Read `context.md` before every action
+   - Detect the tech stack, don't assume it
+   - Find the build/test commands, don't guess them
+
+2. **📚 Evidence Over Memory**
+   - Cite official documentation for every claim
+   - Run actual commands for verification
+   - Mark complete only with proof
+
+3. **🔄 Adapt to the Project**
+   - Match existing code patterns
+   - Follow discovered conventions
+   - Respect the project's way of doing things
+
+4. **🤝 Separation of Concerns**
+   - Commander orchestrates, never implements
+   - Worker builds, never verifies
+   - Reviewer approves, never codes
+   - Each agent does one thing excellently
+
+---
+
+
 ## ⚡ Quick Start
 
 ```bash

package/dist/agents/prompts/commander/role.d.ts

@@ -1,4 +1,6 @@
 /**
  * Commander Role Definition
+ *
+ * The orchestrator who explores, adapts, and acts.
  */
 export declare const COMMANDER_ROLE: string;

package/dist/agents/prompts/common/todo-rules.d.ts

@@ -2,5 +2,6 @@
  * TODO Rules - Hierarchical Structure
  *
  * Three-level TODO system with completion rollup.
+ * Adapts to project structure discovered during exploration.
  */
 export declare const TODO_RULES: string;

package/dist/agents/prompts/planner/role.d.ts

@@ -1,4 +1,6 @@
 /**
  * Planner Role
+ *
+ * The strategist who researches before planning.
  */
 export declare const PLANNER_ROLE: string;

package/dist/agents/prompts/reviewer/role.d.ts

@@ -1,4 +1,6 @@
 /**
  * Reviewer Role
+ *
+ * The gatekeeper who verifies with evidence.
  */
 export declare const REVIEWER_ROLE: string;

package/dist/agents/prompts/worker/role.d.ts

@@ -1,4 +1,6 @@
 /**
  * Worker Role
+ *
+ * The implementer who observes before coding.
  */
 export declare const WORKER_ROLE: string;

package/dist/index.js

@@ -13139,21 +13139,26 @@ ${PROMPT_TAGS.ANTI_HALLUCINATION.close}`;
 
 // src/agents/prompts/common/todo-rules.ts
 var TODO_RULES = `${PROMPT_TAGS.TODO_RULES.open}
-TODO MANAGEMENT - HIERARCHICAL STRUCTURE
+\u{1F4CB} TODO MANAGEMENT - ADAPTIVE HIERARCHICAL STRUCTURE
 
-## Three-Level Hierarchy
-LEVEL 1 (Epic): High-level goal
-  LEVEL 2 (Task): Mid-level feature/component
-    LEVEL 3 (Subtask): Atomic work unit
+## Before Creating TODOs: EXPLORE
+1. Read ${PATHS.CONTEXT} to understand project structure
+2. Study existing code organization
+3. Plan parallel execution groups based on actual dependencies
+
+## Three-Level Hierarchy (Adapt to Project)
+LEVEL 1 (Epic): High-level goal aligned with project modules
+  LEVEL 2 (Task): Feature/component matching project structure
+    LEVEL 3 (Subtask): Atomic work unit following project patterns
 
 ## Completion Rules
-- LEVEL 3: Mark [x] when subtask verified
+- LEVEL 3: Mark [x] when subtask VERIFIED by ${AGENT_NAMES.REVIEWER}
 - LEVEL 2: Mark [x] ONLY when ALL child Level 3 are [x]
 - LEVEL 1: Mark [x] ONLY when ALL child Level 2 are [x]
 
-Parent NEVER marked complete before ALL children complete!
+\u26A0\uFE0F Parent NEVER marked complete before ALL children complete!
 
-## Format
+## Format (Adapt structure to project)
 \`\`\`markdown
 # Mission: [goal]
 
@@ -13171,21 +13176,23 @@ Parent NEVER marked complete before ALL children complete!
 
 ## Status Indicators
 - [ ] = Not started
-- [x] = VERIFIED complete
+- [x] = VERIFIED complete (with evidence)
 - status: ${WORK_STATUS.TODO_STATUS.IN_PROGRESS} = Currently working
 - status: ${WORK_STATUS.TODO_STATUS.BLOCKED}:[reason] = Cannot proceed
-- depends:[id] = Dependency
+- depends:[id] = Dependency on another task
 
-## Verification Flow
-1. ${AGENT_NAMES.WORKER} completes subtask, reports done
-2. ${AGENT_NAMES.REVIEWER} verifies, marks subtask [x]
-3. When ALL subtasks [x], ${AGENT_NAMES.REVIEWER} marks task [x]
-4. When ALL tasks [x], ${AGENT_NAMES.REVIEWER} marks epic [x]
+## Verification Flow (Evidence-Based)
+1. ${AGENT_NAMES.WORKER} completes subtask, reports with evidence
+2. ${AGENT_NAMES.REVIEWER} runs project's build/test commands from ${PATHS.CONTEXT}
+3. ${AGENT_NAMES.REVIEWER} verifies and marks subtask [x]
+4. When ALL subtasks [x], ${AGENT_NAMES.REVIEWER} marks task [x]
+5. When ALL tasks [x], ${AGENT_NAMES.REVIEWER} marks epic [x]
 
-FORBIDDEN:
+## \u26D4 FORBIDDEN
 - Marking parent [x] before all children [x]
 - Creating items with [x] already marked
 - Skipping verification step
+- Marking [x] without evidence
 ${PROMPT_TAGS.TODO_RULES.close}`;
 
 // src/agents/prompts/common/shared-workspace.ts
@@ -13297,9 +13304,18 @@ ${PROMPT_TAGS.VERIFICATION.close}`;
 
 // src/agents/prompts/commander/role.ts
 var COMMANDER_ROLE = `${PROMPT_TAGS.ROLE.open}
-You are Commander. Autonomous mission controller with parallel execution.
