@mstar-harness/dsh 3.8.1 → 3.8.3

package/harness-skills/mstar-host/references/dsh.md

@@ -119,9 +119,9 @@ or a custom profile).
   doneAt — deterministic, documented heuristic, only provably
   cross-iteration events are dropped, no historical back-scan of resumed
   long logs; the sidebar chip title is captured at open time; a docked
-  body renders nothing while `tab.visible === false`. Panel acceptance is
-  dual-track: in-loop browser harness verification against the rebuilt
-  bundle plus user-restart final GUI acceptance.
+  body renders nothing while `tab.visible === false`. Routine panel QA uses affected unit evidence only. Real-browser rebuilt-bundle
+  verification or user-restart GUI acceptance belongs to an explicitly requested
+  independent **`mstar-e2e`** workflow (`/amazing-e2e-check`), never an iteration QA gate.
 
 ## Skill loading
 
@@ -138,6 +138,7 @@ or a custom profile).
 | dsh tool | Harness use |
 |----------|-------------|
 | **`subagent`** | Primary dispatch — the model-facing delegation tool the dispatch gate matches (default `toolName`; a renamed instance must be declared via Config `dispatchTools`) |
+| **`workflow`** | Read-only N≥3 fan-out — one run, one conversation `workflow-run` node (§ Read-only fan-out via the `workflow` tool; scripts → `references/dsh-workflow-scripts.md`) |
 | **`mstar_iteration_gate`** | Evaluate the iteration phase gate in-app (`evaluatePhaseGate` — `mstar iteration gate` parity) |
 | **`mstar_sdd_workspace`** / **`mstar_sdd_task_brief`** | SDD workspace resolve + task brief extraction (`mstar sdd …` parity) |
 | **`mstar_*_validate`** | On-demand seam validators (design-md / audit / compound / roles) |
@@ -208,59 +209,107 @@ tab's `EventLogPage` log page are pure consumers of this evidence.
   `beforeDispatch` followed by the identical text as an in-loop subagent tool
   call) records two dispatch events — the surfaces are mutually exclusive by
   design; the double record is documented, not deduplicated.
-- **File / bounds**: events append to `{HARNESS_DIR}/agent-flow.jsonl` (JSON
-  Lines, one event per line; harness dirs are gitignored by convention). The
-  ledger assumes ONE dsh process writes each harness dir (single-writer):
-  concurrent dsh sessions on the same repo can lose events (the append itself
-  is near-atomic O_APPEND, but truncation is a read-modify-write) — the loss
-  only under-reports actual flow in the panel, never a gate impact. After each
-  append the file truncates to the most recent **500** events; truncation is
-  size-gated (≈500 lines' typical size — small files stay append-only) and
-  performed as an atomic temp-file rename. The catalog read returns the
-  latest-first view with a default window of **50** and a role × outcome
-  summary. A MISSING file reads as the empty view ("no actual dispatches yet"
-  — recording starts at plan merge); an unreadable file is absent evidence;
-  malformed lines are skipped, never fatal.
+- **File / bounds**: events append to the ACTIVE workflow dir —
+  `{HARNESS_DIR}/workflows/<id>/agent-flow.jsonl` (JSON Lines, one event per
+  line; harness dirs are gitignored by convention) — never the harness root:
+  with no active lifecycle the record is SKIPPED with a one-time warn. The
+  append and the size-gated truncating read-modify-write form ONE critical
+  section behind a per-workflow lockdir, so a second dsh session sharing the
+  active lifecycle cannot silently drop the other writer's lines (steady state
+  stays one writer per workflow dir); any loss only under-reports actual flow
+  in the panel, never a gate impact. After each append the file truncates to
+  the most recent **500** events; truncation is size-gated (≈500 lines'
+  typical size — small files stay append-only) and performed as an atomic
+  temp-file rename. The catalog read returns the latest-first view with a
+  default window of **50** and a role × outcome summary. A MISSING file reads
+  as the empty view ("no actual dispatches yet" — recording starts at plan
+  merge); an unreadable file is absent evidence; malformed lines are skipped,
+  never fatal.
 - **Settle = real completion pairing, never faked**: `tools/post-execute`
   IS part of the
   verified dsh-tools registry surface (`runPostExecute` dispatches the
   waterfall for every tool call — verified against the upstream source and
   pinned by a real-call probe). The pairing listener matches dispatch TOOLS
-  (Config `dispatchTools`, default `['subagent']`), looks up the exec's
-  `callId` in the apply-scoped pairing store, and branches on the verified
-  result shapes:
-  - `{ kind: 'background', taskId }` → store `taskId → dispatchRef`; the REAL
-    settle arrives via `ctx.tasks.onTaskDone` (terminal mapping
-    completed → ok / killed → denied / failed → error, `durationMs` when
-    available), wired through `ctx.inject(['tasks'])`.
+  (Config `dispatchTools`, default `['subagent', 'subagent_fork']`), looks up
+  the exec's agent-namespaced call key in the apply-scoped pairing store, and
+  branches on the verified result shapes:
+  - `{ kind: 'background', jobId }` (the registry job id, `<kind>-N`) → store
+    `jobId → dispatchRef` and the bounded job id as the ref's `taskRef`; the
+    REAL settle arrives via `ctx.inject(['jobs'])` → `jobs.onJobDone`
+    (terminal mapping completed → ok / killed → denied / failed → error,
+    `durationMs` when available). A background value without a valid `jobId` →
+    nothing mappable (no settle).
   - `{ kind: 'continuable', subagentId }` → no terminal signal this round →
-    no settle (documented limit — the child owns its turns).
+    no settle (documented limit — the child owns its turns); the value
+    authorizes the child-identity join below and nothing is copied onto a
+    settle.
   - any other successful value (foreground included) → settle `ok`; a failed
-    result (`isError`) → settle `error`.
