peaks-loop 4.0.0-beta.6 → 4.0.0-beta.7

package/CHANGELOG.md

@@ -2,6 +2,68 @@
 
 ## [Unreleased]
 
+## 4.0.0-beta.7 — 2026-07-09
+
+### Added — zcode adapter (9th IDE) + runtime model detection
+
+- **`peaks ide model --current` CLI** — `src/cli/commands/ide-commands.ts` (80 lines) + register in `src/cli/program.ts`. Outputs `{ modelId, detected, registeredAdapters }`. Available as a top-level primitive (not adapter-specific).
+- **`detectCurrentIdeModel()` service** — `src/services/ide/current-model-detector.ts` (43 lines). Walks the adapter registry in insertion order, calls each adapter's optional `detectCurrentModel()` with try/catch isolation — one bad adapter cannot poison the chain.
+- **zcode-adapter `detectCurrentModel()` method** — `src/services/ide/adapters/zcode-adapter.ts`. Reads `~/.zcode/v2/config.json`, resolves active provider via 4-tier priority chain (env override → non-`builtin:` provider → first enabled → first provider), returns the model's id. Live test: `peaks ide model --current` against a real z-code installation → `"modelId": "M3"`.
+- **`IdeAdapter` interface optional field** — `readonly detectCurrentModel?: () => Promise<string | undefined>` in `src/services/ide/ide-types.ts`. Back-compat: 8 existing adapters without the field are unaffected.
+- **`getStrongestModelIdAsync()` async variant** — `src/services/config/model-routing.ts`. Sync `getStrongestModelId()` is unchanged (rd-service compat); new async variant lets future async callers fall back to runtime probe instead of `'claude-opus-4-7'`.
+
+### Changed — install no longer writes hardcoded default model
+
+- **`scripts/install-skills.mjs`** — removed `model: 'sonnet'` default and `providers.minimax.model` default from `createConfigDefaults()`. After install, `~/.peaks/config.json` no longer contains a Claude model recommendation; users set their own via `peaks config set model <id>` (or leave unset and the back-compat fallback applies).
+- **`STRONGEST_MODEL_ID` constant removed** — replaced by `getStrongestModelId(config?)` function in `src/services/config/model-routing.ts`. Resolution order: `config.model` → `PEAKS_STRONGEST_MODEL_DEFAULT` env var → back-compat `'claude-opus-4-7'`.
+- **`workflow-router-service.ts` / `minimax-worker-service.ts` / `rd-service.ts`** — threaded dynamic modelId through 16 call sites in `workflow-router-service.ts`, replaced literal `'claude-opus-4-7'` in `minimax-worker-service.ts` (×2) and `rd-service.ts`.
+
+### Added — zcode IDE (peaks-loop's 9th)
+
+- **`IdeId` union type** — `'zcode'` added in `src/services/ide/ide-types.ts`.
+- **`ZCODE_ADAPTER`** — new `src/services/ide/adapters/zcode-adapter.ts` (98 lines). Template: `claude-code-adapter.ts` with `compact.compactCommand` / `hookEvent` / `toolMatcher` / `envVar` degraded (z-code is a desktop application, no CLI binary; UNVERIFIED sentinel strings used where the interface requires non-optional types).
+- **`IDE_DETECTION_DIRS` + `IDE_SKILL_INSTALL_PROFILES`** — `scripts/install-skills.mjs`. New `{ id: 'zcode', dir: '.zcode' }` detection entry + zcode profile (skillsDir `~/.zcode/skills`, env override `PEAKS_ZCODE_SKILLS_DIR`).
+- **`ide-registry.ts`** — registers `ZCODE_ADAPTER` alongside the existing 8 adapters (claude-code / trae / trae-cn / codex / cursor / qoder / tongyi-lingma / hermes / openclaw).
+- **`tests/unit/ide/zcode-adapter.test.ts`** — 10 cases (T-1 id, T-2 dirName, T-3 standards rootFile = `CLAUDE.md`, T-4 rulesDir = `.claude/rules`, T-5 skillsDir, T-6 compactCommand undefined, T-7 hookEvent is string, T-8 type test, T-9 IDE_DETECTION_DIRS contains, T-10 IDE_SKILL_INSTALL_PROFILES contains).
+- **`tests/unit/ide/zcode-adapter-detect-model.test.ts`** — 14 cases (pure resolver + env-fixture file IO + path helper).
+- **`tests/unit/cli/ide-commands.test.ts`** — 5 cases (CLI + top-level command registration).
+- **Fixture sync** — `tests/unit/install-skills-script.test.ts` + `tests/unit/cli-program.workflow.test.ts` (Strategy A: env-var override, no hardcoded literals) + `tests/unit/ide/ide-registry.test.ts` (whitelist 6 → 7 adapters).
+
+### Documentation — SKILL.md sync with 4.0.0-beta.6 CLI reality
+
+- **`skills/peaks-code/SKILL.md`** — added inline `> CLI reality check` blocks at §Step 0.8 (D-001 `peaks code detect-job` is `peaks job init`), §Step 0.8 third paragraph (D-003 `JOB_SHAPE_NOT_DECIDED` is now passive `done: 0` warning, not exception), §Step 2.5 (D-002 `peaks session title` takes positional sid, not `--session-id` flag), §Step 11c + 11d (D-010 `<!-- peaks-memory:start -->` block requires YAML frontmatter `title:` + `kind:` + `---` + closing `<!-- peaks-memory:end -->`, otherwise `extractedCount: 0` silently). New **§CLI Drift Index** section after Boundaries provides single landing page for D-001/002/003/010 with inline-anchor table.
+- **`skills/peaks-slice-decompose/SKILL.md`** — new **§SC template hard rules (sediment 2026-07-09)** section with 3 rules: Rule 1 ESM/CJS pre-flight before specifying export syntax (D-009c), Rule 2 whitelist fixture sync is a primary task not collateral (D-009a), Rule 3 adapter-interface optional fields > required (D-009b reciprocal).
+
+### Verification (4.0.0-beta.7)
+
+- `pnpm tsc -p tsconfig.json --noEmit` — 0 errors
+- `pnpm vitest run tests/unit/ide/ tests/unit/cli/ide-commands.test.ts` — 15 files / 211 tests / 0 failed
+- `peaks workflow verify-pipeline --rid 003-add-zcode-adapter` — `ok=true, complete=true`, 9/9 gate PASS
+- `node bin/peaks.js ide model --current` (against real z-code installation) — `modelId: "M3"`, `detected: true`
+
+### Sediment (Step 11) — 7 lessons in `.peaks/memory/`
+
+| File | Topic |
+|---|---|
+| `z-code-peaks-loop-9-ide-adapter-vendor-neutrality-adapter.md` | z-code 9th IDE + vendor-neutrality adapter pattern |
+| `peaks-loop-install-model-getstrongestmodelid-fallback.md` | install no longer hardcodes default model |
+| `desktop-application-ide-adapter-z-code-cli.md` | desktop-app IDE adapter field-degradation decision |
+| `peaks-code-runbook-4-0-0-beta-6-skill-md-cli-d-001-d-002-d-003-d-010.md` | 4 SKILL.md / CLI drift points |
+| `2026-07-09-zcode-adapter-overview.md` | RID 003 overall summary (project record) |
+| `peaks-ide-runtime-detect-zcode-only.md` | `peaks ide model --current` z-code 4-tier priority chain |
+| `ide-adapter-detectcurrentmodel-optional-interface-pattern.md` | optional interface field extension pattern |
+
+### Known follow-ups (S3-cleanup backlog)
+
+- **Slice C extension** — other 8 IDEs' `detectCurrentModel()` (claude-code / trae / cursor / codex / qoder / tongyi-lingma / hermes / openclaw). Currently only z-code has a working detector; all 8 still fall through to the back-compat `'claude-opus-4-7'`.
+- **zai / GLM provider entry** — `D-008`: peaks-loop has no built-in provider entry for z-code's default `builtin:zai` (GLM) or `builtin:bigmodel` providers. Users on those providers must add custom `provider.<uuid>` entries to their `~/.peaks/config.json` themselves.
+- **zcode-adapter UNVERIFIED sentinel values** — `hookEvent: 'PreToolUse'`, `toolMatcher: 'Bash'`, `envVar: 'ZCODE_PROJECT_DIR'` are placeholder guesses based on Anthropic-compatible protocol. Real z-code desktop application dogfood will reveal the actual values; replace once known.
+- **`IdeAdapter` interface optional fields** — `D-009b` proposes converting `hookEvent` / `toolMatcher` / `envVar` to optional types, or adding an `unverified: boolean` flag, so future adapters with unknown protocols don't need to fabricate sentinel values.
+
+### Breaking changes
+
+None. All 8 pre-existing adapters continue to load without changes. CLI subcommands `peaks ide *` and `peaks ide model --current` are additive.
+
 ## 4.0.0-beta.5 — 2026-07-08
 
