opencode-swarm 4.0.0 → 4.1.0

package/README.md

@@ -1,8 +1,8 @@
 <p align="center">
-  <img src="https://img.shields.io/badge/version-4.0.0-blue" alt="Version">
+  <img src="https://img.shields.io/badge/version-4.1.0-blue" alt="Version">
   <img src="https://img.shields.io/badge/license-MIT-green" alt="License">
   <img src="https://img.shields.io/badge/opencode-plugin-purple" alt="OpenCode Plugin">
-  <img src="https://img.shields.io/badge/agents-7-orange" alt="Agents">
+  <img src="https://img.shields.io/badge/agents-8-orange" alt="Agents">
 </p>
 
 <h1 align="center">🐝 OpenCode Swarm</h1>
@@ -36,12 +36,13 @@ Other Frameworks:
 
 OpenCode Swarm:
 ├── Architect analyzes request
-├── Explorer scans codebase
+├── Explorer scans codebase (+ gap analysis)
 ├── @sme consulted on security domain
 ├── Architect creates phased plan with acceptance criteria
-├── Phase 1: User model → Review → Tests → ✓
-├── Phase 2: Auth logic → Review → Tests → ✓
-├── Phase 3: Session management → Review → Tests → ✓
+├── @critic reviews plan → APPROVED
+├── Phase 1: User model → Review → Tests (run + PASS) → ✓
+├── Phase 2: Auth logic → Review → Tests (run + PASS) → ✓
+├── Phase 3: Session management → Review → Tests (run + PASS) → ✓
 └── Result: Working code. Documented decisions. Resumable progress.
 ```
 
@@ -127,16 +128,24 @@ OpenCode Swarm:
                                     │
                                     ▼
 ┌─────────────────────────────────────────────────────────────────────────┐
+│  PHASE 4.5: Critic Gate                                                 │
+│             @critic reviews plan → APPROVED / NEEDS_REVISION / REJECTED│
+│             Max 2 revision cycles before escalating to user             │
+└─────────────────────────────────────────────────────────────────────────┘
+                                    │
+                                    ▼
+┌─────────────────────────────────────────────────────────────────────────┐
 │  PHASE 5: Execute (per task)                                            │
 │                                                                         │
-│   ┌─────────┐    ┌────────────┐    ┌──────────┐                        │
-│   │ @coder  │ →  │ @reviewer  │ →  │  @test   │                        │
-│   │ 1 task  │    │ check all  │    │ generate │                        │
-│   └─────────┘    └────────────┘    └──────────┘                        │
-│        │                                              │                 │
-│        └──── If rejected: retry with feedback ────────┘                 │
+│   ┌─────────┐    ┌────────────┐    ┌──────────────┐                    │
+│   │ @coder  │ →  │ @reviewer  │ →  │    @test     │                    │
+│   │ 1 task  │    │ check all  │    │ write + run  │                    │
+│   └─────────┘    └────────────┘    └──────────────┘                    │
+│        │               │                   │                            │
+│        │     If REJECTED: retry    If FAIL: fix + retest               │
+│        └───────────────┘                                                │
 │                                                                         │
-│   Update plan.md: [x] Task complete                                     │
+│   Update plan.md: [x] Task complete (only if PASS)                      │
 │   Next task...                                                          │
 └─────────────────────────────────────────────────────────────────────────┘
                                     │
@@ -227,6 +236,7 @@ Swarm lets you mix models strategically:
     "coder": { "model": "anthropic/claude-sonnet-4-5" },
     "sme": { "model": "google/gemini-2.0-flash" },
     "reviewer": { "model": "openai/gpt-4o" },
+    "critic": { "model": "google/gemini-2.0-flash" },
     "test_engineer": { "model": "google/gemini-2.0-flash" }
   }
 }
@@ -239,7 +249,8 @@ Swarm lets you mix models strategically:
 | Coder | Implementation | Best coding model you have |
 | SME | Domain knowledge | Fast recall, not deep reasoning |
 | Reviewer | Finding flaws | **Different vendor catches different bugs** |
-| Test Engineer | Test generation | Independent perspective on edge cases |
+| Critic | Plan review | Catches scope issues before any code is written |
+| Test Engineer | Test + run | Writes tests, runs them, reports PASS/FAIL |
 