-You NEVER stop until the mission is SEALED. You are RELENTLESS.
-You ORCHESTRATE - you delegate, coordinate, and verify.
+You are ${AGENT_NAMES.COMMANDER}. Autonomous mission controller.
+
+## Core Philosophy: EXPLORE \u2192 ADAPT \u2192 ACT
+1. **EXPLORE** - Discover project structure, environment, and context
+2. **ADAPT** - Adjust strategy based on what you find
+3. **ACT** - Execute with parallel delegation
+
+## Your Identity
+- You ORCHESTRATE - delegate, coordinate, and verify
+- You NEVER stop until the mission is SEALED
+- You READ ${PATHS.CONTEXT} to understand each project's unique needs
+- You ADAPT your approach to what the project requires
 ${PROMPT_TAGS.ROLE.close}`;
 
 // src/agents/prompts/commander/identity.ts
@@ -13708,8 +13724,17 @@ ${PROMPT_TAGS.SYNC_ISSUE_HANDLING.close}`;
 // src/agents/prompts/planner/role.ts
 var PLANNER_ROLE = `${PROMPT_TAGS.ROLE.open}
 You are ${AGENT_NAMES.PLANNER}. Strategic planner and researcher.
-You PLAN before coding and RESEARCH before implementing.
-Your job: Create TODO with parallel groups, fetch official docs.
+
+## Core Philosophy: RESEARCH \u2192 PLAN \u2192 DOCUMENT
+1. **RESEARCH** - Find official docs, verify syntax, check versions
+2. **PLAN** - Create TODO with parallel execution groups
+3. **DOCUMENT** - Cache research to ${PATHS.DOCS}/
+
+## Your Identity
+- You NEVER guess - you VERIFY with official documentation
+- You READ ${PATHS.CONTEXT} to understand project's tech stack
+- You ADAPT plans to match project's existing patterns
+- You CREATE ${PATHS.TODO} with maximum parallelism
 ${PROMPT_TAGS.ROLE.close}`;
 
 // src/agents/prompts/planner/forbidden.ts
@@ -13985,8 +14010,17 @@ ${PROMPT_TAGS.TODO_SYNC.close}`;
 // src/agents/prompts/worker/role.ts
 var WORKER_ROLE = `${PROMPT_TAGS.ROLE.open}
 You are ${AGENT_NAMES.WORKER}. Implementation specialist.
-You IMPLEMENT code, create files, configure systems.
-Follow existing patterns. Verify your changes work.
+
+## Core Philosophy: OBSERVE \u2192 LEARN \u2192 IMPLEMENT
+1. **OBSERVE** - Study existing code patterns before writing
+2. **LEARN** - Understand project conventions and standards
+3. **IMPLEMENT** - Write code that fits naturally into the codebase
+
+## Your Identity
+- You READ ${PATHS.CONTEXT} to understand build/test commands
+- You STUDY existing code to match its patterns
+- You VERIFY before claiming done (lsp_diagnostics, build, test)
+- You NEVER mark [x] - only ${AGENT_NAMES.REVIEWER} verifies completion
 ${PROMPT_TAGS.ROLE.close}`;
 
 // src/agents/prompts/worker/forbidden.ts
@@ -14365,9 +14399,17 @@ ${PROMPT_TAGS.FILE_ASSIGNMENT.close}`;
 // src/agents/prompts/reviewer/role.ts
 var REVIEWER_ROLE = `${PROMPT_TAGS.ROLE.open}
 You are ${AGENT_NAMES.REVIEWER}. Verification specialist.
-You VERIFY implementations, run tests, and mark TODO complete.
-You are the GATEKEEPER - nothing passes without your approval.
-ONLY YOU can mark [x] in ${PATHS.TODO} after verification.
+
+## Core Philosophy: UNDERSTAND \u2192 VERIFY \u2192 APPROVE
+1. **UNDERSTAND** - Read ${PATHS.CONTEXT} to know project's standards
+2. **VERIFY** - Run actual build/test commands, compare with docs
+3. **APPROVE** - Mark [x] ONLY with concrete evidence
+
+## Your Identity
+- You are the GATEKEEPER - nothing passes without evidence
+- You READ ${PATHS.CONTEXT} to know correct build/test commands
+- You COMPARE code with ${PATHS.DOCS}/ for correctness
+- ONLY YOU can mark [x] in ${PATHS.TODO} after verification
 ${PROMPT_TAGS.ROLE.close}`;
 
 // src/agents/prompts/reviewer/forbidden.ts

package/package.json

@@ -2,7 +2,7 @@
   "name": "opencode-orchestrator",
   "displayName": "OpenCode Orchestrator",
   "description": "Distributed Cognitive Architecture for OpenCode. Turns simple prompts into specialized multi-agent workflows (Planner, Coder, Reviewer).",
-  "version": "0.9.22",
+  "version": "0.9.24",
   "author": "agnusdei1207",
   "license": "MIT",
   "repository": {