release(0.3.61): add ws-megabrain vibe-coding skill

.agents/skills/ws-activity-report/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: ws-activity-report
-version: 0.3.60
+version: 0.3.61
 description: >-
   Timesheet entries (date, start, end, description) for ws-spec-to-pr /
   ws-spec-to-pr-lite deliveries. Start = earliest bootstrap file creation in

.agents/skills/ws-changelog/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: ws-changelog
-version: 0.3.60
+version: 0.3.61
 description: Summarized task history writer — appends concise, structured task completion records to the project changelog file.
 invocation_names:
   - changelog

.agents/skills/ws-check-harness/SKILL.md

@@ -2,7 +2,7 @@
 name: ws-check-harness
 description: Meta-harness integrity auditor — scans routing, links, portability, integrity digests, instruction duplication, role clarity, and skill composition topology.
 disable-model-invocation: true
-version: 0.3.60
+version: 0.3.61
 invocation_names:
   - check-harness
   - ws-check-harness

.agents/skills/ws-check-workflows/SKILL.md

@@ -1,7 +1,7 @@
 ---
 name: ws-check-workflows
 description: Workflow FSM simulation runner — validates step continuity, state isolation, provider dispatch, and artifact transitions across standard, lite, and multi-spec pipelines.
-version: 0.3.60
+version: 0.3.61
 disable-model-invocation: true
 invocation_names:
   - check-workflows

.agents/skills/ws-classify-complexity/SKILL.md

@@ -1,7 +1,7 @@
 ---
 name: ws-classify-complexity
 description: Pipeline complexity classifier — analyzes a spec against config.json dagThresholds and recommends ws-spec-to-pr-lite or ws-spec-to-pr (standard).
-version: 0.3.60
+version: 0.3.61
 invocation_names:
   - classify-complexity
   - ws-classify-complexity

.agents/skills/ws-cleanup/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: ws-cleanup
-version: 0.3.60
+version: 0.3.61
 disable-model-invocation: true
 description: >-
   Lists disposable workflow leftovers (telemetry, .runtime, fix-pr temps,

.agents/skills/ws-code-review/SKILL.md

@@ -1,7 +1,7 @@
 ---
 name: ws-code-review
 description: Local two-phase code review with fix → re-review loops (max 3). Trigger when reviewing a branch/diff before ship, or when orch Step 6 / lite Step 3 runs.
-version: 0.3.60
+version: 0.3.61
 disable-model-invocation: true
 invocation_names:
   - code-review

.agents/skills/ws-configure-project/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: ws-configure-project
-version: 0.3.60
+version: 0.3.61
 description: Project configuration wizard — detects project settings and interviews config.json sections (including preview.dryRunCommand and optional specMemo).
 invocation_names:
   - configure-project

.agents/skills/ws-configure-project/scripts/configure_autoload.py

@@ -49,6 +49,7 @@ def ensure_utf8_stdio() -> None:
     ("ws-changelog", "Every task completion — append-only history"),
     ("ws-fable-method", "Every prompt — structured investigate/act/verify when non-trivial"),
     ("ws-tdah", "Every prompt — action-first shape + judgment"),
+    ("ws-megabrain", "Every prompt — vibe-coding implementer (no spec required); defer when orch owns the session"),
 ]
 
 # Author-machine absolute paths (Windows drive, UNC, POSIX home/opt). Never emit these.

.agents/skills/ws-doctor/SKILL.md

@@ -1,7 +1,7 @@
 ---
 name: ws-doctor
 description: Workflow skills diagnostic inspector — read-only diagnose of path errors, tool/script recipes, config switches, and missing references across installed ws-* skills.
-version: 0.3.60
+version: 0.3.61
 disable-model-invocation: true
 invocation_names:
   - doctor

.agents/skills/ws-fable-domain/SKILL.md

@@ -1,7 +1,7 @@
 ---
 name: ws-fable-domain
 description: Domain adapter generator & schemas — binds domain authority, minimum evidence sets, observation rules, and fraud definitions for DevOps, Data, Research, and Ops.
-version: 0.3.60
+version: 0.3.61
 invocation_names:
   - ws-fable-domain
   - /ws-fable-domain

.agents/skills/ws-fable-judge/SKILL.md

@@ -1,7 +1,7 @@
 ---
 name: ws-fable-judge
 description: Adversarial audit of claimed work against git diffs and re-run verifications. Trigger after claimed completion, during local review, or before merge.
-version: 0.3.60
+version: 0.3.61
 invocation_names:
   - ws-fable-judge
   - /ws-fable-judge