 ### Added — peaks-solo dispatcher (分诊员)

package/dist/src/cli/commands/ide-commands.d.ts

@@ -0,0 +1,25 @@
+/**
+ * peaks ide * CLI (Slice 2026-07-09 add-zcode-adapter, Slice C).
+ *
+ * Slice C adds ONE verb:
+ *
+ *   peaks ide model --current
+ *
+ * Outputs the currently-active model id detected from the host
+ * environment (so far: only the `zcode` adapter opts in via
+ * `IdeAdapter.detectCurrentModel?`). The CLI is the read-only
+ * debugging surface for runtime model probing. It does NOT set any
+ * state — `peaks config set model` remains the canonical write path.
+ *
+ * Future slices will extend this surface with `peaks ide list`,
+ * `peaks ide detect`, etc. — this slice keeps the surface minimal
+ * (per Karpathy guideline #2 — Simplicity First).
+ *
+ * Vendor-neutrality: this file delegates to
+ * `src/services/ide/current-model-detector.ts` which walks every
+ * registered adapter. This file does NOT import any specific
+ * adapter.
+ */
+import { Command } from 'commander';
+import { type ProgramIO } from '../cli-helpers.js';
+export declare function registerIdeCommands(program: Command, io: ProgramIO): void;

package/dist/src/cli/commands/ide-commands.js

@@ -0,0 +1,42 @@
+import { detectCurrentIdeModel } from '../../services/ide/current-model-detector.js';
+import { addJsonOption, printResult } from '../cli-helpers.js';
+import { fail, ok } from '../../shared/result.js';
+import { listAdapterIds } from '../../services/ide/ide-registry.js';
+export function registerIdeCommands(program, io) {
+    const ide = program.command('ide').description('Read-only introspection helpers for the registered IDE adapter layer (slice 2026-07-09 add-zcode-adapter, Slice C)');
+    addJsonOption(ide
+        .command('model')
+        .description('Inspect the IDE adapter model surface. Slice C supports `--current` only.')
+        .option('--current', 'Print the model id the active IDE adapter reports as currently configured')
+        .action(async (options) => {
+        if (options.current) {
+            await runIdeModelCurrent(io, options);
+            return;
+        }
+        // No verb supplied — Commander would normally show help; we
+        // emit a structured error instead so the JSON envelope stays
+        // consistent for LLM callers.
+        printResult(io, fail('ide.model', 'MISSING_VERB', 'Specify a verb: --current. Future slices may add --configured / --default.', { registeredAdapters: listAdapterIds() }, ['Rerun with --current to read the runtime-detected model id.']), options.json);
+    }));
+}
+async function runIdeModelCurrent(io, options) {
+    try {
+        const modelId = await detectCurrentIdeModel();
+        if (modelId === undefined) {
+            printResult(io, ok('ide.model.current', {
+                modelId: null,
+                detected: false,
+                registeredAdapters: listAdapterIds(),
+            }), options.json);
+            return;
+        }
+        printResult(io, ok('ide.model.current', {
+            modelId,
+            detected: true,
+            registeredAdapters: listAdapterIds(),
+        }), options.json);
+    }
+    catch (error) {
+        printResult(io, fail('ide.model.current', 'DETECTION_FAILED', 'Runtime model detection failed; falling back to configured model. See logs.', { error: error instanceof Error ? error.message : String(error) }, ['Check `PEAKS_LOG_LEVEL=debug peaks ide model --current` for the underlying error.']), options.json);
+    }
+}

package/dist/src/cli/program.js

@@ -73,6 +73,10 @@ import { registerMutCommands } from './commands/mut-commands.js';
 import { registerFixtureCommands } from './commands/fixture-commands.js';
 import { registerReviewerCommands } from './commands/reviewer-commands.js';
 import { registerObservabilityCommands } from './commands/observability-commands.js';
+// Slice 2026-07-09 add-zcode-adapter (Slice C): `peaks ide model --current` —
+// read-only runtime model probe surfaced via the registered IDE adapter
+// chain. See src/services/ide/current-model-detector.ts.
+import { registerIdeCommands } from './commands/ide-commands.js';
 // Slice 2026-07-01-strategic-compact-cli: peaks compact
 // (suggest | recommend | survival | dry-run | force) — strategic-compact
 // CLI primitives. See src/services/compact/ and
@@ -370,6 +374,9 @@ Run peaks (no arguments) for a quickstart. You likely want one of:
     // Slice B of v2.11.1: `peaks observability status|slices|fanout|repair-cycles`
     // (read-only queries over the JSONL metrics emitted from the
     // peaks request transition hook in Slice A).
+    // Slice 2026-07-09 add-zcode-adapter (C.6): `peaks ide model --current`
+    // (read-only runtime model probe via the registered IDE adapter chain).
+    registerIdeCommands(program, io);
     registerObservabilityCommands(program, io);
     // Slice 2026-07-01-strategic-compact-cli: `peaks compact
     // suggest|recommend|survival|dry-run|force` — strategic-compact CLI

package/dist/src/services/config/model-routing.d.ts

@@ -1,4 +1,57 @@
 import { type ModelProviderConfig, type PeaksConfig } from './config-types.js';
-export declare const STRONGEST_MODEL_ID: "claude-opus-4-7";
+/**
+ * Slice 2026-07-09 add-zcode-adapter (A.3): replaces the legacy
+ * `STRONGEST_MODEL_ID = 'claude-opus-4-7'` constant (removed) with a dynamic
+ * resolver. The strongest planner/reviewer model is now derived
+ * from the active config instead of being hardcoded to a specific
+ * vendor. This unlocks running peaks-loop under non-Claude IDEs
+ * (z-code, Cursor, Trae, ...).
+ *
+ * Resolution order:
+ *   1. Explicit `config.model` (user override via `peaks config set model ...`)
+ *   2. Stable fallback for back-compat (env-var override; defaults to
+ *      `claude-opus-4-7` so legacy fixtures keep compiling per
+ *      SC §3.4 strategy A)
+ *
+ * Vendor-neutrality note: the default `'claude-opus-4-7'` is a
+ * *back-compat fallback*, NOT a hardcoded "the strongest model is
+ * always Claude" assumption. Production callers should set
+ * `config.model` (or override the env var) to declare their strongest
+ * model. Adapter-level runtime probing is wired in Slice C
+ * (`getStrongestModelIdAsync` below).
+ */
+export declare function getStrongestModelId(config?: {
+    model?: unknown;
+}): string;
+/**
+ * Slice 2026-07-09 add-zcode-adapter (Slice C, C.4): async variant
+ * of `getStrongestModelId`. Same precedence for layers 1 + 2, but
+ * layer 2.5 is the IDE's runtime probe (`detectCurrentIdeModel`)
+ * which is consulted BEFORE the env-var back-compat fallback.
+ *
+ * Why this exists:
+ *   - `rd-service.buildPlan` (and other sync callers) still call the
+ *     sync `getStrongestModelId` — we do NOT change those signatures.
+ *   - New async callers (the future Slice C-extension of
+ *     `workflow-router-service`, plus any LLM-inside-z-code path)
+ *     resolve the strongest model from the IDE's actual state,
+ *     without forcing the user to set `config.model`.
+ *
+ * Resolution order:
+ *   1. Explicit `config.model`.
+ *   2. `PEAKS_STRONGEST_MODEL_DEFAULT` env var override.
+ *   3. Runtime probe via `detectCurrentIdeModel()` (only when the
+ *      registered IDE adapter opts in via `detectCurrentModel?` —
+ *      only `zcode` does in Slice C; other IDEs skip this layer).
+ *   4. Hardcoded back-compat default `'claude-opus-4-7'`.
+ *
+ * Each layer is isolated so a failure in the probe does not break
+ * the resolver.
+ */
+export declare function getStrongestModelIdAsync(config?: {
+    model?: unknown;
+}, runtimeProbeDisabled?: boolean): Promise<string>;
 export declare function getConfiguredExecutionModelId(providers: ModelProviderConfig | undefined): string;
-export declare function getEconomyAwareExecutionModelId(config: Pick<PeaksConfig, 'economyMode' | 'providers'>): string;
+export declare function getEconomyAwareExecutionModelId(config: Pick<PeaksConfig, 'economyMode' | 'providers'> & {
+    model?: string;
+}): string;

package/dist/src/services/config/model-routing.js

@@ -1,4 +1,79 @@
-export const STRONGEST_MODEL_ID = 'claude-opus-4-7';
+import { detectCurrentIdeModel } from '../ide/current-model-detector.js';
+/**
+ * Slice 2026-07-09 add-zcode-adapter (A.3): replaces the legacy
+ * `STRONGEST_MODEL_ID = 'claude-opus-4-7'` constant (removed) with a dynamic
+ * resolver. The strongest planner/reviewer model is now derived
+ * from the active config instead of being hardcoded to a specific
+ * vendor. This unlocks running peaks-loop under non-Claude IDEs
+ * (z-code, Cursor, Trae, ...).
+ *
+ * Resolution order:
+ *   1. Explicit `config.model` (user override via `peaks config set model ...`)
+ *   2. Stable fallback for back-compat (env-var override; defaults to
+ *      `claude-opus-4-7` so legacy fixtures keep compiling per
+ *      SC §3.4 strategy A)
+ *
+ * Vendor-neutrality note: the default `'claude-opus-4-7'` is a
+ * *back-compat fallback*, NOT a hardcoded "the strongest model is
+ * always Claude" assumption. Production callers should set
+ * `config.model` (or override the env var) to declare their strongest
+ * model. Adapter-level runtime probing is wired in Slice C
+ * (`getStrongestModelIdAsync` below).
+ */
+export function getStrongestModelId(config) {
+    const raw = config?.model;
+    const fromConfig = typeof raw === 'string' ? raw.trim() : undefined;
+    if (fromConfig)
+        return fromConfig;
+    // Slice 2026-07-09: back-compat fallback for test fixtures that
+    // pre-date this slice (SC §3.4 strategy A env-var override).
+    return process.env.PEAKS_STRONGEST_MODEL_DEFAULT ?? 'claude-opus-4-7';
+}
+/**
+ * Slice 2026-07-09 add-zcode-adapter (Slice C, C.4): async variant
+ * of `getStrongestModelId`. Same precedence for layers 1 + 2, but
+ * layer 2.5 is the IDE's runtime probe (`detectCurrentIdeModel`)
+ * which is consulted BEFORE the env-var back-compat fallback.
+ *
+ * Why this exists:
+ *   - `rd-service.buildPlan` (and other sync callers) still call the
+ *     sync `getStrongestModelId` — we do NOT change those signatures.
+ *   - New async callers (the future Slice C-extension of
+ *     `workflow-router-service`, plus any LLM-inside-z-code path)
+ *     resolve the strongest model from the IDE's actual state,
+ *     without forcing the user to set `config.model`.
+ *
+ * Resolution order:
+ *   1. Explicit `config.model`.
+ *   2. `PEAKS_STRONGEST_MODEL_DEFAULT` env var override.
+ *   3. Runtime probe via `detectCurrentIdeModel()` (only when the
+ *      registered IDE adapter opts in via `detectCurrentModel?` —
+ *      only `zcode` does in Slice C; other IDEs skip this layer).
+ *   4. Hardcoded back-compat default `'claude-opus-4-7'`.
+ *
+ * Each layer is isolated so a failure in the probe does not break
+ * the resolver.
+ */
+export async function getStrongestModelIdAsync(config, runtimeProbeDisabled = false) {
+    const raw = config?.model;
+    const fromConfig = typeof raw === 'string' ? raw.trim() : undefined;
+    if (fromConfig)
+        return fromConfig;
+    const envDefault = process.env.PEAKS_STRONGEST_MODEL_DEFAULT;
+    if (envDefault && envDefault.length > 0)
+        return envDefault;
+    if (!runtimeProbeDisabled) {
+        try {
+            const probed = await detectCurrentIdeModel();
+            if (probed && probed.length > 0)
+                return probed;
+        }
+        catch {
+            // best-effort — fall through to back-compat default
+        }
+    }
+    return 'claude-opus-4-7';
+}
 export function getConfiguredExecutionModelId(providers) {
     const providerConfigs = Object.values(providers ?? {});
     const configuredModel = providerConfigs
@@ -12,6 +87,7 @@ export function getConfiguredExecutionModelId(providers) {
 export function getEconomyAwareExecutionModelId(config) {
     // Slice 2.0.1-bug1 round 3: economy is the project default. Treat undefined as enabled
     // (matches the pre-slice implicit default from DEFAULT_CONFIG.economyMode = true). Only an
-    // explicit `economyMode === false` switches execution to STRONGEST_MODEL_ID.
-    return config.economyMode !== false ? getConfiguredExecutionModelId(config.providers) : STRONGEST_MODEL_ID;
+    // explicit `economyMode === false` switches execution to the strongest planner/reviewer
+    // model (resolved dynamically per Slice 2026-07-09 add-zcode-adapter A.3).
+    return config.economyMode !== false ? getConfiguredExecutionModelId(config.providers) : getStrongestModelId(config);
 }