+    result (`isError` or an `error` payload) → settle `error`. A returned
+    foreground `runId` is the settle's `childId`, extracted independently of
+    the outcome (an error settle keeps its identity without becoming `ok`).
   Pairing is apply-scoped (in-memory `callId → dispatchRef` /
-  `taskId → dispatchRef` maps created in the entry `apply`; an HMR restart
+  `jobId → dispatchRef` maps created in the entry `apply`; an HMR restart
   resets them, and completions outside the window stay unpaired). Every
   PAIRED settle carries the paired dispatch's identity (`role`/`planId`/
   `taskId` — same field names + semantics as the dispatch event; the registry
-  background-task id is never written as `taskId`, `taskRef` is reserved for
-  it). Unpaired payloads (non-dispatch tools, calls outside the pairing
-  window) record NOTHING — the ledger stays dispatch-only, never a
-  fabricated settle.
+  job id is never written as `taskId` — `taskId` stays the Assignment `Task N`
+  tag, `taskRef` is reserved for the registry id). Unpaired payloads
+  (non-dispatch tools, calls outside the pairing window) record NOTHING — the
+  ledger stays dispatch-only, never a fabricated settle.
+- **Child identity (`subagent-link`, nonterminal)**: the child session id is
+  published upstream as a PARENT-OWNED `subagent/catalog` session event
+  (`{ version: 0, childId, childCreatedAt, mode, label }`; `label` = the
+  delegation `description`), appended by the tool body — for the continuable
+  path BEFORE the tool returns. The join spans a per-dispatch CALL WINDOW: the
+  pre-execute reserves the first raw-label slot under the live parent Session
+  object and captures its `seq` as the window start; a valid `background` /
+  `continuable` result at `tools/post-execute` makes that exact candidate
+  eligible (every other outcome retires it to a tombstone; a duplicate label
+  was already refused a candidate at reservation). Eligibility walks
+  `eventAt(seq)` over `[fromSeq, end)`, where `end` is the session's `seq`
+  CAPTURED when the candidate became eligible — recovering a catalog appended
+  before the tool returned — while ONE root-context `session/event` observer
+  feeds the same matcher for later arrivals against the session's CURRENT
+  `seq`: only that live observer follows the session forward, so a catalog
+  appended after eligibility still joins through it (the catch-up scan stays
+  frozen at its captured endpoint). A matched
+  candidate is consumed once and appends `{ v: 1, ts, kind: 'subagent-link',
+  agent?, childId, label, role, planId?, taskId?, taskRef? }` to the
+  DISPATCH's own workflow dir (`ts` = observation time), correlating the
+  catalog child back to the dispatch identity mstar recorded. It is an
+  IDENTITY record, NOT a completion: no `outcome`, no `verdict`, no `paired`
+  marker. A background one-shot link also carries its registry `taskRef`; a
+  continuable link omits it. Settle rows carry an optional `childId` — a
+  foreground `runId`, or for background only when the join has already
+  supplied one.
+- **Join bounds (honest degrade)**: NO row when the label is missing/empty,
+  the dispatch unpaired, a background result carries no valid `jobId` (the
+  reserved candidate is retired — nothing mappable: no settle, no link),
+  the catalog version unknown, the mode not matching the result kind,
+  a continuable catalog naming a different child than the tool returned,
+  the slot map at capacity (500 labels per parent Session), or the slot
+  already consumed. The join is apply-scoped: no whole-history cold
+  scan and no `session/created` backfill — a catalog written before apply
+  (constructor seeds) can never label a new dispatch. Duplicate labels are
+  deterministic best-effort (first reservation + first matching catalog wins),
+  NOT proof of unique ownership. Not every provider emits a catalog — a remote
+  run without a `localAgent` produces none, so a dispatch may legitimately
+  have no link row.
 - **Catalog**: `state.agentFlow` carries the ledger view (`events` ≤ 50,
   latest-first, + `summary`); the model-facing `<mstar_engine_status>` text
   renders ONE compact `agent flow: …` line only when events > 0 (role totals
   top-5 + latest dispatch with HH:MM — the event detail lives in the
-  structured source, never the model text). A ledger record (dispatch/settle)
-  invalidates the affected workspace's TTL cache entry IMMEDIATELY
-  (apply-scoped `harnessDir → cache key` reverse map + invalidation closure)
-  → the next pre-step rebuilds and (digest
-  text change) re-injects the row — the 60 s TTL no longer bounds
-  ledger-change latency; it still bounds non-ledger staleness.
+  structured source, never the model text). A ledger record
+  (dispatch/settle/link) invalidates the affected workspace's TTL cache entry
+  IMMEDIATELY (apply-scoped `harnessDir → cache key` reverse map +
+  invalidation closure) → the next pre-step rebuilds and (digest text change)
+  re-injects the row — the 60 s TTL no longer bounds ledger-change latency; it
+  still bounds non-ledger staleness.
 - **Maintainer view**: change the ledger shape (event schema, bounds, settle
   seam) and update the projections together — `gates/agent-flow.ts` (record /
-  read / settle listener), `gates/catalog.ts` (agent-flow line + `source`
-  view) and `client/panel/graph/project-graph.ts` (the ZoneView flow/agents
-  projection) — the panel renders ONLY what the evidence shows.
+  read / settle / catalog-join listeners), `gates/catalog.ts` (agent-flow line
+  + `source` view) and `client/panel/graph/project-graph.ts` (the ZoneView
+  flow/agents projection) — the panel renders ONLY what the evidence shows.
 
 ## PM dispatch
 
@@ -299,22 +348,102 @@ the "queued messages" dock instead of reaching the parent (observed on dsh).
 The closing message is the guaranteed delivery channel; reserve `report` for
 MID-turn findings that change what the parent should do next.
 
