opencode-resolve 0.1.5 → 0.1.6

package/README.ko.md

@@ -232,7 +232,7 @@ opencode plugin opencode-resolve --global --force
 
 **Companion 2 — `@slkiser/opencode-quota` (사용량/quota 추적)**
 
-> _그리고 **`@slkiser/opencode-quota`** — 컨텍스트 윈도우를 오염시키지 않고 실시간 토큰/quota 사용량을 보여줍니다. GLM coding-plan, OpenAI Plus/Pro, Qwen 등 지원. opencode-resolve가 실제로 얼마나 토큰 절약하는지 눈으로 확인하는 데 유용. 설치할까요? (권장)_
+> _그리고 **`@slkiser/opencode-quota`** — 컨텍스트 윈도우를 오염시키지 않고 실시간 토큰/quota 사용량을 보여줍니다. GLM coding-plan, OpenAI Plus/Pro, Qwen 등 지원. 내 사용량/잔여 quota 확인용으로 유용. 설치할까요? (권장)_
 
 사용자가 **예** (각 질문마다):
 
@@ -301,18 +301,15 @@ provider가 **하나도 설정되지 않은 경우** 진행을 멈추고, `openc
 
 > _`coder`와 `explorer`를 `zai-coding-plan/glm-5.1`로, `resolver`/`reviewer`/`deep-reviewer`를 `openai/gpt-5.5`로 핀닝합니다. 진행할까요?_
 
-#### 3d-bis. `coder` 모델 기준 `maxParallelSubagents` 결정
+#### 3d-bis. 사용자가 명시적으로 hard cap 원할 때만 `maxParallelSubagents` 설정
 
-opencode-resolve의 정체성은 **토큰 절약** — resolver는 빠르게 계획하고 유형별로 subagent를 디스패치하며, 병렬 캡은 역할당입니다. 캡은 **coder 모델 하나만 보고** 정합니다 (coder가 fan-out 주체이므로 rate budget이 거기서 결정):
+기본 흐름은 `maxParallelSubagents`를 `resolve.json`에 **쓰지 않습니다**. resolver prompt에 soft 가이드라인이 들어있어요 — 진짜 독립적인 작업이면 coder fan out, rate-limit 에러 보이면 backoff, explorer 무제한, reviewer/deep-reviewer/planner는 본질적으로 singleton. oh-my-openagent (per-model semaphore, 사실상 default 무제한) 와 OpenCode core (built-in concurrency 룰 없음) 의 처리와 동일한 결, rate limit을 이미 인지한 모델을 과제약하지 않음.
 
-| `coder` 모델 ID에 포함된 | `maxParallelSubagents` |
-|---|---|
-| `glm` (어떤 GLM이든 — 보통 GLM coding-plan) | **2** |
-| 그 외 (GPT, Claude, Gemini, local, …) | **4** |
+`maxParallelSubagents`는 사용자가 hard cap을 명시적으로 원할 때만 물어보세요. 흔한 케이스: GLM coding-plan 사용자가 "burst 절대 안 됨" 보장 원할 때. 권장 멘트:
 
-GLM의 제약은 GLM coding-plan의 계정당 rate limit 때문 — 강력한 GLM이라도 burst 디스패치 시 throttle 걸림. 그 외 family는 4를 잘 받아냄. strong/reasoner 모델은 보지 마세요 — 캡은 doer를 따름.
+> _(optional, 관련된 경우만)_ _`coder` 모델이 GLM이네요 — coding-plan은 burst 시 throttle 걸립니다. 안전하게 `maxParallelSubagents: 2`로 핀닝할까요?_
 
-단일(A) / 2단계(B) 케이스도 결국 coder가 쓰는 모델 하나만 보고 적용.
+그 외엔 필드 자체를 생략.
 
 #### 3e. `~/.config/opencode/resolve.json` 작성
 
@@ -348,8 +345,7 @@ GLM의 제약은 GLM coding-plan의 계정당 rate limit 때문 — 강력한 GL
     "researcher":    { "enabled": false }
   },
   "autoApprove": true,
-  "autoUpdate": true,
-  "maxParallelSubagents": 2
+  "autoUpdate": true
 }
 ```
 
@@ -382,7 +378,7 @@ GLM의 제약은 GLM coding-plan의 계정당 rate limit 때문 — 강력한 GL
 }
 ```
 
-`<provider>/<모델>` 플레이스홀더는 모두 사용자가 고른 **정확한** ID로 교체. `maxParallelSubagents`는 **coder 모델 ID에 `glm` 포함되면 2, 그 외 4**.
+`<provider>/<모델>` 플레이스홀더는 모두 사용자가 고른 **정확한** ID로 교체. **`maxParallelSubagents`는 사용자가 명시적으로 hard cap 요청한 경우에만 추가** (3d-bis 참고).
 