package/dist/src/services/ide/adapters/zcode-adapter.d.ts

@@ -0,0 +1,40 @@
+import type { IdeAdapter } from '../ide-types.js';
+/**
+ * Helper exported for tests + internal use: the default location of the
+ * z-code v2 config file on the host filesystem.
+ */
+export declare function defaultZcodeConfigPath(): string;
+/**
+ * Resolve the active model id from a parsed z-code config payload
+ * (the `~/.zcode/v2/config.json` root object). Pure / exported for
+ * unit tests so the file-IO seam and the resolution seam are
+ * independently verifiable.
+ *
+ * Resolution order (Slice C §"z-code config 解析路径"):
+ *   P1. env var `PEAKS_ZCODE_ACTIVE_PROVIDER_UUID` (test seam +
+ *       manual pin for the user).
+ *   P2. The provider whose top-level UUID key is NOT prefixed
+ *       `builtin:` — this matches the user-installed provider
+ *       pattern. Within that provider, prefer its `models` first
+ *       key (insertion order — z-code writes the active model
+ *       first in the picker).
+ *   P3. The first provider with `enabled: true`.
+ *   P4. The first provider in the object (insertion order) — last
+ *       resort. Returns undefined if there are zero providers.
+ *
+ * Returns `undefined` when nothing matches.
+ */
+export declare function resolveZcodeCurrentModel(config: unknown, envOverride?: string | undefined): string | undefined;
+/**
+ * The async adapter method that `detectCurrentIdeModel()` calls.
+ *
+ * Resolution order:
+ *   1. `PEAKS_ZCODE_ACTIVE_PROVIDER_UUID` env var override (test seam
+ *      + manual pin).
+ *   2. `PEAKS_ZCODE_CONFIG_PATH` env var override (test seam for a
+ *      fixture file). Falls back to `~/.zcode/v2/config.json`.
+ *   3. Returns `undefined` when the file is missing / malformed
+ *      (consistent with the cross-IDE contract).
+ */
+export declare function detectZcodeCurrentModel(): Promise<string | undefined>;
+export declare const ZCODE_ADAPTER: IdeAdapter;