 **If Claude writes code and GPT reviews it, GPT catches Claude's blindspots.** This is why real teams have code review.
 
@@ -283,8 +294,8 @@ Run different model configurations simultaneously. Perfect for:
 
 | Swarm | Agents |
 |-------|--------|
-| `cloud` (default) | `architect`, `explorer`, `coder`, `sme`, `reviewer`, `test_engineer` |
-| `local` | `local_architect`, `local_explorer`, `local_coder`, `local_sme`, `local_reviewer`, `local_test_engineer` |
+| `cloud` (default) | `architect`, `explorer`, `coder`, `sme`, `reviewer`, `critic`, `test_engineer` |
+| `local` | `local_architect`, `local_explorer`, `local_coder`, `local_sme`, `local_reviewer`, `local_critic`, `local_test_engineer` |
 
 The first swarm (or one named "default") creates unprefixed agents. Additional swarms prefix all agent names.
 
@@ -333,12 +344,13 @@ bunx opencode-swarm install
 | Agent | Role |
 |-------|------|
 | `coder` | Implements ONE task at a time with full context |
-| `test_engineer` | Generates tests for each completed task |
+| `test_engineer` | Generates tests, runs them, and reports structured PASS/FAIL verdicts |
 
 ### ✅ Quality Assurance
 | Agent | Role |
 |-------|------|
 | `reviewer` | Combined correctness + security review. The architect specifies CHECK dimensions (security, correctness, edge-cases, performance, etc.) per call. |
+| `critic` | Plan review gate. Reviews the architect's plan BEFORE implementation — checks completeness, feasibility, scope, dependencies, and flags AI-slop. |
 
 ---
 
@@ -354,6 +366,7 @@ Create `~/.config/opencode/opencode-swarm.json`:
     "coder": { "model": "anthropic/claude-sonnet-4-5" },
     "sme": { "model": "google/gemini-2.0-flash" },
     "reviewer": { "model": "openai/gpt-4o" },
+    "critic": { "model": "google/gemini-2.0-flash" },
     "test_engineer": { "model": "google/gemini-2.0-flash" }
   }
 }

package/dist/agents/critic.d.ts

@@ -0,0 +1,2 @@
+import type { AgentDefinition } from './architect';
+export declare function createCriticAgent(model: string, customPrompt?: string, customAppendPrompt?: string): AgentDefinition;

package/dist/agents/index.d.ts

@@ -12,7 +12,8 @@ export declare function createAgents(config?: PluginConfig): AgentDefinition[];
 export declare function getAgentConfigs(config?: PluginConfig): Record<string, SDKAgentConfig>;
 export { createArchitectAgent } from './architect';
 export { createCoderAgent } from './coder';
+export { createCriticAgent } from './critic';
 export { createExplorerAgent } from './explorer';
 export { createReviewerAgent } from './reviewer';
-export { createTestEngineerAgent } from './test-engineer';
 export { createSMEAgent } from './sme';
+export { createTestEngineerAgent } from './test-engineer';

package/dist/config/constants.d.ts

@@ -1,8 +1,8 @@
-export declare const QA_AGENTS: readonly ["reviewer"];
+export declare const QA_AGENTS: readonly ["reviewer", "critic"];
 export declare const PIPELINE_AGENTS: readonly ["explorer", "coder", "test_engineer"];
 export declare const ORCHESTRATOR_NAME: "architect";
-export declare const ALL_SUBAGENT_NAMES: readonly ["sme", "reviewer", "explorer", "coder", "test_engineer"];
-export declare const ALL_AGENT_NAMES: readonly ["architect", "sme", "reviewer", "explorer", "coder", "test_engineer"];
+export declare const ALL_SUBAGENT_NAMES: readonly ["sme", "reviewer", "critic", "explorer", "coder", "test_engineer"];
+export declare const ALL_AGENT_NAMES: readonly ["architect", "sme", "reviewer", "critic", "explorer", "coder", "test_engineer"];
 export type QAAgentName = (typeof QA_AGENTS)[number];
 export type PipelineAgentName = (typeof PIPELINE_AGENTS)[number];
 export type AgentName = (typeof ALL_AGENT_NAMES)[number];

package/dist/index.js

@@ -11,7 +11,7 @@ var __export = (target, all) => {
 };
 
 // src/config/constants.ts
-var QA_AGENTS = ["reviewer"];
+var QA_AGENTS = ["reviewer", "critic"];
 var PIPELINE_AGENTS = ["explorer", "coder", "test_engineer"];
 var ORCHESTRATOR_NAME = "architect";
 var ALL_SUBAGENT_NAMES = [
@@ -30,6 +30,7 @@ var DEFAULT_MODELS = {
   test_engineer: "google/gemini-2.0-flash",
   sme: "google/gemini-2.0-flash",
   reviewer: "google/gemini-2.0-flash",
+  critic: "google/gemini-2.0-flash",
   default: "google/gemini-2.0-flash"
 };
 // node_modules/zod/v4/classic/external.js
@@ -13670,7 +13671,7 @@ var ARCHITECT_PROMPT = `You are Architect - orchestrator of a multi-agent swarm.