+### Progress discipline — native workflow, never `/goal`
+
+dsh progress is driven by the **native workflow**: the workflow snapshot phases
++ the dispatch gates + **subagent settle notifications**. mstar **stops arming**
+a goal on dsh — the surviving bridge is advisory-only (no `create` / `edit` /
+`complete` / `pause` / `resume`, no goal read) — so no goal round loop drives
+mstar work here. Never drive a dsh session with a `/goal` objective or a goal
+round loop: `goal-round-driver` opens a round whenever the goal is active +
+armed and the agent is idle, and it knows nothing about running subagents, so
+an operator who arms `/goal` manually can still get rounds firing while a
+dispatched child owns the critical path.
+
+**Phase 2 continuous execution is a PM-local loop**: dispatch → **wait for the
+child's settle notification** → next dispatch. When a dispatched child owns the
+critical path, the correct action is to **wait** — not to open another unit of
+work against the same worktree.
+
 ### QC default
 
-- **`Execution mode: sdd`**: **N=3** `subagent` dispatches — one per QC seat
-  (`qc-specialist`, `qc-specialist-2`, `qc-specialist-3`), each body **Act as**
-  the respective QC role + QC skill load. **MUST dispatch all three with
-  `run_in_background: true` in one message** → the seats run CONCURRENTLY
-  (background children; wall ≈ single seat); foreground (no
-  `run_in_background`) runs serially (wall ≈ 3× single seat) and does NOT
-  count as parallel tri. Cannot emit required **N** → **`Blocked`**.
+- **`Execution mode: sdd`**: **N=3** seats — one per QC seat (`qc-specialist`,
+  `qc-specialist-2`, `qc-specialist-3`), each body **Act as** the respective QC
+  role + QC skill load. Read-only fan-out of N≥3 on dsh uses the native
+  **`workflow`** tool (the `mstar-qc-tri` script — § Read-only fan-out via the
+  `workflow` tool): one run, three concurrent children, one conversation
+  `workflow-run` node. When the tool is not mounted (the `ptc` preset hides it),
+  fall back to the `subagent` path below. **The `subagent` path MUST dispatch all
+  three with `run_in_background: true` in one message** → the seats run
+  CONCURRENTLY (background children; wall ≈ single seat); foreground (no
+  `run_in_background`) runs serially (wall ≈ 3× single seat) and does NOT count
+  as parallel tri. Cannot emit required **N** → **`Blocked`**.
 - **`inline`**: **N=1**.
 
-### SDD implement (serial)
-
-- **`Execution mode: sdd`**: one implementer `subagent` dispatch per task id;
-  task reviewer = a separate dispatch (SDD review role) — no sticky resume
-  unless the host's continuable-subagent id is available and recorded.
+### SDD implement
+
+- **`Execution mode: sdd`**: one implementer `subagent` dispatch per ready task id;
+  independent tasks use isolated tracks and `run_in_background: true` before
+  waiting, per **`mstar-sdd`** § Ready-task scheduling. Task reviewer is a fresh
+  separate dispatch; sticky resume is limited to one sequential owner track
+  with a recorded continuable-subagent id.
+
+## Read-only fan-out via the `workflow` tool
+
+dsh also exposes the upstream **`workflow`** tool
+(`@deepseek-ai/dsh-tool-workflow`, mounted by the shipped agent presets; the
+`ptc` preset disables it in favour of its own orchestration surface). It runs a
+model-written plain-JavaScript script that fans children out inside ONE run; the
+run is recorded as durable `tool-workflow/*` session events and the stock dsh UI
+(`dsh-client-ui-workflow-run`) folds them into one conversation **`workflow-run`**
+node the operator expands by phase and member. Use it for **read-only fan-out of
+N ≥ 3 seats** — plan QC tri, large-repo audit categories, `/amazing-pr-review
+deep` seats — and copy the `script` + `meta` + `args` from this skill →
+`references/dsh-workflow-scripts.md`. For **1–2** delegations keep **`subagent`**
+(the tool's own guidance): the two-seat default tier of `/amazing-pr-review`
+shows two subagent cards and no `workflow-run` node, and that is expected.
+
+**Read-only only.** A workflow child is a delegated child (the shipped `spawn`
+provider pins `approval: never` for the whole delegation), and the run has **no
+per-child pre-start veto seam** — so a script is never the channel for writable
+work; writable fan-out stays on `subagent` behind the dispatch and lease gates.
+Seats return findings in their result payload and must never depend on writing
+files — the caller persists the seat reports.
+
+**Every `agent()` prompt starts with the Assignment header** — `## Assignment`
+plus `Execute as` / `Delegation` / `Task category` as the first lines:
+
+```markdown
+## Assignment
+
+Execute as: qc-specialist
+Delegation: forbidden
+Task category: audit
+```
+
+Role binding on dsh is prompt-only (there is no `agent` field), and the same
+engine grammar is what the role-persona channel parses
+(`packages/dsh/src/gates/role-persona.ts` reads only the header region) — so
+`Execute as: qc-specialist` resolves the QC role persona for that child. Keep
+body-quoted field examples out of the header region, and never pass the deferred
+`agentType` option: the engine rejects it loudly.
+
+| Operator types | N | Tool | `meta.name` | Operator sees |
+|---|---|---|---|---|
+| `/codebase-audit` (large repo) | ≥3 | native `workflow` | `mstar-audit-fanout` | conversation `workflow-run` node |
+| `/amazing-pr-review deep` | ≥3 | native `workflow` | `mstar-pr-seats` | same |
+| `/amazing-pr-review` default tier | 2 | `subagent` | — | two subagent cards, no node (expected) |
+| Plan QC tri (PM already in session, no extra slash) | 3 | native `workflow` | `mstar-qc-tri` | conversation `workflow-run` node |
+| Any 1–2 read-only delegation | 1–2 | `subagent` | — | expected |
+
+`meta.name` is the gate identity — keep it kebab-case and on the recommended
+list. With the default Config (`workflowNames` unset) every name is *unknown*,
+which under the default `workflowGate: warn` is one `workflow.name.unknown`
+**advisory that the run survives** — acceptable on a first run, not a failure. A
+production overlay may set `workflowNames: ['mstar-qc-tri', 'mstar-audit-fanout',
+'mstar-pr-seats']` (and, separately, `workflowGate: hard`); both are operator
+choices, never mstar defaults. The same run also reaches the panel's 事件记录 tab
+through the agent-flow ledger (§ Agent-flow ledger).
 