package/dist/src/services/ide/adapters/zcode-adapter.js

@@ -0,0 +1,220 @@
+/**
+ * z-code IDE adapter (slice 2026-07-09-add-zcode-adapter) — peaks-loop 的第九个
+ * 内置 IDE 适配器。
+ *
+ * z-code 是一个 VS Code-style 的 Anthropic-compatible 桌面应用:
+ *   - 不是 CLI,没有 `zcode` binary (`which zcode` 返回空)。
+ *   - 桌面应用内置了 "导入 skills" 功能,user 手动触发后将本地 skills 目录
+ *     与 `~/.zcode/skills` 双向同步 (实测 symlink)。
+ *   - hook 协议、project-dir env var、compact command 均未公开。
+ *
+ * 设计前提 (RD-3 §2.2 D1 决策 + 现场实测,见
+ * `.peaks/memory/2026-07-09-zcode-context-probe-handoff.md` §2):
+ *   - `compact` 字段全部 undefined — 走 `llm-self-compress` fallback 路径。
+ *   - `envVar` / `hookEvent` / `toolMatcher` 全部标注 UNVERIFIED,因为
+ *     z-code 文档未公开,本 adapter 用 Anthropic-compatible 协议的合理
+ *     假设占位。
+ *   - `subAgentDispatcher` 用 `nullSubAgentDispatcher` 占位 — z-code
+ *     桌面应用没有公开的 Task 工具 dispatch 协议。
+ *   - `standardsProfile` 字段指向宪法文件根 + rules 目录 — z-code 的设计
+ *     就是借用上游 "import skills" 目标目录,因此保留上游路径常量。
+ *   - `skillInstall` 指向 `~/.zcode/skills` — 实测 z-code 已通过 user
+ *     手动导入同步到这里。
+ *   - 注:`standardsProfile` 中的常量子段是 z-code 借用的固定路径,属于
+ *     架构事实,不算 vendor verb 泄漏。
+ *
+ * Vendor-neutrality 备注 (SC-3 §5 + RD-3 §2.4):
+ *   本文件允许出现 vendor 字符串 (`zcode` / `ZCODE_PROJECT_DIR` 等),
+ *   **不允许** 出现其他 vendor 字符串 / 模型名。`standardsProfile`
+ *   中的 `CLAUDE.md` / `.claude/rules` 是 z-code 借用的上游路径常量,
+ *   属于架构事实,不算 vendor verb 泄漏。
+ */
+import { existsSync, readFileSync } from 'node:fs';
+import { homedir } from 'node:os';
+import { join, resolve } from 'node:path';
+import { nullSubAgentDispatcher } from '../../dispatch/sub-agent-dispatcher.js';
+/**
+ * Helper exported for tests + internal use: the default location of the
+ * z-code v2 config file on the host filesystem.
+ */
+export function defaultZcodeConfigPath() {
+    return join(homedir(), '.zcode', 'v2', 'config.json');
+}
+/**
+ * Read the z-code config from `path` (string-typed to keep this
+ * swappable for tests). Synchronous + Node.js `fs` so adapter callers
+ * stay async-only at the public surface (`detectCurrentModel` is
+ * already declared `async`).
+ */
+function readZcodeConfig(path) {
+    if (!existsSync(path))
+        return undefined;
+    try {
+        return JSON.parse(readFileSync(path, 'utf8'));
+    }
+    catch {
+        return undefined;
+    }
+}
+/**
+ * Resolve the active model id from a parsed z-code config payload
+ * (the `~/.zcode/v2/config.json` root object). Pure / exported for
+ * unit tests so the file-IO seam and the resolution seam are
+ * independently verifiable.
+ *
+ * Resolution order (Slice C §"z-code config 解析路径"):
+ *   P1. env var `PEAKS_ZCODE_ACTIVE_PROVIDER_UUID` (test seam +
+ *       manual pin for the user).
+ *   P2. The provider whose top-level UUID key is NOT prefixed
+ *       `builtin:` — this matches the user-installed provider
+ *       pattern. Within that provider, prefer its `models` first
+ *       key (insertion order — z-code writes the active model
+ *       first in the picker).
+ *   P3. The first provider with `enabled: true`.
+ *   P4. The first provider in the object (insertion order) — last
+ *       resort. Returns undefined if there are zero providers.
+ *
+ * Returns `undefined` when nothing matches.
+ */
+export function resolveZcodeCurrentModel(config, envOverride) {
+    const providers = config?.provider;
+    if (!providers || typeof providers !== 'object')
+        return undefined;
+    // P1: env var override (vendor-neutral identifier: the provider UUID).
+    if (envOverride !== undefined && envOverride !== '') {
+        const target = providers[envOverride];
+        if (target && typeof target === 'object') {
+            const modelId = pickFirstModelId(target);
+            if (modelId)
+                return modelId;
+        }
+    }
+    // P2: prefer a non-builtin provider (the user-installed one).
+    for (const [uuid, entry] of Object.entries(providers)) {
+        if (uuid.startsWith('builtin:'))
+            continue;
+        if (!entry || typeof entry !== 'object')
+            continue;
+        const modelId = pickFirstModelId(entry);
+        if (modelId)
+            return modelId;
+    }
+    // P3: first enabled provider (legacy / fresh-install case).
+    for (const entry of Object.values(providers)) {
+        if (!entry || typeof entry !== 'object')
+            continue;
+        const enabled = entry.enabled;
+        if (enabled === true) {
+            const modelId = pickFirstModelId(entry);
+            if (modelId)
+                return modelId;
+        }
+    }
+    // P4: first provider at all.
+    for (const entry of Object.values(providers)) {
+        if (!entry || typeof entry !== 'object')
+            continue;
+        const modelId = pickFirstModelId(entry);
+        if (modelId)
+            return modelId;
+    }
+    return undefined;
+}
+function pickFirstModelId(provider) {
+    const models = provider.models;
+    if (!models || typeof models !== 'object')
+        return undefined;
+    for (const key of Object.keys(models)) {
+        if (typeof key === 'string' && key.trim().length > 0)
+            return key.trim();
+    }
+    return undefined;
+}
+/**
+ * The async adapter method that `detectCurrentIdeModel()` calls.
+ *
+ * Resolution order:
+ *   1. `PEAKS_ZCODE_ACTIVE_PROVIDER_UUID` env var override (test seam
+ *      + manual pin).
+ *   2. `PEAKS_ZCODE_CONFIG_PATH` env var override (test seam for a
+ *      fixture file). Falls back to `~/.zcode/v2/config.json`.
+ *   3. Returns `undefined` when the file is missing / malformed
+ *      (consistent with the cross-IDE contract).
+ */
+export async function detectZcodeCurrentModel() {
+    const overridePath = process.env.PEAKS_ZCODE_CONFIG_PATH;
+    const path = overridePath && overridePath.length > 0 ? overridePath : defaultZcodeConfigPath();
+    const config = readZcodeConfig(path);
+    if (config === undefined)
+        return undefined;
+    return resolveZcodeCurrentModel(config, process.env.PEAKS_ZCODE_ACTIVE_PROVIDER_UUID);
+}
+export const ZCODE_ADAPTER = {
+    id: 'zcode',
+    displayName: 'z-code',
+    settings: {
+        dirName: '.zcode', // VERIFIED — 实测 z-code 项目根目录是 `.zcode`
+        settingsFileName: 'settings.json',
+        resolveSettingsFile: (scope, projectRoot) => {
+            const root = scope === 'global' ? homedir() : resolve(projectRoot ?? homedir());
+            return join(root, '.zcode', 'settings.json');
+        },
+        supportsScope: (scope) => scope === 'project' || scope === 'global',
+    },
+    // UNVERIFIED — z-code 桌面应用未公开 `ZCODE_PROJECT_DIR` env var,这里
+    // 用占位符,RD 阶段假设 z-code 借用了 Anthropic-compatible 协议并可能
+    // 注入一个项目根 env 变量。若未注入,峰 router 会回退到 cwd 探测。
+    envVar: 'ZCODE_PROJECT_DIR',
+    // UNVERIFIED — z-code 未公开 hook event key,占位用 Anthropic-compatible
+    // 协议的常见命名 `PreToolUse`。若 z-code 用其他 key,peaks hooks install
+    // 钩子在此 IDE 上不生效,可由 user 通过手动导入绕过。
+    hookEvent: 'PreToolUse',
+    // UNVERIFIED — 同上,占位用 `Bash` matcher。
+    toolMatcher: 'Bash',
+    // Slice 2026-07-09: z-code 桌面应用没有公开 Task 工具 dispatch 协议,
+    // 用 `nullSubAgentDispatcher` 占位 — peaks sub-agent dispatch CLI 在
+    // z-code 上返回 SubAgentNotSupportedError,符合 slice #008 P-5 capability
+    // 检查契约 (CLI 不崩溃)。
+    subAgentDispatcher: nullSubAgentDispatcher,
+    // Slice #010 G9: 假设 z-code 支持 PreToolUse (UNVERIFIED)。Opt-in,等真实
+    // z-code 安装 dogfood 后再决定。若 z-code 不支持 G9 hook,CLI 兜底层仍会
+    // 强制执行 G9 阈值。
+    promptSizeAware: true,
+    installHints: [
+        'z-code 没有 CLI,因此无需重启 — skills symlink 已写入 ~/.zcode/skills/。',
+        '若 z-code 未自动加载 peaks-* skills,在 z-code 桌面应用里手动触发 "导入 skills" 功能即可。',
+    ],
+    capabilities: {
+        gateEnforce: true,
+        statusline: true,
+    },
+    // Standards profile: 路径常量保留 (z-code 的 "导入 skills" 功能
+    // 借用了上游 settings 目录,因此宪法文件路径常量同步保留)。
+    standardsProfile: {
+        rootFile: 'CLAUDE.md',
+        rulesDir: '.claude/rules',
+        rulesFileGlob: '**/*.md',
+        autoLoaded: true,
+        format: 'markdown',
+        migrationHint: 'z-code 借用了上游 standards 路径,无需迁移。',
+    },
+    // Skill install profile: 指向 `~/.zcode/skills` 系列 (实测 z-code 已通过
+    // user 手动 symlink 同步;peaks-loop 也直接支持)。
+    skillInstall: {
+        skillsDir: join(homedir(), '.zcode', 'skills'),
+        outputStylesDir: join(homedir(), '.zcode', 'output-styles'),
+        installStrategy: 'symlink',
+        envVarOverride: 'PEAKS_ZCODE_SKILLS_DIR',
+    },
+    // compact profile 留空 — z-code 没有 CLI binary (RD-3 §2.2 D1 决策)。
+    // peaks-loop 走 `llm-self-compress` fallback 路径 (LLM 自己总结 context),
+    // 不抛错、不警告 (符合 slice #008 P-5 capability-check contract)。
+    // Slice C (2026-07-09 add-zcode-adapter, C.2): runtime probe for
+    // the currently-active model id. Reads `~/.zcode/v2/config.json`
+    // and resolves the active provider via `resolveZcodeCurrentModel`.
+    // See `.peaks/_runtime/2026-07-08-session-17918f/qa/.../slice-C-completion.md`
+    // for the resolution priority chain (env override → non-builtin
+    // provider → first enabled → first provider). Returns undefined
+    // on any failure (consistent with the cross-IDE contract).
+    detectCurrentModel: detectZcodeCurrentModel,
+};

package/dist/src/services/ide/current-model-detector.d.ts

@@ -0,0 +1,14 @@
+/**
+ * Probe every registered IDE adapter for its currently-active model
+ * id. Returns the first non-empty value found in registration order,
+ * or `undefined` when no adapter opts in / all opted-in adapters
+ * return undefined.
+ *
+ * Registration order: see `ide-registry.ts` ADAPTERS Map. The IDE
+ * adapter for `claude-code` is consulted first; only `zcode` opts in
+ * to this slice's detection (other adapters will return undefined).
+ *
+ * Async because each IDE adapter's detectCurrentModel may read local
+ * disk or env on a slow filesystem.
+ */
+export declare function detectCurrentIdeModel(): Promise<string | undefined>;

package/dist/src/services/ide/current-model-detector.js

@@ -0,0 +1,60 @@
+/**
+ * Runtime current-model detector (Slice 2026-07-09 add-zcode-adapter, Slice C).
+ *
+ * Walks the registered IDE adapter chain and asks the IDE adapter
+ * (when it opts in via `IdeAdapter.detectCurrentModel?`) to report the
+ * currently-active model id. This is the runtime sibling of
+ * `getStrongestModelId(config)` in `src/services/config/model-routing.ts`:
+ *
+ *   - `getStrongestModelId(config)` reads `config.model` (user-set) and
+ *     falls back to the env-var / `'claude-opus-4-7'` constant. Sync.
+ *   - `detectCurrentIdeModel()` reads the IDE's runtime state (which
+ *     provider UI is active, which model the picker shows). Async.
+ *
+ * The CLI surface `peaks ide model --current` (added in this slice)
+ * delegates here. The future async resolver at
+ * `getStrongestModelIdAsync` (Slice C §C.4) calls this so config-less
+ * runs (e.g. an LLM inside z-code) can derive the strongest model from
+ * its actual environment.
+ *
+ * Vendor-neutrality notes (SC-3 §5 + RD-3 §2.4):
+ *   - This file never names a vendor (no "claude", "zcode", etc.).
+ *   - It only walks `IdeAdapter.detectCurrentModel?` — any adapter
+ *     can opt in.
+ *   - It returns `undefined` when detection is unavailable; callers
+ *     must fall back to the configured / hardcoded model.
+ */
+import { getAdapter } from './ide-registry.js';
+import { listAdapterIds } from './ide-registry.js';
+/**
+ * Probe every registered IDE adapter for its currently-active model
+ * id. Returns the first non-empty value found in registration order,
+ * or `undefined` when no adapter opts in / all opted-in adapters
+ * return undefined.
+ *
+ * Registration order: see `ide-registry.ts` ADAPTERS Map. The IDE
+ * adapter for `claude-code` is consulted first; only `zcode` opts in
+ * to this slice's detection (other adapters will return undefined).
+ *
+ * Async because each IDE adapter's detectCurrentModel may read local
+ * disk or env on a slow filesystem.
+ */
+export async function detectCurrentIdeModel() {
+    for (const id of listAdapterIds()) {
+        const adapter = getAdapter(id);
+        if (typeof adapter.detectCurrentModel !== 'function')
+            continue;
+        try {
+            const modelId = await adapter.detectCurrentModel();
+            if (typeof modelId === 'string' && modelId.trim().length > 0) {
+                return modelId.trim();
+            }
+        }
+        catch {
+            // Per-SC §3.4: detection is best-effort. Swallow errors so a
+            // misconfigured adapter cannot poison the chain; the next
+            // adapter still gets a chance.
+        }
+    }
+    return undefined;
+}