 `<provider>/<모델>` 플레이스홀더는 모두 사용자가 3b/3d에서 고른 **정확한** ID 문자열로 교체 — 임의 생성 금지, 자동완성 금지, 버전 드리프트 금지. 고른 모델을 `provider/model` 문자열로 매핑할 수 없으면 추측하지 말고 사용자에게 다시 물어보세요.
 

package/README.md

@@ -232,7 +232,7 @@ Before moving to model selection, ask the user about two **independent** compani
 
 **Companion 2 — `@slkiser/opencode-quota` (quota & usage tracking)**
 
-> _And **`@slkiser/opencode-quota`** — shows your live token/quota usage without polluting the context window. Supports GLM coding-plan, OpenAI Plus/Pro, Qwen, and more. Useful for actually seeing the token savings opencode-resolve gives you. Install it? (recommended)_
+> _And **`@slkiser/opencode-quota`** — shows your live token/quota usage without polluting the context window. Supports GLM coding-plan, OpenAI Plus/Pro, Qwen, and more. Useful for keeping an eye on your own usage / remaining quota. Install it? (recommended)_
 
 For each **yes** answer:
 
@@ -301,18 +301,15 @@ Confirm each pick back to the user before writing the file. Example:
 
 > _I'll pin `coder` and `explorer` to `zai-coding-plan/glm-5.1`, and `resolver`/`reviewer`/`deep-reviewer` to `openai/gpt-5.5`. Proceed?_
 
-#### 3d-bis. Pick `maxParallelSubagents` from the `coder` model
+#### 3d-bis. Skip `maxParallelSubagents` unless the user explicitly wants a hard cap
 
-opencode-resolve's identity is **token efficiency** — the resolver plans fast and dispatches subagents by type, and the parallel cap is per-role. The cap is determined by the `coder` model only — coder is the fan-out subject, so its rate budget is what matters:
+The default flow does **not** write `maxParallelSubagents` to `resolve.json`. The resolver prompt ships with soft fan-out guidance — fan out coders when work is genuinely independent, back off on rate-limit errors, explorer is unrestricted, reviewer/deep-reviewer/planner are singletons by nature. This matches how oh-my-openagent (per-model semaphore, default no real cap) and OpenCode core (no built-in concurrency rule) handle the same concern, and avoids overconstraining a model that's already aware of rate limits.
 
-| `coder` model ID contains | `maxParallelSubagents` |
-|---|---|
-| `glm` (any GLM model — most often the GLM coding-plan) | **2** |
-| anything else (GPT, Claude, Gemini, local, …) | **4** |
+Only ask about `maxParallelSubagents` if the user explicitly mentions wanting to cap fan-out. Common case: GLM coding-plan users who want to **guarantee** they never burst beyond 1 or 2 coder calls. Suggested wording:
 
-The GLM constraint comes from the GLM coding-plan's per-account rate limits — even a powerful GLM gets throttled under burst dispatch. Every other family tested tolerates 4 well and the resolver gets meaningful speedup. Don't look at the strong/reasoner model — the cap follows the doer, not the judge.
+> _(optional, ask only if relevant)_ _Your `coder` model is GLM — the coding-plan throttles under bursts. Pin `maxParallelSubagents: 2` to be safe?_
 
-For single-tier (A) and two-tier (B), apply the same rule to whichever model the `coder` agent ends up using.
+Otherwise, omit the field entirely.
 
 #### 3e. Write `~/.config/opencode/resolve.json`
 
@@ -348,8 +345,7 @@ For single-tier (A) and two-tier (B), apply the same rule to whichever model the
     "researcher":    { "enabled": false }
   },
   "autoApprove": true,
-  "autoUpdate": true,
-  "maxParallelSubagents": 2
+  "autoUpdate": true
 }
 ```
 
@@ -382,7 +378,7 @@ For single-tier (A) and two-tier (B), apply the same rule to whichever model the
 }
 ```
 
-Replace every `<provider>/<model>` placeholder with the **exact** ID strings the user picked. Set `maxParallelSubagents` to **2** if the `coder` model ID contains `glm`, otherwise **4**.
+Replace every `<provider>/<model>` placeholder with the **exact** ID strings the user picked. **Do not** add `maxParallelSubagents` unless the user explicitly asked for a hard cap (see 3d-bis).
 
