opencode-pair-autonomy 1.0.0 → 1.0.1

package/README.md

@@ -1,4 +1,4 @@
-# oh-my-pair-code
+# opencode-pair-autonomy
 
 OpenCode harness with opinionated agent orchestration. One coordinator, eight specialized workers, automatic verify+review pipeline.
 
@@ -13,17 +13,17 @@ OpenCode harness with opinionated agent orchestration. One coordinator, eight sp
 
 ## Agents
 
-| Agent | Role | Model |
-|-------|------|-------|
-| **Yang** | Coordinator — plans, argues, delegates | claude-opus-4-6 |
-| **Thorfinn** | General implementation | claude-sonnet-4-6 |
-| **Ginko** | Web and doc research | claude-sonnet-4-6 |
-| **Kaiki** | Senior code review (read-only) | claude-opus-4-6 |
-| **Odokawa** | Cross-model review (read-only) | gpt-5.4 |
-| **Ozen** | Build, test, lint verification | claude-sonnet-4-6 |
-| **Skull Knight** | Scoped failure repair | claude-sonnet-4-6 |
-| **Paprika** | Frontend, Figma, browser testing | claude-sonnet-4-6 |
-| **Rajdhani** | Fast codebase exploration | claude-sonnet-4-6 |
+| Agent            | Role                                   | Model             |
+| ---------------- | -------------------------------------- | ----------------- |
+| **Yang**         | Coordinator — plans, argues, delegates | claude-opus-4-6   |
+| **Thorfinn**     | General implementation                 | claude-sonnet-4-6 |
+| **Ginko**        | Web and doc research                   | claude-sonnet-4-6 |
+| **Kaiki**        | Senior code review (read-only)         | claude-opus-4-6   |
+| **Odokawa**      | Cross-model review (read-only)         | gpt-5.4           |
+| **Ozen**         | Build, test, lint verification         | claude-sonnet-4-6 |
+| **Skull Knight** | Scoped failure repair                  | claude-sonnet-4-6 |
+| **Paprika**      | Frontend, Figma, browser testing       | claude-sonnet-4-6 |
+| **Rajdhani**     | Fast codebase exploration              | claude-sonnet-4-6 |
 
 ## MCP Servers
 
@@ -32,14 +32,14 @@ OpenCode harness with opinionated agent orchestration. One coordinator, eight sp
 ## Quick start
 
 ```bash
-bunx --bun github:cemalturkcan/oh-my-pair-code install
+bunx opencode-pair-autonomy install
 ```
 
 From source:
 
 ```bash
-git clone https://github.com/cemalturkcan/oh-my-pair-code.git
-cd oh-my-pair-code
+git clone https://github.com/cemalturkcan/opencode-pair-autonomy.git
+cd opencode-pair-autonomy
 bun install && bun run build && bun link
 opencode-pair-autonomy install
 ```
@@ -69,14 +69,14 @@ opencode-pair-autonomy init
 
 ## Hooks
 
-| Hook | What it does |
-|------|-------------|
-| `session.created` | Prepare session context injection |
-| `chat.message` | Inject mode, project docs, session memory (coordinator) or project facts (workers) |
-| `tool.execute.before` | Plan mode gate, long-running command detection |
-| `tool.execute.after` | Comment guard, file tracking, compact suggestions |
-| `session.idle` | Save session summary, promote learned patterns, cleanup old sessions |
-| `session.compacting` | Pre-compact observation snapshot |
+| Hook                  | What it does                                                                       |
+| --------------------- | ---------------------------------------------------------------------------------- |
+| `session.created`     | Prepare session context injection                                                  |
+| `chat.message`        | Inject mode, project docs, session memory (coordinator) or project facts (workers) |
+| `tool.execute.before` | Plan mode gate, long-running command detection                                     |
+| `tool.execute.after`  | Comment guard, file tracking, compact suggestions                                  |
+| `session.idle`        | Save session summary, promote learned patterns, cleanup old sessions               |
+| `session.compacting`  | Pre-compact observation snapshot                                                   |
 
 ## What install changes
 

package/dist/hooks/runtime.d.ts

@@ -29,7 +29,7 @@ declare function estimateTokens(chunks: string[]): number;
 export declare function resolveHookProfile(config: HarnessConfig): HookProfile;
 export declare function profileMatches(profile: HookProfile, allowed: HookProfile | HookProfile[]): boolean;
 /**
- * Primary (user-facing) agents that should receive session-context injection.
+ * Primary (user-facing) agent that should receive session-context injection.
  * Subagents spawned via Task tool should NOT get previous-session context
  * injected into their system prompt — it causes session mixing.
  */

package/dist/index.js

@@ -14802,17 +14802,6 @@ rajdhani (Sunny Boy) \u2014 sonnet-4-6 none
   Fast codebase recon. Send him first when entering unfamiliar territory.
 </WorkerCatalog>
 `;
-var DELEGATION_PRECISION = `
-<DelegationPrecision>
-Before delegating, read relevant files yourself. Your worker prompt MUST include:
-- Exact file paths and line numbers.
-- Specific type/function names involved.
-- "Change THIS, not THAT" when ambiguity exists.
-- Context not in the files (user intent, constraints).
-
-For broad scope (5+ files unknown), spawn rajdhani first for recon.
-</DelegationPrecision>
-`;
 var AUTOMATIC_WORKFLOW = `
 <AutomaticWorkflow>
 After implementation is complete:
@@ -14831,8 +14820,8 @@ You operate in two modes, controlled by /go and /plan commands:
 