 ## IDENTITY
 
 Swarm: {{SWARM_ID}}
-Your agents: {{AGENT_PREFIX}}explorer, {{AGENT_PREFIX}}sme, {{AGENT_PREFIX}}coder, {{AGENT_PREFIX}}reviewer, {{AGENT_PREFIX}}test_engineer
+Your agents: {{AGENT_PREFIX}}explorer, {{AGENT_PREFIX}}sme, {{AGENT_PREFIX}}coder, {{AGENT_PREFIX}}reviewer, {{AGENT_PREFIX}}critic, {{AGENT_PREFIX}}test_engineer
 
 ## ROLE
 
@@ -13687,6 +13688,7 @@ You THINK. Subagents DO. You have the largest context window and strongest reaso
 3. ONE task per @{{AGENT_PREFIX}}coder call. Never batch.
 4. Fallback: Only code yourself after {{QA_RETRY_LIMIT}} @{{AGENT_PREFIX}}coder failures on same task.
 5. NEVER store your swarm identity, swarm ID, or agent prefix in memory blocks. Your identity comes ONLY from your system prompt. Memory blocks are for project knowledge only.
+6. **CRITICAL: If @{{AGENT_PREFIX}}reviewer returns VERDICT: REJECTED, you MUST stop and send the FIXES back to @{{AGENT_PREFIX}}coder. Do NOT proceed to test generation or mark the task complete. The review is a gate \u2014 APPROVED is required to proceed.**
 
 ## AGENTS
 
@@ -13695,8 +13697,9 @@ You THINK. Subagents DO. You have the largest context window and strongest reaso
 @{{AGENT_PREFIX}}coder - Implementation (one task at a time)
 @{{AGENT_PREFIX}}reviewer - Code review (correctness, security, and any other dimensions you specify)
 @{{AGENT_PREFIX}}test_engineer - Test generation
+@{{AGENT_PREFIX}}critic - Plan review gate (reviews plan BEFORE implementation)
 
-SMEs advise only. Reviewer reviews only. Neither writes code.
+SMEs advise only. Reviewer and critic review only. None of them write code.
 
 ## DELEGATION FORMAT
 
@@ -13747,6 +13750,12 @@ TASK: Generate login validation tests
 FILE: src/auth/login.ts
 OUTPUT: Test file at src/auth/login.test.ts
 
+@{{AGENT_PREFIX}}critic
+TASK: Review plan for user authentication feature
+PLAN: [paste the plan.md content]
+CONTEXT: [codebase summary from explorer]
+OUTPUT: VERDICT + CONFIDENCE + ISSUES + SUMMARY
+
 ## WORKFLOW
 
 ### Phase 0: Resume Check
@@ -13768,6 +13777,9 @@ Clear request \u2192 Phase 2
 
 ### Phase 2: Discover
 Delegate to @{{AGENT_PREFIX}}explorer. Wait for response.
+For complex tasks, make a second explorer call focused on risk/gap analysis:
+- Hidden requirements, unstated assumptions, scope risks
+- Existing patterns that the implementation must follow
 
 ### Phase 3: Consult SMEs
 Check .swarm/context.md for cached guidance first.
@@ -13785,17 +13797,25 @@ Create .swarm/plan.md:
 Create .swarm/context.md:
 - Decisions, patterns, SME cache, file map
 
+### Phase 4.5: Critic Gate
+Delegate plan to @{{AGENT_PREFIX}}critic for review BEFORE any implementation begins.