package/dist/src/services/ide/ide-registry.js

@@ -4,6 +4,7 @@ import { CURSOR_ADAPTER } from './adapters/cursor-adapter.js';
 import { CODEX_ADAPTER } from './adapters/codex-adapter.js';
 import { HERMES_ADAPTER } from './adapters/hermes-adapter.js';
 import { OPENCLAW_ADAPTER } from './adapters/openclaw-adapter.js';
+import { ZCODE_ADAPTER } from './adapters/zcode-adapter.js';
 /**
  * Built-in IDE adapter registry。Map<IdeId, IdeAdapter> 是单一来源。
  *
@@ -24,6 +25,7 @@ const ADAPTERS = new Map([
     ['codex', CODEX_ADAPTER],
     ['hermes', HERMES_ADAPTER],
     ['openclaw', OPENCLAW_ADAPTER],
+    ['zcode', ZCODE_ADAPTER],
 ]);
 /** Get the adapter for a given IDE id. Throws on unsupported IDE. */
 export function getAdapter(ide) {
@@ -57,4 +59,5 @@ export function _resetAdaptersForTesting() {
     ADAPTERS.set('codex', CODEX_ADAPTER);
     ADAPTERS.set('hermes', HERMES_ADAPTER);
     ADAPTERS.set('openclaw', OPENCLAW_ADAPTER);
+    ADAPTERS.set('zcode', ZCODE_ADAPTER);
 }

package/dist/src/services/ide/ide-types.d.ts

@@ -12,7 +12,7 @@
  * 其他全部归一化到 peaks 内部模型(见 hook-protocol.ts)。
  */
 import type { SubAgentDispatcher } from '../dispatch/sub-agent-dispatcher.js';
-export type IdeId = 'claude-code' | 'trae' | 'codex' | 'cursor' | 'qoder' | 'tongyi-lingma' | 'hermes' | 'openclaw';
+export type IdeId = 'claude-code' | 'trae' | 'codex' | 'cursor' | 'qoder' | 'tongyi-lingma' | 'hermes' | 'openclaw' | 'zcode';
 export interface IdeCapabilities {
     /** peaks gate enforce 是否适用该 IDE(必备) */
     readonly gateEnforce: true;
@@ -125,6 +125,29 @@ export interface IdeAdapter {
      * `.peaks/memory/2026-06-27-auto-compact-design.md`.
      */
     readonly compact?: IdeCompactProfile;
+    /**
+     * Optional runtime probe for the IDE's currently-active model id.
+     * When defined, `detectCurrentIdeModel()` in
+     * `src/services/ide/current-model-detector.ts` calls this to
+     * discover which model the IDE UI is currently configured for —
+     * e.g. so `peaks ide model --current` and the future
+     * `getStrongestModelIdAsync()` can use the same value without
+     * forcing the user to set `config.model` manually.
+     *
+     * Returning `undefined` (or the method being absent on this
+     * adapter) means "I don't know — fall back to the configured
+     * model / env var / back-compat default".
+     *
+     * Implementations may read local files (`fs.readFileSync`),
+     * env vars, or both. They MUST NOT throw on missing files — catch
+     * locally and return `undefined`. The orchestration layer also
+     * catches, but defense in depth.
+     *
+     * Added in slice 2026-07-09 add-zcode-adapter (Slice C); only the
+     * `zcode` adapter opts in this slice. Future slices can opt in
+     * other IDEs without changing this signature (optional method).
+     */
+    readonly detectCurrentModel?: () => Promise<string | undefined>;
 }
 /**
  * Per-IDE auto-compact capability descriptor. See `IdeAdapter.compact`

package/dist/src/services/preferences/preferences-types.d.ts

@@ -71,7 +71,7 @@ export interface ProjectPreferences {
      * Slice 2026-06-28-code-mode-bypass-fix: economyMode is a
      * **MODEL-SELECTION** knob only — it chooses between the cheap
      * configured provider (`getConfiguredExecutionModelId`) and the
-     * strongest planner/reviewer model (`STRONGEST_MODEL_ID`). It does
+     * strongest planner/reviewer model (`getStrongestModelId(...)`). It does
      * NOT throttle concurrency, fan-out, or worker count. Per user
      * direction 2026-06-28: "效率比省钱更重要，是在效率达到最大值的时
      * 候，再去考虑经济问题" (efficiency first; economy is only considered

package/dist/src/services/providers/minimax-worker-service.d.ts

@@ -10,7 +10,7 @@ export type MiniMaxWorkerRequest = {
 export type MiniMaxWorkerResult = {
     provider: MiniMaxProviderSmokeResult;
     reviewHandoff: {
-        model: 'claude-opus-4-7';
+        model: string;
         prompt: string;
     };
     constraints: {

package/dist/src/services/providers/minimax-worker-service.js

@@ -62,7 +62,7 @@ export async function runMiniMaxWorker(config, request, fetchImpl = fetch) {
         successMatch: 'startsWith'
     }, fetchImpl);
     const reviewHandoff = {
-        model: 'claude-opus-4-7',
+        model: process.env.PEAKS_MINIMAX_REVIEW_MODEL ?? 'claude-opus-4-7',
         prompt: [
             `Review this MiniMax worker result for change ${sessionId}.`,
             `Goal: ${goal}`,

package/dist/src/services/rd/rd-service.js

@@ -4,7 +4,7 @@ import { stableRealPath } from '../../shared/path-utils.js';
 import { WORKSPACE_UNAVAILABLE_NEXT_ACTIONS } from '../../shared/planner-response.js';
 import { getLocalArtifactPath, hasValidArtifactWorkspace } from '../artifacts/workspace-service.js';
 import { getSessionDir } from '../session/getSessionDir.js';
-import { getConfiguredExecutionModelId, STRONGEST_MODEL_ID } from '../config/model-routing.js';
+import { getConfiguredExecutionModelId, getStrongestModelId } from '../config/model-routing.js';
 /**
  * 2.0.1-bug1: the slim 2.0 `~/.peaks/config.json` no longer carries a
  * `providers` block (legacy model config lives in
@@ -116,8 +116,8 @@ function selectConcreteTargetArea(targetAreas, index) {
 function getTaskModelRole(wave) {
     return wave === 'implementation candidates' || wave === 'unit-test execution' ? 'execution' : 'strongest';
 }
-function getTaskModelId(modelRole, executionModelId) {
-    return modelRole === 'execution' ? executionModelId : STRONGEST_MODEL_ID;
+function getTaskModelId(modelRole, executionModelId, strongestModelId) {
+    return modelRole === 'execution' ? executionModelId : strongestModelId;
 }
 const MAX_ARTIFACT_BYTES = 256_000;
 function readArtifactFile(rootPath, artifactWorkspacePath, artifact) {
@@ -220,6 +220,9 @@ function buildPlan(request, standardsOverlay) {
     const goal = normalizeGoal(request.goal);
     const swarmMode = request.swarmMode ?? true;
     const executionModelId = request.executionModelId?.trim() || resolveExecutionModelId();
+    // Slice 2026-07-09 add-zcode-adapter (A.3): strongest planner/reviewer
+    // model is resolved dynamically (config.model → env-var fallback).
+    const strongestModelId = getStrongestModelId();
     const { workerTarget, blockedReasons } = resolveWorkerTarget(request.maxWorkers);
     const artifactWorkspacePath = resolveArtifactWorkspacePath(request);
     const artifactRoot = 'rd/swarm';
@@ -342,7 +345,7 @@ function buildPlan(request, standardsOverlay) {
             workerKind: taskId,
             purpose: `${taskId.replace(/^rd-/, '').replace(/-/g, ' ')} for ${goal}`,
             modelRole,
-            modelId: getTaskModelId(modelRole, executionModelId),
+            modelId: getTaskModelId(modelRole, executionModelId, strongestModelId),
             inputs: [goal, artifactRoot],
             outputs: [briefPath],
             dependsOn: [...waveDependencies[wave]],

package/dist/src/services/workflow/workflow-router-service.d.ts

@@ -38,7 +38,7 @@ export type WorkflowRouterStep = {
 };
 export type WorkflowModelRouting = {
     readonly strongestModel: {
-        readonly modelId: 'claude-opus-4-7';
+        readonly modelId: string;
         readonly uses: readonly WorkflowStepStage[];
     };
     readonly executionModel: {