 [Mode: Planning] (default at session start)
 - Discuss, argue, read files, create plan with TodoWrite.
-- You CANNOT spawn workers or execute implementation tools.
-- The system will block those attempts and remind you.
+- You CAN spawn read-only workers: rajdhani (scout), ginko (research), kaiki (review), odokawa (review).
+- You CANNOT spawn implementation workers (thorfinn, ozen, skull-knight, paprika) or use edit/write/patch tools.
 - When your plan is ready, tell the user and wait for /go.
 
 [Mode: Executing] (after /go)
@@ -14843,13 +14832,6 @@ You operate in two modes, controlled by /go and /plan commands:
 - After everything is done, mode returns to Planning.
 </PlanMode>
 `;
-var DIRECT_ACTION = `
-<DirectAction>
-Direct (no worker): Read/Glob/Grep/fff, research (context7/websearch/grep_app), database (pg-mcp/ssh-mcp/mariadb), read-only git, trivial edits (typo, config value, \u22645 lines in one file).
-Delegate: new files, 2+ files, >5 lines, build/test/lint, any filesystem-modifying bash.
-When in doubt, delegate.
-</DirectAction>
-`;
 var INPUT_HANDLING = `
 <InputHandling>
 On large paste: acknowledge immediately, process, respond. Never go silent.
@@ -14879,26 +14861,7 @@ Found: tell worker to skill_use first. Not found: proceed without.
 After novel implementations, suggest /create-skill.
 </SkillManagement>
 `;
-function buildCoordinatorPrompt(promptAppend) {
-  const sections = [
-    COORDINATOR_CORE,
-    RESPONSE_DISCIPLINE,
-    buildMcpCatalog(),
-    WORKER_CATALOG,
-    DELEGATION_PRECISION,
-    AUTOMATIC_WORKFLOW,
-    PLAN_MODE,
-    DIRECT_ACTION,
-    INPUT_HANDLING,
-    WORKER_CONTINUATION,
-    PARALLEL_SAFETY,
-    ACTION_SAFETY,
-    SKILL_MANAGEMENT
-  ];
-  return withPromptAppend(sections.join(`
-`), promptAppend);
-}
-var DELEGATION_EXP = `
+var DELEGATION = `
 <Delegation>
 ## Your Role
 
@@ -14977,13 +14940,13 @@ Use rajdhani when you need to understand an unfamiliar area of the codebase.
 Reading 1-2 files yourself is fine. For broader exploration, scout first \u2014 its compact report lets you write better worker prompts.
 </Delegation>
 `;
-function buildCoordinatorPromptExp(promptAppend) {
+function buildCoordinatorPrompt(promptAppend) {
   const sections = [
     COORDINATOR_CORE,
     RESPONSE_DISCIPLINE,
     buildMcpCatalog(),
     WORKER_CATALOG,
-    DELEGATION_EXP,
+    DELEGATION,
     AUTOMATIC_WORKFLOW,
     PLAN_MODE,
     INPUT_HANDLING,
@@ -15271,16 +15234,6 @@ function createHarnessAgents(config2) {
       tools: COORDINATOR_DISABLED_TOOLS,
       permission: { task: COORDINATOR_TASK_PERMISSIONS }
     }, overrides.yang),
-    "yang-exp": withOverride({
-      mode: "primary",
-      description: "Yang Wenli (Experimental) \u2014 Judgment-based delegation coordinator.",
-      model: "anthropic/claude-opus-4-6",
-      variant: "max",
-      prompt: buildCoordinatorPromptExp(overrides["yang-exp"]?.prompt_append),
-      color: "#7B68EE",
-      tools: COORDINATOR_DISABLED_TOOLS,
-      permission: { task: COORDINATOR_TASK_PERMISSIONS }
-    }, overrides["yang-exp"]),
     thorfinn: withOverride({
       mode: "subagent",
       hidden: true,
@@ -16389,7 +16342,7 @@ function resolveHookProfile(config2) {
 function profileMatches(profile, allowed) {
   return (Array.isArray(allowed) ? allowed : [allowed]).includes(profile);
 }
-var PRIMARY_AGENTS = new Set(["yang", "yang-exp"]);
+var PRIMARY_AGENTS = new Set(["yang"]);
 function resolveSessionID(value) {
   if (!value || typeof value !== "object")
     return;
@@ -16943,8 +16896,19 @@ function safeHook(name, hook) {
 var NODE_COMMAND_RE = /^(npm|pnpm|yarn|bun|npx|bunx|node|tsc|tsx|vite|next|nuxt|vitest|jest|eslint|prettier)\b/;
 var NODE_MODULES_BIN_RE = /node_modules\/\.bin\//;
 var PLAN_MODE_BLOCKED_TOOLS = new Set(["task", "edit", "write", "patch"]);
-function isAgentOrTaskTool(tool) {
-  return tool === "task" || tool.startsWith("task_");
+var PLAN_MODE_ALLOWED_AGENTS = new Set([
+  "ginko",
+  "kaiki",
+  "odokawa",
+  "rajdhani"
+]);
+function isBlockedTaskInPlanMode(tool, args) {
+  if (tool !== "task" && !tool.startsWith("task_"))
+    return false;
+  const target = typeof args.subagent_type === "string" ? args.subagent_type : typeof args.agent === "string" ? args.agent : undefined;
+  if (target && PLAN_MODE_ALLOWED_AGENTS.has(target))
+    return false;
+  return true;
 }
 function isNodeCommand(command) {
   return NODE_COMMAND_RE.test(command.trim()) || NODE_MODULES_BIN_RE.test(command);
@@ -16967,7 +16931,7 @@ function createPreToolUseHook(config2, runtime, profile) {
         runtime.incrementToolCount(sessionID);
       }
       if (sessionID && agent && PRIMARY_AGENTS.has(agent) && tool && runtime.getPlanMode(sessionID) === "planning") {
-        const blocked = PLAN_MODE_BLOCKED_TOOLS.has(tool) || isAgentOrTaskTool(tool);
+        const blocked = PLAN_MODE_BLOCKED_TOOLS.has(tool) || isBlockedTaskInPlanMode(tool, args);
         if (blocked) {
           const count = runtime.incrementPlanModeBlock(sessionID);
           const msg = count >= 3 ? "[PlanMode] STILL in planning mode. You have attempted execution tools multiple times. STOP trying. Complete your plan with TodoWrite. The user will /go to start execution." : "[PlanMode] You are in planning mode. Cannot use execution tools. Use Read/Glob/Grep/TodoWrite to continue planning. The user will /go to start execution.";