 ## Commands and skills paths
 

package/harness-skills/mstar-host/references/kimi.md

@@ -97,10 +97,10 @@ Harness **dispatch** on Kimi = **one or more `Agent` tool calls** with correct *
 
 Cannot emit required **N** → **`Blocked`**.
 
-### SDD implement (serial)
+### SDD implement
 
-- **`Execution mode: sdd`**: one implementer **`Agent`** per task id; task reviewer = new **`Agent`** with **Act as `code-reviewer`** (Kimi L2 review; not qc-specialist*), always via generic fallback `subagent_type: "coder"` per C5 — no sticky resume unless host adds it later. Serial rule → **`parallel-dispatch.md`** § SDD implement.
-- **Never** multiple implementer Agents in one message for the same plan.
+- **`Execution mode: sdd`**: one implementer **`Agent`** per task id; task reviewer = new **`Agent`** with **Act as `code-reviewer`** (Kimi L2 review; not qc-specialist*), always via generic fallback `subagent_type: "coder"` per C5 — no sticky resume unless host adds it later. Ready-task scheduling → **`parallel-dispatch.md`** § SDD implement.
+- Independent ready implementers use isolated parallel tracks; never share a writable worktree or session.
 
 ## Clarify
 

package/harness-skills/mstar-host/references/omp.md

@@ -188,10 +188,10 @@ Harness **dispatch** on omp = **one or more `task` tool calls** with correct **`
 
 Cannot emit required **N** → **`Blocked`**.
 
-### SDD implement (serial)
+### SDD implement
 
-- **`Execution mode: sdd`**: one implementer `task` entry per task id with `agent` matching the implementer role when listed; task reviewer = new entry with `agent: "code-reviewer"` (omp L2 review; not qc-specialist*) or `agent: "reviewer"`/`"task"` as fallback + C5b — no sticky resume unless host resume/id is available and recorded. Serial rule → **`parallel-dispatch.md`** § SDD implement.
-- **Never** multiple implementer entries in one message for the same plan.
+- **`Execution mode: sdd`**: one implementer `task` entry per task id with `agent` matching the implementer role when listed; task reviewer = new entry with `agent: "code-reviewer"` (omp L2 review; not qc-specialist*) or `agent: "reviewer"`/`"task"` as fallback + C5b — no sticky resume unless host resume/id is available and recorded. Ready-task scheduling → **`parallel-dispatch.md`** § SDD implement.
+- Independent ready implementers use isolated parallel tracks; never share a writable worktree or session.
 
 ## Clarify
 

package/harness-skills/mstar-host/references/parallel-dispatch.md

@@ -46,16 +46,16 @@ Formal iteration Phase 2 uses the same SDD + tri rule — not a separate carve-o
 
 （**SDD 默认**已在上一节；本节仅覆盖显式 tri 的非 SDD 场景。）
 
-## SDD implement (serial — not parallel)
+## SDD implement
 
-- **`Execution mode: sdd`**: implementer and task reviewer dispatches are **one at a time** per task.
-- **Never** multiple implementer Tasks in one message for the same plan.
-- See **`mstar-sdd`**.
+- Follow **`mstar-sdd`** § Ready-task scheduling: independent ready tasks run concurrently after per-track worktree and artifact isolation; fresh implementers, one fresh reviewer per task.
+- Serialize only actual dependencies, overlapping writers, one sticky session, and integration merges. PM alone updates shared progress.
+- When the host only exposes separate asynchronous starts, issue every ready call before waiting for any result; same-message packaging is required only when supported.
 
 ## QC targeted re-review (after fixes)
 
 - Assignment: **`QC re-review: targeted — reviewers: <role-ids>`** → **N** = listed seats only (1–3), **one** dispatch turn with **N** invocations.
-- Do **not** default to three invocations after a routine fix round.
+- Do **not** default to three invocations after a routine fix round. Each listed seat receives only its findings and fix delta; tri seat count never expands review scope.
 - Post-dispatch: verify only **dispatched** seats returned; PM updates same bundle `qc-consolidated.md` and durable plan summary (see `mstar-artifacts/references/plan-files-and-reports.md`).
 
 ## Self-check before send
@@ -65,4 +65,4 @@ Formal iteration Phase 2 uses the same SDD + tri rule — not a separate carve-o
 3. Dispatch message contains **exactly `N`** invocation calls?
 4. **Each** invocation carries its role-binding field set to **`Execute as`** (omp `agent` / Cursor `subagent_type` / OpenCode `subagent` / Kimi·ZCode `subagent_type`)? A bare `task`/prompt item with no role field = **incomplete**, even at **N=1**.
 5. QC initial: **`Execution mode: sdd`** → **N=3**? **`inline`** → **N=1**? Targeted re-review → **N** = Assignment reviewer count?
-6. SDD implement → **serial** (never batch implementers); sticky = **resume** same implementer, not parallel
+6. SDD implement → independent ready tasks isolated and concurrent? Sticky resume limited to one sequential owner track?

package/harness-skills/mstar-host/references/zcode.md

@@ -67,7 +67,7 @@ ZCode C5/C5b SSOT is **this file** — do **not** load `_shared/host-role-bindin
 3. **Skill load list** — instruct the subagent to read `mstar-roles` → `references/<role-id>.md` (or shared reference + parameters) and topic skills per that reference.
 4. **`subagent_type`** — bare Morning Star role id per C5; `general-purpose` fallback.
 