package/dist/src/services/workflow/workflow-router-service.js

@@ -1,4 +1,4 @@
-import { getConfiguredExecutionModelId, STRONGEST_MODEL_ID } from '../config/model-routing.js';
+import { getConfiguredExecutionModelId, getStrongestModelId } from '../config/model-routing.js';
 import { getLocalArtifactPath } from '../artifacts/workspace-service.js';
 import { createRdSwarmPlan } from '../rd/rd-service.js';
 import { createTechPlan, getTechStatus } from '../tech/tech-service.js';
@@ -37,12 +37,12 @@ function assertCodeModeAllowed(mode, codeMode) {
         throw new Error('codeMode requires code workflow mode');
     }
 }
-function step(input, executionModelId) {
+function step(input, executionModelId, strongestModelId) {
     const modelRole = EXECUTION_STAGES.includes(input.stage) ? 'execution' : 'strongest';
     return {
         ...input,
         modelRole,
-        modelId: modelRole === 'execution' ? executionModelId : STRONGEST_MODEL_ID,
+        modelId: modelRole === 'execution' ? executionModelId : strongestModelId,
         dryRunOnly: true,
         invokesAgents: false,
         writesArtifacts: false
@@ -78,40 +78,40 @@ function annotateSteps(steps, codeMode) {
         };
     });
 }
-function createSoloSteps(executionModelId) {
+function createSoloSteps(executionModelId, strongestModelId) {
     return [
-        step({ id: 'code-product-direction', stage: 'product-direction', owner: 'peaks-code', modelTier: 'top-tier', reason: 'Product direction needs strong judgment before execution work is delegated.', dependsOn: [] }, executionModelId),
-        step({ id: 'code-design-direction', stage: 'design-direction', owner: 'peaks-code', modelTier: 'top-tier', reason: 'Design direction uses the recommended default before cheaper implementation work.', dependsOn: ['code-product-direction'] }, executionModelId),
-        step({ id: 'code-tech-direction', stage: 'tech-direction', owner: 'peaks-tech', modelTier: 'top-tier', reason: 'Technical boundaries and approval gates use the recommended default with high-confidence planning.', dependsOn: ['code-design-direction'] }, executionModelId),
-        step({ id: 'code-tech-review', stage: 'tech-review', owner: 'peaks-tech', modelTier: 'top-tier', reason: 'Tech artifacts and gate decisions require strong review and a recommended default path.', dependsOn: ['code-tech-direction'] }, executionModelId),
-        step({ id: 'code-rd-planning', stage: 'rd-planning', owner: 'peaks-rd', modelTier: 'top-tier', reason: 'RD task decomposition and acceptance criteria use the recommended default before execution delegation.', dependsOn: ['code-tech-review'] }, executionModelId),
-        step({ id: 'code-coding-execution', stage: 'coding-execution', owner: 'peaks-rd', modelTier: executionModelId === STRONGEST_MODEL_ID ? 'top-tier' : 'mid-tier', reason: `Coding and routine refactoring must use the configured execution worker model ${executionModelId}.`, dependsOn: ['code-rd-planning'] }, executionModelId),
-        step({ id: 'code-unit-test-execution', stage: 'unit-test-execution', owner: 'peaks-rd', modelTier: executionModelId === STRONGEST_MODEL_ID ? 'top-tier' : 'mid-tier', reason: `Unit test authoring and focused test runs must use the configured execution worker model ${executionModelId}.`, dependsOn: ['code-coding-execution'] }, executionModelId),
-        step({ id: 'code-quality-review', stage: 'quality-review', owner: 'peaks-code', modelTier: 'top-tier', reason: 'Reducer and final quality gates need strong synthesis and risk review.', dependsOn: ['code-unit-test-execution'] }, executionModelId)
+        step({ id: 'code-product-direction', stage: 'product-direction', owner: 'peaks-code', modelTier: 'top-tier', reason: 'Product direction needs strong judgment before execution work is delegated.', dependsOn: [] }, executionModelId, strongestModelId),
+        step({ id: 'code-design-direction', stage: 'design-direction', owner: 'peaks-code', modelTier: 'top-tier', reason: 'Design direction uses the recommended default before cheaper implementation work.', dependsOn: ['code-product-direction'] }, executionModelId, strongestModelId),
+        step({ id: 'code-tech-direction', stage: 'tech-direction', owner: 'peaks-tech', modelTier: 'top-tier', reason: 'Technical boundaries and approval gates use the recommended default with high-confidence planning.', dependsOn: ['code-design-direction'] }, executionModelId, strongestModelId),
+        step({ id: 'code-tech-review', stage: 'tech-review', owner: 'peaks-tech', modelTier: 'top-tier', reason: 'Tech artifacts and gate decisions require strong review and a recommended default path.', dependsOn: ['code-tech-direction'] }, executionModelId, strongestModelId),
+        step({ id: 'code-rd-planning', stage: 'rd-planning', owner: 'peaks-rd', modelTier: 'top-tier', reason: 'RD task decomposition and acceptance criteria use the recommended default before execution delegation.', dependsOn: ['code-tech-review'] }, executionModelId, strongestModelId),
+        step({ id: 'code-coding-execution', stage: 'coding-execution', owner: 'peaks-rd', modelTier: executionModelId === strongestModelId ? 'top-tier' : 'mid-tier', reason: `Coding and routine refactoring must use the configured execution worker model ${executionModelId}.`, dependsOn: ['code-rd-planning'] }, executionModelId, strongestModelId),
+        step({ id: 'code-unit-test-execution', stage: 'unit-test-execution', owner: 'peaks-rd', modelTier: executionModelId === strongestModelId ? 'top-tier' : 'mid-tier', reason: `Unit test authoring and focused test runs must use the configured execution worker model ${executionModelId}.`, dependsOn: ['code-coding-execution'] }, executionModelId, strongestModelId),
+        step({ id: 'code-quality-review', stage: 'quality-review', owner: 'peaks-code', modelTier: 'top-tier', reason: 'Reducer and final quality gates need strong synthesis and risk review.', dependsOn: ['code-unit-test-execution'] }, executionModelId, strongestModelId)
     ];
 }
-function createSoloStepsForMode(codeMode, executionModelId) {
-    return annotateSteps(createSoloSteps(executionModelId), codeMode);
+function createSoloStepsForMode(codeMode, executionModelId, strongestModelId) {
+    return annotateSteps(createSoloSteps(executionModelId, strongestModelId), codeMode);
 }
-function createTeamSteps(executionModelId) {
+function createTeamSteps(executionModelId, strongestModelId) {
     return [
-        step({ id: 'team-product-direction', stage: 'product-direction', owner: 'human', modelTier: 'top-tier', reason: 'Team product direction should stay on the governed planning path.', dependsOn: [] }, executionModelId),
-        step({ id: 'team-design-direction', stage: 'design-direction', owner: 'human', modelTier: 'top-tier', reason: 'Team design direction should preserve reviewability and accountability.', dependsOn: ['team-product-direction'] }, executionModelId),
-        step({ id: 'team-tech-direction', stage: 'tech-direction', owner: 'peaks-tech', modelTier: 'top-tier', reason: 'Team technical plans should remain strongly governed before RD execution.', dependsOn: ['team-design-direction'] }, executionModelId),
-        step({ id: 'team-tech-review', stage: 'tech-review', owner: 'peaks-tech', modelTier: 'top-tier', reason: 'Team tech approval requires strong review before execution.', dependsOn: ['team-tech-direction'] }, executionModelId),
-        step({ id: 'team-rd-planning', stage: 'rd-planning', owner: 'peaks-rd', modelTier: 'top-tier', reason: 'Team RD task decomposition remains on the governed strongest-model path.', dependsOn: ['team-tech-review'] }, executionModelId),
-        step({ id: 'team-coding-execution', stage: 'coding-execution', owner: 'peaks-rd', modelTier: executionModelId === STRONGEST_MODEL_ID ? 'top-tier' : 'mid-tier', reason: `Bounded coding tasks must use the configured execution worker model ${executionModelId}.`, dependsOn: ['team-rd-planning'] }, executionModelId),
-        step({ id: 'team-unit-test-execution', stage: 'unit-test-execution', owner: 'peaks-rd', modelTier: executionModelId === STRONGEST_MODEL_ID ? 'top-tier' : 'mid-tier', reason: `Unit-test tasks must use the configured execution worker model ${executionModelId}.`, dependsOn: ['team-coding-execution'] }, executionModelId),
-        step({ id: 'team-quality-review', stage: 'quality-review', owner: 'peaks-rd', modelTier: 'top-tier', reason: 'Team RD outputs still need reducer and quality review gates.', dependsOn: ['team-unit-test-execution'] }, executionModelId)
+        step({ id: 'team-product-direction', stage: 'product-direction', owner: 'human', modelTier: 'top-tier', reason: 'Team product direction should stay on the governed planning path.', dependsOn: [] }, executionModelId, strongestModelId),
+        step({ id: 'team-design-direction', stage: 'design-direction', owner: 'human', modelTier: 'top-tier', reason: 'Team design direction should preserve reviewability and accountability.', dependsOn: ['team-product-direction'] }, executionModelId, strongestModelId),
+        step({ id: 'team-tech-direction', stage: 'tech-direction', owner: 'peaks-tech', modelTier: 'top-tier', reason: 'Team technical plans should remain strongly governed before RD execution.', dependsOn: ['team-design-direction'] }, executionModelId, strongestModelId),
+        step({ id: 'team-tech-review', stage: 'tech-review', owner: 'peaks-tech', modelTier: 'top-tier', reason: 'Team tech approval requires strong review before execution.', dependsOn: ['team-tech-direction'] }, executionModelId, strongestModelId),
+        step({ id: 'team-rd-planning', stage: 'rd-planning', owner: 'peaks-rd', modelTier: 'top-tier', reason: 'Team RD task decomposition remains on the governed strongest-model path.', dependsOn: ['team-tech-review'] }, executionModelId, strongestModelId),
+        step({ id: 'team-coding-execution', stage: 'coding-execution', owner: 'peaks-rd', modelTier: executionModelId === strongestModelId ? 'top-tier' : 'mid-tier', reason: `Bounded coding tasks must use the configured execution worker model ${executionModelId}.`, dependsOn: ['team-rd-planning'] }, executionModelId, strongestModelId),
+        step({ id: 'team-unit-test-execution', stage: 'unit-test-execution', owner: 'peaks-rd', modelTier: executionModelId === strongestModelId ? 'top-tier' : 'mid-tier', reason: `Unit-test tasks must use the configured execution worker model ${executionModelId}.`, dependsOn: ['team-coding-execution'] }, executionModelId, strongestModelId),
+        step({ id: 'team-quality-review', stage: 'quality-review', owner: 'peaks-rd', modelTier: 'top-tier', reason: 'Team RD outputs still need reducer and quality review gates.', dependsOn: ['team-unit-test-execution'] }, executionModelId, strongestModelId)
     ];
 }
 function uniqueStrings(values) {
     return [...new Set(values)];
 }
-function createModelRouting(steps, executionModelId) {
+function createModelRouting(steps, executionModelId, strongestModelId) {
     return {
         strongestModel: {
-            modelId: STRONGEST_MODEL_ID,
+            modelId: strongestModelId,
             uses: steps.filter((step) => step.modelRole === 'strongest').map((step) => step.stage)
         },
         executionModel: {
@@ -135,10 +135,10 @@ function getTechPlanBlockedReasons(techPlan) {
 function getTechPlanNextActions(techPlan) {
     return [...techPlan.nextActions];
 }
-function createModeStatus(economyMode, swarmMode, executionModelId, executionModelSource) {
+function createModeStatus(economyMode, swarmMode, executionModelId, executionModelSource, strongestModelId) {
     const economySummary = economyMode
         ? `Economy mode enabled: code worker and test worker strictly use ${executionModelId} from config providers.`
-        : `Economy mode disabled: code worker and test worker use ${STRONGEST_MODEL_ID}, matching planner/reviewer.`;
+        : `Economy mode disabled: code worker and test worker use ${strongestModelId}, matching planner/reviewer.`;
     const swarmSummary = swarmMode
         ? 'Swarm mode enabled: peaks-rd coding, unit-test, and peaks-qa quality work must be represented as swarm worker tasks.'
         : 'Swarm mode disabled: swarm worker graph generation is bypassed.';
@@ -183,8 +183,14 @@ export function createWorkflowRouterPlan(request) {
     // object (`config: { providers: {} }`) still surfaces the "must be
     // configured" error from `getConfiguredExecutionModelId`.
     const effectiveProviders = request.config?.providers ?? { minimax: { model: 'minimax-2.7' } };
-    const executionModelId = economyMode !== false ? getConfiguredExecutionModelId(effectiveProviders) : STRONGEST_MODEL_ID;
-    const modeStatus = createModeStatus(economyMode, swarmMode, executionModelId, economyMode ? 'config.providers' : 'planner-reviewer-strongest-model');
+    // Slice 2026-07-09 add-zcode-adapter (A.3): the strongest
+    // planner/reviewer model is now resolved via `getStrongestModelId`,
+    // which reads `config.model` (if present) or falls back to the
+    // env-var-overridable default. This decouples the strongest-model
+    // pick from any specific vendor constant.
+    const strongestModelId = getStrongestModelId(request.config);
+    const executionModelId = economyMode !== false ? getConfiguredExecutionModelId(effectiveProviders) : strongestModelId;
+    const modeStatus = createModeStatus(economyMode, swarmMode, executionModelId, economyMode ? 'config.providers' : 'planner-reviewer-strongest-model', strongestModelId);
     const codeMode = getCodeMode(request.mode, request.codeMode);
     const decisionProfile = getDecisionProfileSummary(request.mode, codeMode);
     const artifactWorkspacePath = request.artifactWorkspacePath ?? (request.workspace ? getLocalArtifactPath(request.workspace) : undefined);
@@ -195,7 +201,7 @@ export function createWorkflowRouterPlan(request) {
     const techStatus = getTechStatus({ sessionId: request.sessionId, ...sharedWorkspaceOptions });
     const techPlan = createTechPlan({ sessionId: request.sessionId, goal, swarm: swarmMode, dryRun: true, ...sharedWorkspaceOptions });
     const rdPlan = createRdSwarmPlan({ skill: 'rd', sessionId: request.sessionId, goal, maxWorkers, swarmMode, executionModelId, dryRun: true, ...sharedWorkspaceOptions });
-    const steps = codeMode ? createSoloStepsForMode(codeMode, executionModelId) : createTeamSteps(executionModelId);
+    const steps = codeMode ? createSoloStepsForMode(codeMode, executionModelId, strongestModelId) : createTeamSteps(executionModelId, strongestModelId);
     const blockedReasons = uniqueStrings([
         ...techStatus.blockedReasons,
         ...getTechPlanBlockedReasons(techPlan),
@@ -213,7 +219,7 @@ export function createWorkflowRouterPlan(request) {
         decisionProfile,
         dryRun: true,
         routePolicy: request.mode === 'code' ? 'code-broad-multi-model' : 'team-rd-limited-multi-model',
-        modelRouting: createModelRouting(steps, executionModelId),
+        modelRouting: createModelRouting(steps, executionModelId, strongestModelId),
         modelAssignments: createModelAssignments(steps),
         modeStatus,
         techStatus,