package/dist/prompts/coordinator.d.ts

@@ -1,2 +1 @@
 export declare function buildCoordinatorPrompt(promptAppend?: string): string;
-export declare function buildCoordinatorPromptExp(promptAppend?: string): string;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "opencode-pair-autonomy",
-  "version": "1.0.0",
+  "version": "1.0.1",
   "description": "OpenCode harness with opinionated agent orchestration. One coordinator, eight specialized workers, automatic verify+review pipeline.",
   "type": "module",
   "main": "dist/index.js",
@@ -88,7 +88,7 @@
   },
   "repository": {
     "type": "git",
-    "url": "git+https://github.com/cemalturkcan/oh-my-pair-code.git"
+    "url": "git+https://github.com/cemalturkcan/opencode-pair-autonomy.git"
   },
   "license": "MIT",
   "dependencies": {

package/vendor/skills/opencode-plugin-dev/SKILL.md

@@ -4,14 +4,17 @@ description: Build and refine OpenCode harness, MCP, plugin, prompt, and agent-t
 ---
 
 ## Purpose
-Use this skill for work in `oh-my-pair-code`, related OpenCode plugins, and MCP or agent-tooling repositories in this workspace.
+
+Use this skill for work in `opencode-pair-autonomy`, related OpenCode plugins, and MCP or agent-tooling repositories in this workspace.
 
 ## Use When
+
 - The task touches prompts, tool wiring, plugin install flows, MCP servers, vendor wrappers, or agent coordination.
 - The repo uses TypeScript with `@modelcontextprotocol/sdk`, Zod, CLI wiring, plugin manifests, or prompt assembly.
 - You need to preserve harness behavior while extending skill usage, tool exposure, or automation flows.
 
 ## Working Method
+
 1. Inspect existing prompt layers, install flow, and tool contracts before changing behavior.
 2. Reuse current naming, file placement, and config conventions for plugins, MCP wrappers, and prompts.
 3. Keep changes additive and architecture-preserving unless the user explicitly requests a structural shift.
@@ -19,6 +22,7 @@ Use this skill for work in `oh-my-pair-code`, related OpenCode plugins, and MCP
 5. Surface user-visible behavior changes clearly, especially when they affect session restart, tool availability, or external installs.
 
 ## Repo Conventions To Prefer
+
 - Keep prompt policy explicit, concrete, and action-oriented.
 - Prefer dedicated tools over shell when the harness already exposes a safer interface.
 - Preserve non-interactive shell assumptions for installs, git, and automation.
@@ -26,6 +30,7 @@ Use this skill for work in `oh-my-pair-code`, related OpenCode plugins, and MCP
 - Avoid hidden magic; make automation behavior legible in code and config.
 
 ## Guardrails
+
 - Do not silently broaden permissions, change security posture, or alter external side effects.
 - Do not add new dependencies when current tooling or plugins already solve the problem.
 - Do not make prompt changes that encourage architecture drift or speculative behavior.