-Paste-only Assignment **without** an invoke call is **not** dispatch. Anti-recursion NEVER: leaf executors are already `Execute as` — no recursive invoke of the same role; Assignment wins (`Delegation: forbidden` unless stated). **Never** multiple implementer invokes in one message for the same plan (SDD serial → **`parallel-dispatch.md`** § SDD implement).
+Paste-only Assignment **without** an invoke call is **not** dispatch. Anti-recursion NEVER: leaf executors are already `Execute as` — no recursive invoke of the same role; Assignment wins (`Delegation: forbidden` unless stated). Independent ready implementers may run concurrently after isolation; scheduling → **`parallel-dispatch.md`** § SDD implement.
 
 ZCode invoke shape (same turn):
 
@@ -119,10 +119,10 @@ Harness **dispatch** on ZCode = **one or more `Agent` tool calls** with correct
 
 Cannot emit required **N** → **`Blocked`**.
 
-### SDD implement (serial)
+### SDD implement
 
-- **`Execution mode: sdd`**: one implementer **`Agent`** per task id (bare role id per C5, `general-purpose` fallback); task reviewer = new **`Agent`** with **Act as `code-reviewer`** (`subagent_type: "code-reviewer"`, `general-purpose` fallback; ZCode L2 review; not qc-specialist*), always with C5b prompt binding — no sticky resume unless host adds it later. Serial rule → **`parallel-dispatch.md`** § SDD implement.
-- **Never** multiple implementer Agents in one message for the same plan.
+- **`Execution mode: sdd`**: one implementer **`Agent`** per task id (bare role id per C5, `general-purpose` fallback); task reviewer = new **`Agent`** with **Act as `code-reviewer`** (`subagent_type: "code-reviewer"`, `general-purpose` fallback; ZCode L2 review; not qc-specialist*), always with C5b prompt binding — no sticky resume unless host adds it later. Ready-task scheduling → **`parallel-dispatch.md`** § SDD implement.
+- Independent ready implementers use isolated parallel tracks; never share a writable worktree or session.
 
 ## Clarify
 

package/harness-skills/mstar-iteration/SKILL.md

@@ -81,7 +81,7 @@ Phase 5: PR merge-ready loop —— 至 mergeable + CI 全绿 + reviews resolved
 - 未知 → 读 `mstar-*`；仅 **`Blocked`**、secrets、不可逆范围缺口、branch metadata 缺失、或 Phase 5 多轮仍 blocked 时升级用户
 - 实际 Git ≠ `working_branch` → **同轮**更新 plan + snapshot + `execution_lease.working_branch`（如适用）
 - **跨 plan implement 并行安全闸**与 **integration merge 串行** → `references/phase-2-worktree-lease.md` §2.0 #5 /「Multi-plan parallelism」（**无论** `Worktree mode: waived`）
-- plan 内 SDD task **串行** — phase-2 reference §2.4、§2.5、`mstar-sdd` Continuous execution
+- plan 内 SDD 独立 ready tasks **并行**，真实依赖与共享写目标串行 — phase-2 reference §2.4、§2.5、`mstar-sdd` Ready-task scheduling
 - **zero-residual（默认）**：单 plan QC findings 尽量当轮清干净；仅真 blocker 才 defer（须 Durable Roadmap）— 见 **`mstar-artifacts`** Findings cleanup modes
 - iteration 命令共享的 PM invariants / preflight / todos / STOP → **`references/command-shared-invariants.md`**
 

package/harness-skills/mstar-iteration/references/phase-1-prepare.md

@@ -145,10 +145,10 @@ Phase 1 与 §1.6 须遵守 **`references/iteration-artifact-boundaries.md`**（
 派发机制 → **`mstar-dispatch-gates`**（specialist review-and-edit dispatch，**顺序链**）。PM **不得**将迭代 harness 文档 commit 到 `spec_integration_branch`，直到：
 
 1. **product-manager** → **architect** → **writing-specialist** 已按序 invoke 编辑 compass、plans、`{SPECS_DIR}/` 与 **`{ITERATION_DIR}/<iteration-id>/`** package（guides/specs，按需）；**不得**在 start 链向 `{KNOWLEDGE_DIR}/` 新增
-2. **writing-specialist** 完成 **corpus hygiene**：全库 `{SPECS_DIR}/` + 既有 `{KNOWLEDGE_DIR}/` 卫生；错放迁回 **`<iteration-id>/`** package；细则 → **`iteration-corpus-hygiene.md`**、**`iteration-artifact-boundaries.md`**
+2. **writing-specialist** 完成 **corpus hygiene**：仅本轮修改的 `{SPECS_DIR}/` / iteration package 与直接相关 knowledge 引用；错放迁回 **`<iteration-id>/`** package；细则 → **`iteration-corpus-hygiene.md`**、**`iteration-artifact-boundaries.md`**
 3. PM 将 compass `status` 设为 `locked`，并确认各 plan 的 Prepare gate（specify / clarify / plan）
 
-**顺序理由**：产品范围与优先级 → 架构与长期契约（specs）→ 行文、规格库卫生与错放纠正（须在 PM/architect 定稿后扫全库 specs）。并行会导致后手重复劳动或覆盖前手未定稿内容。OpenCode：plain role id — **`mstar-host/references/opencode.md`** § Role-mention hygiene。
+**顺序理由**：产品范围与优先级 → 架构与长期契约（specs）→ 行文、规格库卫生与错放纠正（在 PM/architect 定稿后核对受影响文档）。本共享产物链存在真实依赖；独立文档可按 ownership 隔离并行。早期全局探索的既有结果复用，不因每次编辑重新扫全库。OpenCode：plain role id — **`mstar-host/references/opencode.md`** § Role-mention hygiene。
 