.agents/skills/ws-fable-method/SKILL.md

@@ -1,7 +1,7 @@
 ---
 name: ws-fable-method
 description: 7-step structured problem-solving methodology — evidence-based problem classification, primary-source investigation, surgical action, and adversarial verification for complex tasks.
-version: 0.3.60
+version: 0.3.61
 invocation_names:
   - ws-fable-method
   - /ws-fable-method

.agents/skills/ws-fix-pr/SKILL.md

@@ -1,7 +1,7 @@
 ---
 name: ws-fix-pr
 description: Single-pass PR thread fixer — resolves active GitHub or ADO PR review threads, applying targeted code fixes and posting progress reports.
-version: 0.3.60
+version: 0.3.61
 disable-model-invocation: true
 invocation_names:
   - fix-pr

.agents/skills/ws-goal-fix-pr/SKILL.md

@@ -1,7 +1,7 @@
 ---
 name: ws-goal-fix-pr
 description: PR thread convergence loop — orchestrates iterative fix-pr rounds until all open PR review threads are resolved and checks pass.
-version: 0.3.60
+version: 0.3.61
 disable-model-invocation: true
 invocation_names:
   - goal-fix-pr

.agents/skills/ws-goal-loop/SKILL.md

@@ -1,7 +1,7 @@
 ---
 name: ws-goal-loop
 description: Generic convergence loop primitive — stateful loop engine providing sentinel management, heartbeat/settle timers, and re-check convergence control.
-version: 0.3.60
+version: 0.3.61
 disable-model-invocation: true
 invocation_names:
   - goal-loop

.agents/skills/ws-implement-tasks/SKILL.md

@@ -1,7 +1,7 @@
 ---
 name: ws-implement-tasks
 description: Task implementation & fix executor — builds planned features following task DAGs or applies surgical defect fixes from code review findings.
-version: 0.3.60
+version: 0.3.61
 disable-model-invocation: true
 invocation_names:
   - implement-tasks

.agents/skills/ws-karpathy-guidelines/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: ws-karpathy-guidelines
-version: 0.3.60
+version: 0.3.61
 description: Micro code diff hygiene guidelines — reduces LLM coding mistakes through surgical changes, minimal diff footprints, and surfacing assumptions.
 license: MIT
 invocation_names:

.agents/skills/ws-megabrain/SKILL.md

@@ -0,0 +1,90 @@
+---
+name: ws-megabrain
+version: 0.3.61
+description: >-
+  Vibe-coding task implementer without a spec. Scan dirty work when asked what
+  next; route specialists; consume fable/senior/karpathy/tdah. Trigger on
+  /ws-megabrain, /megabrain, vibe coding, implement without spec, plan mode, or research.
+invocation_names:
+  - ws-megabrain
+  - megabrain
+---
+
+# ws-megabrain
+
+> When this skill is loaded, output "ws-megabrain loaded."
+
+Vibe-coding **task implementer**. Specs are optional. Persona: senior product-engineering manager. Autoload: apply this stance on prompt-driven work; **defer** when `ws-spec-to-pr` / lite / `ws-spec-multi` owns the session. Opt out: `stop ws-megabrain`.
+
+Do **not** copy companion protocols into this file. Load them:
+
+| Need | Load |
+|------|------|
+| Investigate / act / verify / report; `plan`; research (Question) | [`../ws-fable-method/SKILL.md`](../ws-fable-method/SKILL.md) |
+| Surgical diffs | [`../ws-karpathy-guidelines/SKILL.md`](../ws-karpathy-guidelines/SKILL.md) |
+| Extras, ambiguity `user-gate`, Code review proof | [`../ws-senior-developer/SKILL.md`](../ws-senior-developer/SKILL.md) |
+| Reply shape | [`../ws-tdah/SKILL.md`](../ws-tdah/SKILL.md) |
+| MEMORY before mutate | `read-memory` ([`../ws-shared/tools.md`](../ws-shared/tools.md)) |
+
+**Entry:** Expand `{sharedDir}` / `{skillsRoot}` / `{plansDir}` / `{specsDir}` / `{reviewsDir}` from [`../ws-shared/tools.md`](../ws-shared/tools.md). Missing config → defaults (`.agents/plans`, `.agents/specs`), `defaults.autoMode` treated as `false`, and gap `config-missing`. Do not run `ws-spec-write` unless the user asked for a spec.
+
+## Invocation
+
+```text
+/ws-megabrain                 implement (default) — optional what-next menu
+/ws-megabrain plan            fable plan-first (named checks) then STOP
+/ws-megabrain research        fable Question — findings, no edits
+/megabrain --focus {path}
+```
+
+## Specialists (Read at most two, only after mode/gate)
+
+**Router**
+
+| Kind | File | Load when |
+|------|------|-----------|
+| `product` | [`references/PRODUCT.md`](references/PRODUCT.md) | Specs, board, refine |
+| `development` | [`references/DEVELOPMENT.md`](references/DEVELOPMENT.md) | Implement, dirty product files |
+| `review` | [`references/REVIEW.md`](references/REVIEW.md) | Review, harness findings |
+| `delivery` | [`references/DELIVERY.md`](references/DELIVERY.md) | Close, leftovers, incomplete ship |
+
+**Domain** (one, from path/title keywords):
+
+| Kind | File | Hits (any) |
+|------|------|------------|
+| `ddd` | [`references/DDD.md`](references/DDD.md) | domain, aggregate, bounded context |
+| `platform` | [`references/PLATFORM.md`](references/PLATFORM.md) | ci, docker, deploy, pipeline |
+| `performance` | [`references/PERFORMANCE.md`](references/PERFORMANCE.md) | latency, cache, profile |
+| `debug` | [`references/DEBUG.md`](references/DEBUG.md) | race, flaky, dump, root cause |
+| `reverse` | [`references/REVERSE.md`](references/REVERSE.md) | undocumented, legacy, reconstruct |
+| `api` | [`references/API.md`](references/API.md) | rest, graphql, grpc, webhook |
+| `data` | [`references/DATA.md`](references/DATA.md) | schema, migration, sql |
+| `qa` | [`references/QA.md`](references/QA.md) | e2e, coverage, test strategy |
+| `refactor` | [`references/REFACTOR.md`](references/REFACTOR.md) | debt, solid, spaghetti |
+| `frontend` | [`references/FRONTEND.md`](references/FRONTEND.md) | ui, css, a11y, bundle |
+| `distributed` | [`references/DISTRIBUTED.md`](references/DISTRIBUTED.md) | queue, kafka, saga, microservice |
+
+No hit → router only. Never Read the whole folder.
+
+## Steps
+
+1. **Bind host** — `askQuestionTool` from `{sharedDir}/host-capabilities.json` or one probe ([`../ws-shared/tools.md`](../ws-shared/tools.md)). `user-gate` / markdown yield ([`../ws-shared/gates.md`](../ws-shared/gates.md)). `autoMode` (`defaults.autoMode`; missing `config.json` → `false`, interactive `user-gate`) → recommended option 1.
+   - Done when: `askQuestionTool` is a tool name or `none`.
+
+2. **Mode** — `plan` → fable **Plan-First** then STOP. `research` / Question wording → fable **Question** (no Act). Else **Task** (implement, no spec required). Autoload with a clear task → skip the menu. `/ws-megabrain` with no task, or "what next" → Step 3. Orch owns session → stop this skill.
+   - Done when: mode is `plan` | `research` | `implement` | `menu` | `defer`.
+
+3. **Scan + gate** (menu only) — Parallel: `git status --porcelain`; branch; last 8 `git log --oneline`; `{plansDir}/*/*.state.md`; `{specsDir}/index.PRD` if present. Rank (Recommended first): unfinished dirty work → dirty no-plan → next board row → review/cleanup. 2–7 options + **Stop**. Each: evidence, `question`|`task`, specialist ids. Cancel → STOP.
+   - Done when: pick, skip (not menu), or STOP.
+
+4. **Specialists** — `Read` 1–2 kind files for the task (keyword or gated option).
+   - Done when: files in context or none needed.
+
+5. **Execute via companions** — Follow **fable** for this mode (triviality, Evidence, Act, Verify, Report). Karpathy before product edits. Senior proof before claiming done. Do not invent a second loop in this file. Do not start spec-to-pr. Commit/push only if the user asked.
+   - Done when: fable's Done for that shape is met.
+
+## Rules
+
+- Specialists after mode/gate, never during a blind scan.
+- No spec ceremony for vibe implement. Plan mode does not mkdir `{us-dir}` unless the user asked for a workflow.
+- Output language: match the user.