package/dist/src/shared/version.d.ts

@@ -1 +1 @@
-export declare const CLI_VERSION = "4.0.0-beta.6";
+export declare const CLI_VERSION = "4.0.0-beta.7";

package/dist/src/shared/version.js

@@ -1 +1 @@
-export const CLI_VERSION = "4.0.0-beta.6";
+export const CLI_VERSION = "4.0.0-beta.7";

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "peaks-loop",
-  "version": "4.0.0-beta.6",
+  "version": "4.0.0-beta.7",
   "description": "Loop Engineering CLI — workflow primitive / loop guards / evaluators / slice orchestration",
   "author": "SquabbyZ",
   "license": "MIT",

package/scripts/install-skills.mjs

@@ -172,21 +172,15 @@ function readPackageVersion(packageRoot = resolvePackageRoot()) {
 }
 
 function createConfigDefaults(packageRoot) {
-  return {
+return {
     version: readPackageVersion(packageRoot),
     currentWorkspace: null,
   workspaces: [],
   language: 'en',
-  model: 'sonnet',
   economyMode: true,
   swarmMode: true,
   tokens: {},
-  providers: {
-    minimax: {
-      model: 'minimax-2.7'
-    }
-  },
-    proxy: {}
+  proxy: {}
   };
 }
 
@@ -381,7 +375,7 @@ function resolveProjectRoot(options) {
  * overrides `PEAKS_CLAUDE_SKILLS_DIR` / `PEAKS_CLAUDE_OUTPUT_STYLES_DIR`
  * continue to work, and the legacy default is preserved.
  */
-const IDE_DETECTION_DIRS = [
+export const IDE_DETECTION_DIRS = [
   { id: 'claude-code', dir: '.claude' },
   { id: 'trae', dir: '.trae' },
   { id: 'trae-cn', dir: '.trae-cn' },
@@ -389,6 +383,7 @@ const IDE_DETECTION_DIRS = [
   { id: 'cursor', dir: '.cursor' },
   { id: 'qoder', dir: '.qoder' },
   { id: 'tongyi-lingma', dir: '.tongyi-lingma' },
+  { id: 'zcode', dir: '.zcode' },
 ];
 
 /**
@@ -404,7 +399,7 @@ const IDE_DETECTION_DIRS = [
  * directory was never populated. 2.0 fixes this by giving
  * all 8 platforms canonical install paths.
  */
-const IDE_SKILL_INSTALL_PROFILES = {
+export const IDE_SKILL_INSTALL_PROFILES = {
   'claude-code': {
     skillsDir: join(homedir(), '.claude', 'skills'),
     outputStylesDir: join(homedir(), '.claude', 'output-styles'),
@@ -469,6 +464,14 @@ const IDE_SKILL_INSTALL_PROFILES = {
     envVar: 'PEAKS_OPENCLAW_SKILLS_DIR',
     outputStylesEnvVar: 'PEAKS_OPENCLAW_OUTPUT_STYLES_DIR',
   },
+  'zcode': {
+    skillsDir: join(homedir(), '.zcode', 'skills'),
+    outputStylesDir: join(homedir(), '.zcode', 'output-styles'),
+    agentsDir: join(homedir(), '.zcode', 'agents'),
+    envVar: 'PEAKS_ZCODE_SKILLS_DIR',
+    outputStylesEnvVar: 'PEAKS_ZCODE_OUTPUT_STYLES_DIR',
+    agentsEnvVar: 'PEAKS_ZCODE_AGENTS_DIR',
+  },
 };
 
 function detectInstalledIdeId(projectRoot) {

package/skills/peaks-code/SKILL.md

@@ -98,9 +98,13 @@ After Step 0.7 returns "fresh", run `peaks upgrade --detect-1x --project <root>
 
 ### Peaks-Loop Step 0.8 — Job 启动 (BLOCKING on LLM judgement — v3.1.1 patch + v3.1.2 mechanical gates)
 
-The CLI is a **recorder + gate** for job-shape (the LLM judges). LLM calls `peaks code detect-job --is-job <bool> --rationale <text> --suggested-job-id <jid> --suggested-strategy <single|rotating> --confidence <high|medium|low>`, which writes `.peaks/_runtime/<sessionId>/job-shape.json`. Downstream steps call `read-job-shape` and refuse if missing.
+> **CLI reality check (D-001 sediment, 2026-07-09):** The CLI surface has changed since this section was last verified. The actual command is **`peaks job init --job-id <jid> --slice-list <list> --main-loop-strategy <single|rotating> [--parallelism-hint <serial|llm-decides>] [--exit-policy <strict|best-effort>] [--project <repo>]`**. The legacy `peaks code detect-job --is-job/--suggested-job-id/--confidence` form described below is **no longer present** in 4.0.0-beta.6. If you find `--is-job` rejected, fall back to `peaks job init`.
 
-**Hard rule (v3.1.1 red-line #10):** LLM MUST NOT skip `peaks code detect-job`. If `read-job-shape` throws `JOB_SHAPE_NOT_DECIDED`, record a decision before proceeding.
+The CLI is a **recorder + gate** for job-shape (the LLM judges). LLM calls `peaks job init --job-id <jid> --slice-list <list> --main-loop-strategy <single|rotating>` (with `--caller-id <id>` and `--project <repo>` in non-Claude-Code environments), which writes `.peaks/_runtime/<sessionId>/job/<jid>/state.json` and emits a `[job-event] {kind: job-started, jobId, total, strategy}` line. Downstream steps call `peaks job status --job-id <jid>` to read state.
+
+**Hard rule (v3.1.1 red-line #10):** LLM MUST NOT skip the job-init call. If `peaks job status` reports `done: 0` and the slice work has begun, the LLM forgot to initialize — re-call `peaks job init` immediately.
+
+> **D-003 sediment (2026-07-09):** The historical `JOB_SHAPE_NOT_DECIDED` exception thrown by `read-job-shape` is no longer wired in 4.0.0-beta.6. The current CLI uses `peaks job status` which reports a passive `done: 0` instead of throwing. Both behaviors enforce the same intent (no work may begin before init) but the failure mode is a soft warning rather than a hard exception. LLM should treat `done: 0` after Step 0.8 as a recoverable miss, not an unrecoverable error.
 