 **完成证据** = 磁盘上的 compass / plans / specs / iteration 文档修订 + specs（与既有 knowledge）卫生/归档（如有）+ 索引与 metadata 更新 + compass `status: locked`。**不**要求单独的迭代审查报告——迭代审查的 SSOT 是被编辑的文档本身，无 per-plan QC 式审计链。
 

package/harness-skills/mstar-iteration/references/phase-2-worktree-lease.md

@@ -149,14 +149,14 @@ mismatch → **STOP**.
 2. **Plan start — feature worktree + branch**：创建/校验 dedicated feature worktree（默认 `<repoRoot>/.worktrees/<plan-id>-<slug>`）；Assignment 须含绝对 `Worktree path` + `Working branch`（与 lease 一致）。plan 内多可写并行轨 → **`mstar-branch-worktree`** **`references/parallel-writable-pre-dispatch.md`**
 3. **Implement → InReview**（产品编辑在 feature worktree；plans / snapshot / iterations / SDD 经 control 绝对路径）：
    - **默认 `Execution mode: sdd`**（多 task plan；hotfix 可 `inline`）。
-   - PM 载入 **`mstar-sdd`** 后，按 plan task 顺序 **串行** per-task 循环（**不是**一次派发 dev 做全部 tasks）：
+   - PM 载入 **`mstar-sdd`** 后，按依赖与 ownership 派发 **独立 ready tasks 并行** 的 per-task 循环（**不是**一次派发 dev 做全部 tasks）：
      1. `mstar sdd workspace <plan-id>` → `{SDD_DIR}`
      2. `mstar sdd task-brief <plan-file> N` → `{SDD_DIR}/task-N-brief.md`；记录 `BASE_SHA`
      3. Dispatch **one** implementer subagent（`references/implementer-prompt.md`：brief 路径 + report 路径 + `Model tier`；**禁止**贴整份 plan）
      4. Implementer `DONE` → `mstar sdd review-package BASE HEAD` → task diff 文件
      5. Dispatch **one** task reviewer subagent（brief + report + diff + Global Constraints）
      6. Fix loop 直至 review clean；append `{SDD_DIR}/progress.md`；更新 snapshot plan 行 / plan checkbox
-     7. Next task
+     7. 放行已满足依赖的 next task；不等待无依赖任务，PM 独占共享 progress / snapshot 写入
    - 每次 Completion Report 后更新 snapshot（`workflows/<id>/snapshot.json`）+ 主 plan
 4. **QC → QA gate**（plan 保持 **`InReview`**；**保留** `execution_lease`）：per-plan 审查链 → **`mstar-sdd`**（L1–L2）+ **`mstar-review-qc/references/review-responsibility-boundaries.md`**（L3 tri / inline 单席；raw reports in `{SDD_DIR}/review/`，durable summary in main plan/snapshot）+ **`QA gate`**（`mandatory` → `qa-engineer`；`pm-acceptance` → PM checklist）。**禁止**在 integration merge 成功前设 `Done` 或删除 `execution_lease`。
 5. **Plan complete — serial merge back**（§2.0 #5 未 waive）：自 **control worktree** claim/resume snapshot 顶层 `integration_merge_lease` → 将 plan feature branch 合并入 `spec_integration_branch`（仅 merge-lease holder；细则 → 下方「Integration merge lease」）→ 记录 merge commit 证据 → 释放 merge lease；**同轮**设 `Done` 并删除 `execution_lease`。merge 失败：保持 `InReview` + 保留 lease，不得标 `Done`。
@@ -177,7 +177,7 @@ mismatch → **STOP**.
 
 | 规则 | 说明 |
 |------|------|
-| 串行 | 同一 plan 内 **one implementer at a time**；每 task 后 **one fresh task reviewer** |
+| 并行 | 独立 ready tasks 各自 fresh implementer + 隔离 worktree；单一 canonical per-plan SDD root 内分离 task artifact 路径，context/progress 仅 PM 串行写；leaf 直接消费不可变绝对路径，不调用共享 context helper；每 task 后一位 fresh reviewer；真实依赖与 merge 串行（`mstar-sdd`） |
 | Sticky（可选） | Assignment **`SDD implementer session: sticky`** + `implementer-session.json`；implementer **resume**，reviewer **fresh** — `mstar-sdd/references/sticky-implementer-session.md` |
 | 文件交接 | brief / report / diff / `progress.md` 在 `{SDD_DIR}`；dispatch prompt **只给路径**，不贴 plan 全文或 task 历史 |
 | Assignment 字段 | 每个 implement dispatch 须含 `Execution mode: sdd`、`SDD dir`、`Model tier`；§2.0 #5 未 waive 时还须含绝对 `Worktree path` + verified `execution_lease`；**禁止**省略 `Model tier` |

package/harness-skills/mstar-review-qc/SKILL.md

@@ -13,16 +13,16 @@ description: "Morning Star QC orchestration — **SDD mandatory plan QC tri-revi
 
 ## L3 是什么（派发前对齐）
 
-- Plan QC seats are **reviewers**: whole-branch **diff / logic / risk** lenses — same family as PR review, not a parallel QA test lane.
+- Plan QC seats are **reviewers**: assigned changed **diff / logic / risk** lenses and directly affected interfaces — same family as PR review, not a parallel QA test lane.
 - **Do not** instruct QC in Assignment to “run the suite / build / lint to confirm” on shared tri cwd; that causes peer `Blocked` and collapses L3 into L4.
-- Runtime proof stays with **implementer evidence** and **`QA gate`** (`qa-engineer` or PM acceptance).
+- Runtime proof stays with scoped **implementer evidence** and **`QA gate`** (targeted unit evidence only). `full tri-review` describes seats, not full-repository review. Scope SSOT → **`mstar-harness-core`** § 定向执行与验证边界; QC never broadens exploration or repeats unchanged L2 evidence.
 