.agents/skills/ws-megabrain/references/API.md

@@ -0,0 +1,25 @@
+# API specialist
+
+Load only when `ws-megabrain` Step 5 selects kind `api`.
+
+## Persona
+
+API and integration specialist.
+
+## Objective
+
+Versioned, documented HTTP/RPC contracts (REST, GraphQL, gRPC as the repo already uses). Webhooks, retries, and third-party adapters that match existing client patterns.
+
+## Pipeline
+
+1. **Contract** — Extend current routes/schemas. No second API style.
+2. **Failure** — Timeouts, retries, and idempotency only where callers already depend on them or the spec requires them.
+3. **Docs** — Update the project's existing API doc or OpenAPI artifact if one exists; do not invent a portal.
+
+## Combine
+
+With `development` or `product`. With `ddd` when the resource is a domain aggregate.
+
+## Output
+
+Endpoint/schema diff plus error and retry behavior named in ACs or tests.

.agents/skills/ws-megabrain/references/DATA.md

@@ -0,0 +1,25 @@
+# DATA specialist
+
+Load only when `ws-megabrain` Step 5 selects kind `data`.
+
+## Persona
+
+Data architect and analytics engineer.
+
+## Objective
+
+Schema, ACID where the DB guarantees it, migrations via configured `verification.migrations*`, and streams only if the stack already has them.
+
+## Pipeline
+
+1. **Source of truth** — Follow current ORM/SQL layout. No extra database.
+2. **Migrate** — Additive first; destructive changes need an explicit spec AC.
+3. **Integrity** — Transactions and constraints that match existing patterns.
+
+## Combine
+
+With `development` or `ddd`. Skip when `stack.database.type` is `none` unless the task is to add a DB (then `product` must own the spec).
+
+## Output
+
+Schema/migration files and the migrate/verify command.

.agents/skills/ws-megabrain/references/DDD.md

@@ -0,0 +1,26 @@
+# DDD specialist
+
+Load only when `ws-megabrain` Step 5 selects kind `ddd`.
+
+## Persona
+
+Domain-driven design architect.
+
+## Objective
+
+Bound contexts, rich domain modeling, isolate business logic from infrastructure, keep a ubiquitous language.
+
+## Pipeline
+
+1. **Context** — Name the bounded context from existing paths/spec, not a new taxonomy.
+2. **Language** — Align symbols with `domain.model` / glossary / current module names. No synonym types.
+3. **Isolate** — Domain rules stay out of hub, UI glue, and IO adapters unless that is already the pattern.
+4. **Model** — Prefer explicit invariants on aggregates over anemic setters. Change consistency boundaries only when the task requires it.
+
+## Combine
+
+Secondary to `development` or `product`. Skip harness-only, cleanup, and pure docs.
+
+## Output
+
+Ubiquitous names in ACs or code; stated context boundary if a new module appears.

.agents/skills/ws-megabrain/references/DEBUG.md

@@ -0,0 +1,25 @@
+# DEBUG specialist
+
+Load only when `ws-megabrain` Step 5 selects kind `debug`.
+
+## Persona
+
+Advanced debugger. Root-cause first.
+
+## Objective
+
+Race conditions, hangs, flaky tests, dumps, and production defects in **this** repo. Reproduce, isolate, fix surgically. Do not reverse-engineer unrelated third-party products.
+
+## Pipeline
+
+1. **Reproduce** — Failing command, log, or test name. Stop if none; ask via `user-gate`.
+2. **Isolate** — Smallest failing path. Check concurrency and shared state.
+3. **Fix** — Minimal change. Add a regression test when the stack has tests.
+
+## Combine
+
+With `development`, `qa`, or `reverse` (undocumented defect). Skip when the issue is missing product scope (use `product`).
+
+## Output
+
+Root cause in one sentence, the fix path, and the regression check.

.agents/skills/ws-megabrain/references/DELIVERY.md

@@ -0,0 +1,25 @@
+# DELIVERY specialist
+
+Load only when `ws-megabrain` Step 5 selects kind `delivery`.
+
+## Persona
+
+Release manager. Close implementation before ship. Stage only what the workflow owns.
+
+## Objective
+
+Unblock close/ship/cleanup: leftover `{plansDir}` dirs, incomplete `shipStatus`, disposable telemetry, or a stuck PR thread batch.
+
+## Pipeline
+
+1. **Close vs ship** — Completed implementation (`status: completed`) before push/PR.
+2. **Stage rules** — Workflow `files_touched` only; never `git add -A`; `{plansDir}` only at close/ship steps.
+3. **Act** — Close/ship/cleanup only inside the blast radius. Stage workflow-owned paths; never `git add -A`. `{plansDir}` only when the user asked to close or commit delivery artifacts.
+
+## Combine
+
+Optional `review` when the blocker is a review thread. Not with `development` unless the option is resume-build (`development` wins; do not Read this file).
+
+## Output
+
+Named next action plus slug/PR pointer. No speculative commits.