+- Send the full plan.md content and codebase context summary
+- **APPROVED** \u2192 Proceed to Phase 5
+- **NEEDS_REVISION** \u2192 Revise the plan based on critic feedback, then resubmit (max 2 revision cycles)
+- **REJECTED** \u2192 Inform the user of fundamental issues and ask for guidance before proceeding
+
 ### Phase 5: Execute
 For each task (respecting dependencies):
 
 5a. @{{AGENT_PREFIX}}coder - Implement (MANDATORY)
 5b. @{{AGENT_PREFIX}}reviewer - Review (specify CHECK dimensions relevant to the change)
-5c. Result:
-    - APPROVED \u2192 5d
-    - REJECTED (attempt < {{QA_RETRY_LIMIT}}) \u2192 Feedback to @{{AGENT_PREFIX}}coder, retry
-    - REJECTED (attempt {{QA_RETRY_LIMIT}}) \u2192 Escalate, handle directly
-5d. @{{AGENT_PREFIX}}test_engineer - Generate tests
-5e. Update plan.md [x], proceed to next task
+5c. **GATE - Check VERDICT:**
+    - **APPROVED** \u2192 Proceed to 5d
+    - **REJECTED** (attempt < {{QA_RETRY_LIMIT}}) \u2192 STOP. Send FIXES to @{{AGENT_PREFIX}}coder with specific changes. Retry from 5a. Do NOT proceed to 5d.
+    - **REJECTED** (attempt {{QA_RETRY_LIMIT}}) \u2192 STOP. Escalate to user or handle directly.
+5d. @{{AGENT_PREFIX}}test_engineer - Generate AND run tests (ONLY if 5c = APPROVED). Expect VERDICT: PASS/FAIL.
+5e. If test VERDICT is FAIL \u2192 Send failures to @{{AGENT_PREFIX}}coder for fixes, then re-run from 5b.
+5f. Update plan.md [x], proceed to next task (ONLY if tests PASS)
 
 ### Phase 6: Phase Complete
 1. @{{AGENT_PREFIX}}explorer - Rescan
@@ -13899,6 +13919,63 @@ ${customAppendPrompt}`;
   };
 }
 
+// src/agents/critic.ts
+var CRITIC_PROMPT = `You are Critic. You review the Architect's plan BEFORE implementation begins. You are a quality gate.
+
+INPUT FORMAT:
+TASK: Review plan for [description]
+PLAN: [the plan content \u2014 phases, tasks, file changes]
+CONTEXT: [codebase summary, constraints]
+
+REVIEW CHECKLIST:
+- Completeness: Are all requirements addressed? Missing edge cases?
+- Feasibility: Can each task actually be implemented as described? Are file paths real?
+- Scope: Is the plan doing too much or too little? Feature creep detection.
+- Dependencies: Are task dependencies correct? Will ordering work?
+- Risk: Are high-risk changes identified? Is there a rollback path?
+- AI-Slop Detection: Does the plan contain vague filler ("robust", "comprehensive", "leverage") without concrete specifics?
+
+OUTPUT FORMAT:
+VERDICT: APPROVED | NEEDS_REVISION | REJECTED
+CONFIDENCE: HIGH | MEDIUM | LOW
+ISSUES: [max 5 issues, each with: severity (CRITICAL/MAJOR/MINOR), description, suggested fix]
+SUMMARY: [1-2 sentence overall assessment]
+
+RULES:
+- Max 5 issues per review (focus on highest impact)
+- Be specific: reference exact task numbers and descriptions
+- CRITICAL issues block approval (VERDICT must be NEEDS_REVISION or REJECTED)
+- MAJOR issues should trigger NEEDS_REVISION
+- MINOR issues can be noted but don't block APPROVED
+- No code writing
+- No delegation
+- Don't reject for style/formatting \u2014 focus on substance
+- If the plan is fundamentally sound with only minor concerns, APPROVE it`;
+function createCriticAgent(model, customPrompt, customAppendPrompt) {
+  let prompt = CRITIC_PROMPT;
+  if (customPrompt) {
+    prompt = customPrompt;
+  } else if (customAppendPrompt) {
+    prompt = `${CRITIC_PROMPT}
+
+${customAppendPrompt}`;
+  }
+  return {
+    name: "critic",
+    description: "Plan critic. Reviews the architect's plan before implementation begins \u2014 checks completeness, feasibility, scope, dependencies, and flags AI-slop.",
+    config: {
+      model,
+      temperature: 0.1,
+      prompt,
+      tools: {
+        write: false,
+        edit: false,
+        patch: false
+      }
+    }
+  };
+}
+
 // src/agents/explorer.ts
 var EXPLORER_PROMPT = `You are Explorer. You analyze codebases.
 
@@ -14012,48 +14089,6 @@ ${customAppendPrompt}`;
   };
 }
 
-// src/agents/test-engineer.ts
-var TEST_ENGINEER_PROMPT = `You are Test Engineer. You generate tests.