 ## 分派时机（与 plan / batch 对齐）
 
 - **`Execution mode: sdd`**：全部 task + L2 task reviewers 完成后 → **强制 tri-review**（`QC mode: full tri-review`，**N=3**）。Assignment 须含 **branch review-package** 路径与 `{SDD_DIR}/review/qcN.md` report paths。PM 汇总 `{SDD_DIR}/review/qc-consolidated.md` 并回写主 plan durable summary。
 - **`Execution mode: inline`**：单席 `qc-specialist` → `{SDD_DIR}/review/qc.md`（**N=1**），或按 hotfix 路由跳过。
 - **After `Request Changes` (default)**：**Targeted re-review** — PM dispatches only seats that **raised** blocking findings; each updates **the same** `{SDD_DIR}/review/qcN.md` (`## Revalidation`, update verdict). **Do not** spawn `qcN-rev2.md` for targeted re-review. Naming → **`mstar-artifacts/references/plan-files-and-reports.md`** § QC 三审触发时机.
-- **Full tri re-review (exception)**：Assignment **`QC re-review: full tri-review`** → new basenames (`qc1-rev2.md` …); PM marks **active wave** in consolidated decision.
+- **Three-seat re-review**：only when all three seats have affected findings. `QC re-review: full tri-review` denotes seat count, never broader scope; each seat still checks its findings and fix delta. New wave basenames may distinguish reports, but do not reopen unchanged review coverage.
 
 > **Engine check (when available):** run `mstar review seats <assignment-file> [--mode sdd|inline|targeted] [--reviewers <role1,role2,...>]` (or `import { executionModeToN, assertTriIdentity } from "@mstar-harness/engine"` in a host hook) to map `Execution mode` to its QC seat count N above and assert tri identity. On `fail` -> do not proceed; fix and re-run. Skill text below remains authoritative when the runtime is absent.
 
@@ -55,7 +55,7 @@ Leaf reviewers apply verdict per **`mstar-roles/references/qc-specialist/report-
 
 ### 覆盖语义（未提及 = 未审查）
 
-- **未提及 = 未审查**：某 finding / severity 项 / 声明未被任何席位报告提及 → 不得在汇总中标记为已解决或通过；如实标注 `unreviewed`，按需转 targeted re-review 或补充席位。
+- **未提及 = 未审查**：某 finding / severity 项 / 声明未被任何席位报告提及 → 不得在汇总中标记为已解决或通过；如实标注 `unreviewed`，仅对受影响项按需转 targeted re-review 或补充席位；不据此重审无关内容。
 - **汇总层零注入**：consolidated 中每条发现可溯源到某 `qcN.md`；PM 不得在汇总层引入席位报告之外的新声明（PM 自身观察走独立 Status Update，不混入 gate 决策输入）。
 - **Unconfirmed 传导**：任一席位 verdict = `Unconfirmed`（`report-template.md` 定义的证据通道失败态）→ gate 决策不得为 `Approve`——先补证据（重发 review-package / 修 diff 基线）再收敛；受影响席位走既有 targeted re-review 机制（同 `qcN.md` `## Revalidation` 原位更新 verdict），不新增 re-review 形态、不改 N 规则。
 

package/harness-skills/mstar-review-qc/references/review-responsibility-boundaries.md

@@ -4,10 +4,10 @@
 
 | Layer | Who | When | Scope | Input |
 |-------|-----|------|-------|--------|
-| **L1** Implementer | dev subagent | Per task | Write code + **run** TDD / verification evidence | `task-N-brief.md` |
+| **L1** Implementer | dev subagent | Per task | Write code + affected unit evidence; non-executable docs/policy may use `scoped-check` | `task-N-brief.md` |
 | **L2** Task reviewer | `code-reviewer` (default; generic fallback when the host agent list lacks it) — PM-dispatched subagent (SDD) | Per task, after implementer | Spec + quality for **one task** (diff-first; no full suite) | brief, report, **task-level** diff |
-| **L3** Plan QC tri (cross-review) | `qc-specialist` + `qc-specialist-2` + `qc-specialist-3` | After **all** tasks on branch | **Code-review seat** — diff, language/logic, security & contract lenses on **whole branch**; **not** the test-execution path | Branch `review-package` MERGE_BASE..HEAD |
-| **L4** QA | `qa-engineer` when **`QA gate: mandatory`**; else PM acceptance | After QC gate | DoD acceptance, residual verify, targeted/full **command** verification, Done recommendation | Review bundle + plan + **L1 evidence** + `status.json` |
+| **L3** Plan QC tri (cross-review) | `qc-specialist` + `qc-specialist-2` + `qc-specialist-3` | After **all** tasks on branch | **Code-review seat** — diff, language/logic, security & contract lenses on **the assigned change and directly affected interfaces**; **not** the test-execution path | Branch `review-package` MERGE_BASE..HEAD |
+| **L4** QA | `qa-engineer` when **`QA gate: mandatory`**; else PM acceptance | After QC gate | DoD acceptance, residual verify, named affected **unit-test** verification only, Done recommendation | Review bundle + plan + **L1 evidence** + `status.json` |
 
 PM sets **`QA gate`** per `mstar-roles/references/project-manager/qa-trigger-matrix.md`. L4 execution when dispatched → `mstar-roles/references/qa-engineer/acceptance-gate.md`.
 