 Replace every `<provider>/<model>` placeholder with the **exact** ID strings the user picked in 3b/3d — no inventing, no autocompletion, no version drift. If you cannot map the picked model to a `provider/model` string, ask the user to clarify rather than guessing.
 

package/dist/index.js

@@ -76,10 +76,14 @@ const VALID_AGENT_KEYS = new Set([
     "permission",
 ]);
 function buildResolverPrompt(maxParallelSubagents) {
-    const limit = Math.max(1, Math.trunc(maxParallelSubagents));
-    const parallelRule = limit === 1
-        ? "CRITICAL: Dispatch at most ONE subagent of each role concurrently. Never run two coders in parallel. Wait for an in-flight coder to finish before dispatching another."
-        : `CRITICAL: Dispatch at most ${limit} subagents of the same role concurrently. Never exceed ${limit} coders in parallel. Wait for in-flight subagents of a given role to finish before dispatching more of that role.`;
+    const explicitLimit = typeof maxParallelSubagents === "number" && Number.isFinite(maxParallelSubagents)
+        ? Math.max(1, Math.trunc(maxParallelSubagents))
+        : undefined;
+    const parallelRule = explicitLimit === undefined
+        ? "Parallel dispatch policy: fan out coder dispatches when the work is genuinely independent (different files, no shared state). Rate limits are per-model — coder/silver-tier models (e.g. GLM coding-plan) can throttle under heavy burst, so back off if you see rate-limit errors. Explorer runs on a lighter model and fans out freely when scope discovery genuinely needs it. Reviewer, deep-reviewer, and planner are dispatched as singletons by their nature."
+        : explicitLimit === 1
+            ? "CRITICAL: User has pinned the coder concurrency cap to 1. Dispatch at most ONE coder concurrently. Wait for an in-flight coder to finish before dispatching another. Explorer is unrestricted (read-only, light model). Reviewer, deep-reviewer, and planner are singletons by nature."
+            : `CRITICAL: User has pinned the coder concurrency cap to ${explicitLimit}. Dispatch at most ${explicitLimit} coders concurrently. Wait for in-flight coders to finish before dispatching more. Explorer is unrestricted (read-only, light model). Reviewer, deep-reviewer, and planner are singletons by nature.`;
     return [
         "You are Resolver, the context-efficient orchestrator agent for OpenCode Resolve.",
         "Your job is to drive the user's task to a verified resolution using minimal context and the fewest LLM calls possible.",
@@ -159,7 +163,7 @@ const DEFAULT_AGENT_CONFIG = {
         color: "#FF7AC6",
         maxSteps: 30,
         description: "Primary orchestrator in the fixed-role verified loop (resolver→coder). Decomposes work into verified checkpoints, dispatches coder, verifies each, and carries forward progress. Internal subagents (explorer, reviewer, deep-reviewer) are available by default but dispatched only when justified.",
-        prompt: buildResolverPrompt(DEFAULT_MAX_PARALLEL_SUBAGENTS),
+        prompt: buildResolverPrompt(undefined),
         permission: {
             edit: "ask",
             bash: "ask",
@@ -389,7 +393,7 @@ function applyResolveConfig(config, resolveConfig) {
     const models = { ...DEFAULT_MODELS, ...resolveConfig.models };
     const defaultModel = typeof config.model === "string" ? config.model : undefined;
     const autoApprove = resolveConfig.autoApprove !== false;
-    const maxParallelSubagents = resolveConfig.maxParallelSubagents ?? DEFAULT_MAX_PARALLEL_SUBAGENTS;
+    const maxParallelSubagents = resolveConfig.maxParallelSubagents;
     config.agent ??= {};
     for (const name of Object.keys(DEFAULT_AGENT_CONFIG)) {
         const override = resolveConfig.agents?.[name];
@@ -451,7 +455,6 @@ function defaultResolveConfig() {
         context7: true,
         commands: false,
         autoApprove: true,
-        maxParallelSubagents: DEFAULT_MAX_PARALLEL_SUBAGENTS,
         autoUpdate: true,
     };
 }

package/opencode-resolve.example.json

@@ -5,7 +5,6 @@
   "commands": false,
   "autoApprove": true,
   "autoUpdate": true,
-  "maxParallelSubagents": 2,
   "models": {},
   "agents": {
     "coder": {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "opencode-resolve",
-  "version": "0.1.5",
+  "version": "0.1.6",
   "description": "OpenCode plugin that adds a small coder/reviewer agent set while preserving native plan/build behavior.",
   "type": "module",
   "homepage": "https://github.com/jshsakura/opencode-resolve#readme",

package/scripts/postinstall.mjs

@@ -13,7 +13,6 @@ const exampleConfigPath = join(root, "opencode-resolve.example.json")
 
 const ADDITIVE_DEFAULTS = {
   autoApprove: true,
-  maxParallelSubagents: 2,
 }
 
 if (process.env.OPENCODE_RESOLVE_SKIP_POSTINSTALL === "1") {