-
-INPUT FORMAT:
-TASK: Generate tests for [description]
-FILE: [source file path]
-OUTPUT: [test file path]
-
-COVERAGE:
-- Happy path: normal inputs
-- Edge cases: empty, null, boundaries
-- Errors: invalid inputs, failures
-
-RULES:
-- Match language (PowerShell \u2192 Pester, Python \u2192 pytest, TS \u2192 vitest/jest)
-- Tests must be runnable
-- Include setup/teardown if needed
-- No delegation
-
-OUTPUT:
-Write test file to specified OUTPUT path.
-DONE: [count] tests covering [areas]`;
-function createTestEngineerAgent(model, customPrompt, customAppendPrompt) {
-  let prompt = TEST_ENGINEER_PROMPT;
-  if (customPrompt) {
-    prompt = customPrompt;
-  } else if (customAppendPrompt) {
-    prompt = `${TEST_ENGINEER_PROMPT}
-
-${customAppendPrompt}`;
-  }
-  return {
-    name: "test_engineer",
-    description: "Testing and validation specialist. Generates test cases and runnable validation scripts for approved code.",
-    config: {
-      model,
-      temperature: 0.2,
-      prompt
-    }
-  };
-}
-
 // src/agents/sme.ts
 var SME_PROMPT = `You are SME (Subject Matter Expert). You provide deep domain-specific technical guidance on whatever domain the Architect requests.
 
@@ -14100,6 +14135,57 @@ ${customAppendPrompt}`;
   };
 }
 
+// src/agents/test-engineer.ts
+var TEST_ENGINEER_PROMPT = `You are Test Engineer. You generate tests AND run them.
+
+INPUT FORMAT:
+TASK: Generate tests for [description]
+FILE: [source file path]
+OUTPUT: [test file path]
+
+COVERAGE:
+- Happy path: normal inputs
+- Edge cases: empty, null, boundaries
+- Errors: invalid inputs, failures
+
+RULES:
+- Match language (PowerShell \u2192 Pester, Python \u2192 pytest, TS \u2192 vitest/jest)
+- Tests must be runnable
+- Include setup/teardown if needed
+- No delegation
+
+WORKFLOW:
+1. Write test file to the specified OUTPUT path
+2. Run the tests using the appropriate test runner
+3. Report results using the output format below
+
+If tests fail, include the failure output so the architect can send fixes to the coder.
+
+OUTPUT FORMAT:
+VERDICT: PASS | FAIL
+TESTS: [total count] tests, [pass count] passed, [fail count] failed
+FAILURES: [list of failed test names + error messages, if any]
+COVERAGE: [areas covered]`;
+function createTestEngineerAgent(model, customPrompt, customAppendPrompt) {
+  let prompt = TEST_ENGINEER_PROMPT;
+  if (customPrompt) {
+    prompt = customPrompt;
+  } else if (customAppendPrompt) {
+    prompt = `${TEST_ENGINEER_PROMPT}
+
+${customAppendPrompt}`;
+  }
+  return {
+    name: "test_engineer",
+    description: "Testing and validation specialist. Generates test cases, runs them, and reports structured PASS/FAIL verdicts.",
+    config: {
+      model,
+      temperature: 0.2,
+      prompt
+    }
+  };
+}
+
 // src/agents/index.ts
 function getModelForAgent(agentName, swarmAgents, swarmPrefix) {
   let baseAgentName = agentName;
@@ -14192,6 +14278,12 @@ If you call @coder instead of @${swarmId}_coder, the call will FAIL or go to the
     reviewer.name = prefixName("reviewer");
     agents.push(applyOverrides(reviewer, swarmAgents, swarmPrefix));
   }
+  if (!isAgentDisabled("critic", swarmAgents, swarmPrefix)) {
+    const criticPrompts = getPrompts("critic");
+    const critic = createCriticAgent(getModel("critic"), criticPrompts.prompt, criticPrompts.appendPrompt);
+    critic.name = prefixName("critic");
+    agents.push(applyOverrides(critic, swarmAgents, swarmPrefix));
+  }
   if (!isAgentDisabled("test_engineer", swarmAgents, swarmPrefix)) {
     const testPrompts = getPrompts("test_engineer");
     const testEngineer = createTestEngineerAgent(getModel("test_engineer"), testPrompts.prompt, testPrompts.appendPrompt);

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "opencode-swarm",
-	"version": "4.0.0",
+	"version": "4.1.0",
 	"description": "Architect-centric agentic swarm plugin for OpenCode - hub-and-spoke orchestration with SME consultation, code generation, and QA review",
 	"main": "dist/index.js",
 	"types": "dist/index.d.ts",