@@ -18,15 +18,17 @@ PM sets **`QA gate`** per `mstar-roles/references/project-manager/qa-trigger-mat
 | Produce runnable test/build evidence | **L1** implementer | **Does not** own |
 | Diff / logic / vulnerability / contract review | **L3** QC (and L2 task reviewer) | **Primary job** |
 | Map DoD → evidence; re-run gaps; close residuals | **L4** QA (or PM acceptance) | **Does not** own |
-| Full / targeted test suites, builds, installs | **L1** and/or **L4** | **NEVER** — leave to QA/dev |
+| Affected unit tests / evidence gaps | **L1** and/or **L4** | **NEVER** — consume evidence |
+| Explicitly user-authorized full local suite | Separate bounded action by implementer/ops; core authorization contract | **NEVER** |
+| Real browser / device / E2E | Explicit independent **`mstar-e2e`** workflow; ops executor | **NEVER**; not an L4 gate |
 
 **Why:** SDD plan QC is **N=3 parallel** on a **shared** `Review cwd`. Build/test/lint toolchains contend for caches and locks and falsely `Blocked` peer reviewers. QC is a **reviewer**, not a second QA lane.
 
-**SDD rule:** Layers L1–L2 run **per task** (serial). Layer L3 is **mandatory full tri-review** (`N=3`) whenever **`Execution mode: sdd`** — single-plan **and** iteration **and** multi-plan iteration. Layer L3 is **not** optional “final single review”.
+**SDD rule:** Layers L1–L2 run **per task**; independent ready tasks use isolated parallel tracks (`mstar-sdd` § Ready-task scheduling). Layer L3 is **mandatory full tri-review** (`N=3`) whenever **`Execution mode: sdd`** — single-plan **and** iteration **and** multi-plan iteration. Layer L3 is **not** optional “final single review”.
 
 **Non-SDD (`Execution mode: inline`):** hotfix / single-stream — plan QC may be **single-seat** (`qc.md`) or skipped per PM routing.
 
-Per-task spec/quality is **done** in L2 before L3. QC seats do not re-derive each task from scratch; they cross-review the **whole branch** for gaps L2 could not see.
+Per-task spec/quality is **done** in L2 before L3. QC seats do not re-derive each task from scratch; they review changed cross-task interfaces for gaps L2 could not see. Full tri denotes three seats, not full scope; do not repeat unchanged L2 work or survey the repository.
 
 ## Plan QC tri (SDD mandatory)
 
@@ -47,7 +49,7 @@ Per-task spec/quality is **done** in L2 before L3. QC seats do not re-derive eac
 | After | Fix dispatch |
 |-------|----------------|
 | Task review Critical/Important | Task fix subagent → task re-review (L2) |
-| Plan QC Critical/Important | **One** fix subagent with **complete** finding list → **targeted** QC re-review (listed seats) |
+| Plan QC Critical/Important | Owned fix assignments, independent ones in parallel; PM retains complete ledger → **targeted** QC re-review (listed seats) |
 
 ## Minor findings
 

package/harness-skills/mstar-roles/SKILL.md

@@ -103,3 +103,5 @@ PM consolidated (tri mode): `{SDD_DIR}/review/qc-consolidated.md` (same folder;
 
 - 角色正文 → `references/<role>.md`（本 skill 内；leaf QC / QA 等子目录见 `references/qc-specialist/`、`references/qa-engineer/`）
 - 全局角色 → `mstar-harness-core` 加载矩阵与专题 skill 索引
+
+- Explicit independent E2E/browser/device requests → `mstar-e2e` (PM orchestrates; `ops-engineer` executes; routine QA does not trigger it).

package/harness-skills/mstar-roles/references/_shared/leaf-executor-core.md

@@ -2,6 +2,15 @@
 
 > Shared by all leaf-executor role references in `mstar-roles/references/`. Each role file references this for the identical Completion Report template, repo-write Git discipline, the shared anti-recursion NEVER section, and plan/documentation rules. **Load selection follows the `mstar-roles` hub § Load Order** (Assignment `Skill presets:` decision): under explicit `none` this boundary plus the role identity carry the load-bearing semantics — no optional topic skill (including `mstar-harness-core`) is required, and `none` never grants delegation or waives gates. Whenever `mstar-harness-core` IS loaded (standard routes, PM rounds, direct topic invocation) it remains the global lifecycle/authority entry. Role-specific NEVER rules, mission, and responsibilities stay in each role file — this file holds only the uniform blocks.
 
+## Assignment scope boundary
+
+Applies under every preset, including explicit `none`. Canonical policy when loaded: `mstar-harness-core` § 定向执行与验证边界.
+
+- Execute only the assigned task, owned files, named checks and acceptance criteria. Read the supplied inputs and relevant knowledge; during implement/fix/QC/QA do not restart whole-repository exploration, review, or scans.
+- Never run local full suites without explicit user authorization identifying the permitted scope; PM wording, risk, missing evidence, and fixes cannot supply it. Full suites belong to CI by default. Do not disguise a full suite as unrelated small checks.
+- Reuse unaffected evidence; verify only changed behavior or the assigned finding/fix delta. QA executes targeted unit tests only; QC runs no test/build/install. Browser/device/E2E belongs to an explicitly requested independent workflow, never routine QA.
+- Stop when the assigned result is evidenced. Do not over-analyze settled questions, invent extra checks, expand into downstream tasks, or repair unrelated findings. Report the concrete missing input/permission to PM if scope is insufficient; preserve completed work.
+
 ## Completion Report
 
 Every leaf executor returns this template (only `**Agent**` and content fields change per role):

package/harness-skills/mstar-roles/references/ops-engineer.md

@@ -23,6 +23,7 @@ If any item below matches, **stop** and return `Blocked` to `project-manager` in
 2. Deploy/runbook execution
 3. Monitoring/alerting integration
 4. Rollback and recovery readiness
+5. Separately requested E2E/browser/device verification — `mstar-e2e` (PM dispatch only; never inferred from routine QA).
 
 ## High-Risk Gate
 
@@ -44,6 +45,8 @@ When assignment is marked `high-risk`:
 
 ## Deliverable Template
 
+For verification-only assignments, use `mstar-e2e` → `references/report-template.md` instead of the Deploy Plan below. The role does not require deployment, production changes, or rollback work when those actions are outside the assignment.
+
 ```markdown
 # Deploy Plan: <release/feature>