 **v3.1.2 mechanical gates** (recorder-only was bypassed twice):
 
@@ -143,7 +147,9 @@ Run `peaks project memories --project <repo> --json` to read decisions / convent
 
 ### Peaks-Loop Step 2.5: Set session title
 
-Extract a short title from the user's first request (8-20 Chinese chars or 4-10 English words). Run `peaks session title` with the active sid. Skip if a title is already set. → `references/skill-presence-and-title.md` (same file as Step 2).
+Extract a short title from the user's first request (8-20 Chinese chars or 4-10 English words). Run `peaks session title <sid> "<title>" --json`. Skip if a title is already set. → `references/skill-presence-and-title.md` (same file as Step 2).
+
+> **CLI reality check (D-002 sediment, 2026-07-09):** The session id is a **positional argument**, not a `--session-id` flag. Correct invocation is `peaks session title 2026-07-08-session-17918f "add-zcode-adapter — peaks-loop 第 9 IDE 接入" --json`. Passing `--session-id` will be rejected with `error: unknown option '--session-id'`.
 
 ## Sub-agent session sharing (MANDATORY — one conversation = one sid)
 
@@ -153,6 +159,19 @@ When peaks-code dispatches a sub-agent (peaks-rd, peaks-qa, peaks-ui, peaks-txt,
 
 Peaks-Loop Code may: identify scenarios, recommend profiles, coordinate role skills through artifacts, coordinate project memory extraction, request user confirmation at risk/commit boundaries. Peaks-Loop Code must NOT silently install hooks / create agents / enable MCP / modify Claude settings / create GitHub repos / bypass role-skill artifacts. → `references/boundaries.md`.
 
+## CLI Drift Index (sediment 2026-07-09)
+
+> **Reading guide:** This document was last verified against CLI surface in **peaks-loop 4.0.0-beta.6**. The CLI has drifted in 4 places since this SKILL.md was last fully aligned. Each drift below is annotated **inline at the relevant step** with a `> CLI reality check` blockquote. When you hit a `error: unknown option ...` error, **read the inline reality check first** before guessing.
+
+| Drift ID | Step | Symptom | Fix | Inline location |
+|---|---|---|---|---|
+| **D-001** | 0.8 | `peaks code detect-job --is-job ...` rejected with `error: unknown option '--is-job'` | Use `peaks job init --job-id <jid> --slice-list <list> --main-loop-strategy <single\|rotating>` | §Step 0.8 first paragraph |
+| **D-002** | 2.5 | `peaks session title --session-id <sid> ...` rejected with `error: unknown option '--session-id'` | sid is positional: `peaks session title <sid> "<title>" --json` | §Step 2.5 |
+| **D-003** | 0.8 | `JOB_SHAPE_NOT_DECIDED` exception expected but never thrown | Current behavior is `peaks job status` reports `done: 0` passively — treat as recoverable miss, not hard error | §Step 0.8 third paragraph |
+| **D-010** | 11c | `peaks memory extract` returns `extractedCount: 0` despite `<!-- peaks-memory:start -->` existing | Block requires YAML frontmatter (`title:` + `kind:` + `---`) + closing `<!-- peaks-memory:end -->`. Bare `peaks-memory:start` is parsed silently but produces no writes | §Step 11c + 11d |
+
+> **Sediment lesson (master record):** `.peaks/memory/peaks-code-runbook-4-0-0-beta-6-skill-md-cli-d-001-d-002-d-003-d-010.md`
+
 ## Peaks-Loop GStack integration
 
 Map gstack stages to Peaks-Loop role artifacts; preserve confirmation gates. → `references/gstack-integration.md` + `references/browser-workflow.md`.
@@ -233,11 +252,13 @@ After final validation, refresh project-local standards via `peaks standards ini
 
 **11c — Canonical extract** (the only CLI that writes `.peaks/memory/`): `peaks memory extract --project <repo> --artifact .peaks/_runtime/<sessionId>/txt/handoff.md --apply --json`. `--apply` is REQUIRED (without it the command only previews — no files land).
 
+> **CLI reality check (D-010 sediment, 2026-07-09):** The `<!-- peaks-memory:start -->` block must be followed immediately by a **YAML frontmatter** (`title: ...`, `kind: lesson | decision | convention`) and a `---` separator. Each block must close with `<!-- peaks-memory:end -->`. A bare `peaks-memory:start` without the YAML fields is parsed but produces no `plannedWrites` — the CLI silently returns `extractedCount: 0`. Full format reference: see any successful extract output (e.g. `.peaks/_runtime/2026-07-08-session-17918f/txt/handoff-003-add-zcode-adapter.md`).
+
 **11d — Gate C (zero-write outcome):** If `extractedCount === 0` after 11c, fire AskUserQuestion:
 
 > "本次 code 未沉淀任何 `.peaks/memory` 文件。可选: (a) 回去在 handoff.md 嵌入至少 1 个 `peaks-memory:start` block 后重试; (b) 显式接受 no-sediment 并记录为 lesson; (c) 取消完成。"
 
-Default option = (a). Code MUST NOT silently accept (b) without user pick. Full bash + flow at `references/runbook.md` §Step 11.
+> **D-010 fix root cause check:** When firing 11d, first inspect whether the block has the YAML frontmatter (`title:` + `kind:` + `---`). If the `<!-- peaks-memory:start -->` exists but no `title:` line follows, fix the block format and re-run 11c — don't ask the user yet. Default option = (a). Code MUST NOT silently accept (b) without user pick. Full bash + flow at `references/runbook.md` §Step 11.
 
 **Why this step exists:** audit 2026-07-03 confirmed 2 consecutive sessions produced zero `.peaks/memory/` files despite completing RD + QA + handoff artifacts; `assisted` mode silently skipped runbook Step 10 (no STOP condition). **Why `peaks memory extract` (not `peaks project memories:extract`):** the artifact-scoped extract is canonical; the batch-scoped sibling is for non-handoff flows. Always use `peaks memory extract --apply`.
 

package/skills/peaks-slice-decompose/SKILL.md

@@ -108,4 +108,64 @@ This skill does NOT modify source code. It documents the v2 envelope produced by
 |---|---|
 | `references/v2-schema.md` | Field-by-field table for `DecompositionResultV2`, `PassResult`, `SliceV2`, `CrossPassEdge.arbitratedBy`, `LlmArbitration`. |
 | `references/granularity-decision.md` | Decision tree for the 4 `--granularity` modes; `shouldSubdivide()` thresholds + tie-break semantics. |
-| `references/cross-pass-edge-interpretation.md` | The 4 edge kinds and how `peaks-rd` should use them for dispatch ordering; `LlmArbitration` trace reader. |
+| `references/cross-pass-edge-interpretation.md` | The 4 edge kinds and how `peaks-rd` should use them for dispatch ordering; `LlmArbitration` trace reader. |
+
+---
+
+## SC template hard rules (sediment 2026-07-09)
+
+These rules are added in response to **D-009a + D-009c** discovered while running RID 003-add-zcode-adapter. They prevent the same class of mistake from re-occurring in future SC drafts.
+
+### Rule 1 — Pre-flight: detect ESM vs CJS before writing export syntax (D-009c)
+
+Before specifying `--caller-id` / export guard syntax in any SC task, run:
+
+```bash
+head -3 package.json | grep '"type"'
+```
+
+- `"type": "module"` → ESM. Use `export const FOO = ...` for any constant that tests need to import. **Do NOT** write `if (typeof module !== 'undefined' && module.exports) { module.exports = { FOO }; }` — that pattern is CJS-only.
+- No `"type"` field (or `"type": "commonjs"`) → CJS. Use the `module.exports` guard.
+
+Why: RID-003 SC §3.5 originally specified the CJS guard for `scripts/install-skills.mjs`. The script is actually ESM (`"type": "module"`). The RD agent had to backtrack and the SC lost 5 minutes of spec clarity.
+
+### Rule 2 — Whitelist fixture sync is NOT collateral (D-009a)
+
+When a slice adds a new entry to a **whitelist fixture** (e.g. `tests/unit/ide/ide-registry.test.ts` asserting `expect(adapters).toHaveLength(7)` after adding the 8th adapter), that fixture sync is a **primary task**, not a "by the way" footnote.
+
+Mandatory structure for any slice that adds to a known set:
+
+```markdown
+### Tasks
+
+| Task | File | Change |
+|---|---|---|
+| T-1 | `src/services/ide/ide-types.ts` | Add `'zcode'` to IdeId |
+| T-2 | `src/services/ide/ide-registry.ts` | Register ZCODE_ADAPTER |
+| T-3 | `tests/unit/ide/ide-registry.test.ts` | Update whitelist: 7 → 8 adapters, add 'zcode' |
+| T-4 | `tests/unit/install-skills-script.test.ts` | Update whitelist: 5 → 6 platforms, add 'zcode' |
+```
+
+If you find a whitelist fixture during execution but didn't pre-list it as a task, treat it as a **NEW TASK** and run the SC ↔ QA repair loop, not as "necessary collateral". Pre-listing prevents surprises at Gate 1.
+
+### Rule 3 — Adapter interface optional fields (D-009b reciprocal)
+
+When extending `IdeAdapter` (or any plugin-interface in `src/services/**/ide/`), prefer **optional fields** over required fields so existing implementations don't need stub updates:
+
+```ts
+// ✅ Optional — back-compat safe
+interface IdeAdapter {
+  readonly detectCurrentModel?: () => Promise<string | undefined>;
+}
+
+// ❌ Required — every existing adapter must add a stub
+interface IdeAdapter {
+  readonly detectCurrentModel: () => Promise<string | undefined>;
+}
+```
+
+If a field MUST be required, document in the SC why each existing implementation must be updated inline, and pre-list those updates as primary tasks (Rule 2 pattern).
+
+---
+
+**Sediment source:** `.peaks/_runtime/2026-07-08-session-17918f/qa/003-add-zcode-adapter/{slice-A-completion,slice-B-completion}.md` D-009a/b/c sections. Master lesson: `.peaks/memory/peaks-code-runbook-4-0-0-beta-6-skill-md-cli-d-001-d-002-d-003-d-010.md`.