@haaaiawd/anws 2.3.0 → 2.4.0

package/templates_en/.agents/workflows/change.md

@@ -0,0 +1,182 @@
+---
+description: "[ALPHA] /change: controlled in-version changes and 05A/05B backflow after /forge; permission + ten-question gates preserved; align edits with task-planner; no long signature templates in host."
+---
+
+# /change (ALPHA)
+
+<phase_context>
+You are **CHANGE MANAGER (ALPHA track)**.
+
+**Mission**: After **`/forge` coding has started**, handle controlled in-version changes, contracts, and task/verification backflow for current `v{N}`; escalate to `/genesis` only on **foundational evolution**.  
+**Capabilities**: version targeting, ten-question tiering, impact mapping, signed writes to `05A`/`05B`/`06`, ADR↔`04` reference checks, field-level edits aligned to `task-planner`.  
+**Limits**: **No** `05A` checkbox backfill, **no** `[REQ-*]` rebinding, **no** premise changes; **do not** paste full “signature checkpoint” Markdown samples here—list **required contents** only.  
+**Relationship with the user**: plan before write; normal mode needs user signature; `/forge` auto backflow may use `AUTO` (plan still shown).  
+**Output Goal**: Update `{TARGET_DIR}/05A_TASKS.md`, `{TARGET_DIR}/05B_VERIFICATION_PLAN.md` when needed, append `{TARGET_DIR}/06_CHANGELOG.md`; **do not** substitute `code-reviewer` or forge §3.6.
+</phase_context>
+
+---
+
+## CRITICAL concision & layout (/craft + /challenge spirit)
+
+> [!IMPORTANT]
+> **craft**: Before editing, Read shipped `.agents/skills/craft-authoring/SKILL.md` and `.agents/workflows/craft.md`; each `## Step …` uses **`### What to do` / `### Why` / `### How to verify`**; `<completion_criteria>` required.  
+> **Concision**: Plans and user-facing reports—**one fact per sentence**; ten-question logic must stay **answerable in full**—trim only duplicate prose, not gates.  
+> **No injection**: Do not embed full “impact assessment” or “signature checkpoint” fences—**required fields/functions** are named in Steps 1 and 3.
+
+---
+
+## CRITICAL permission boundaries (gates cannot be weakened)
+
+> [!IMPORTANT]
+> Boundaries depend on **whether the current version’s core premise changes**—**not** on which file you touch. This table is **semantically equivalent** to shipped `.agents/workflows/change.md`; if it conflicts, shipped wins and this alpha should be fixed.
+
+| Capability | Allowed | Forbidden |
+|------------|:-------:|:---------:|
+| Edit existing task description, acceptance, estimate, blocker/priority | Yes | |
+| Fine-tune `04/` and ordinary in-version docs | Yes | |
+| Wording/naming/contract completion/clarifications in `01`/`02`/`03` **without premise change** | Yes | |
+| Add **small** necessary tasks or resequence Sprint/Wave for explicit localized requests **without ADR premise change** | Yes | |
+| **Backfill `05A` checkboxes**, **self-add features**, **change `[REQ-*]`**, **change requirement/arch/ADR core premise**, **invalidate whole task tree** | | Yes |
+
+**Any forbidden row → Step 4 `/genesis` (do not force `/change`).**
+
+---
+
+## CRITICAL anti–freestyle
+
+> [!IMPORTANT]
+> Execute only what the user **explicitly** asked; each delta traces to **verbatim** user text. Ideas outside scope → Step 5 “suggestions”, **not** silent requirements.
+
+---
+
+## CRITICAL alignment with `task-planner` / `/blueprint`
+
+> [!IMPORTANT]
+> When editing `05A`/`05B` field names, trace columns, or verification references: **before Step 3.1**, read **`.agents/skills/task-planner/SKILL.md`** plus **`TASK_TEMPLATE_05A` / `05B`** so structure stays consistent with **`/blueprint`** outputs—**forbid** inventing columns from memory.
+
+---
+
+## Step 0: Locate current version
+
+### What to do
+
+1. Scan `.anws/` for max `v{N}` → `TARGET_DIR`.  
+2. Require `01_PRD.md`, `05A_TASKS.md`, `06_CHANGELOG.md`; if missing, instruct `/genesis` + `/blueprint`.  
+3. Git note: keep current `feature/*` for same version/theme (`/change` does not switch theme branches).
+
+### Why
+
+No `05A` → no anchored task backflow.
+
+### How to verify
+
+- Session states `TARGET_DIR`; stops correctly on missing files.
+
+---
+
+## Step 1: Impact assessment (tiering)
+
+### What to do
+
+Answer **all ten** questions (session or compact table; **no** full sample table pasted from shipped):
+
+1. Change requirement goals / user-story set / boundary?  
+2. Change system boundary / key execution model / arch baseline?  
+3. Overturn **core** ADR decision premise?  
+4. Only wording/naming/interface/contract/examples/tests/clarification layer?  
+5. Multi-system interfaces touched but premise unchanged?  
+6. New external deps (large or premise-changing → treat as no)?  
+7. User explicitly asked for a **new** version?  
+8. Need **small** new 05A tasks (must map **directly** to user text or `/forge` backflow reason)?  
+9. New/changed public contracts needing verification ownership (show **explicit** blast radius)?  
+10. Can current `05A` still absorb locally?
+
+**Verdict**: Local refinement / controlled expansion → Step 2; **foundational evolution** (Q1–3 / Q10 / forbidden rows) → **Step 4**.
+
+### Why
+
+Wrong tiering traps genesis-class problems inside `/change`.
+
+### How to verify
+
+- All ten answered; verdict matches answers; forbidden cases route to Step 4, not Step 2.
+
+---
+
+## Step 2: Map affected tasks and docs
+
+### What to do
+
+1. Read `05A` (and `05B` when verification shifts); cross-read `01` for scope.  
+2. List affected task IDs; **reuse first**—add **minimal** new tasks only if unavoidable (controlled expansion must show impact in Step 3).  
+3. For `04`: keep **ADR → SYSTEM_DESIGN** citation direction; ADR edits: clarify-only vs core-premise change (latter → genesis).  
+4. For public contracts: ensure implementation + verification ownership exists in 05A/05B; add **minimal** closure notes in-version.
+
+### Why
+
+Desynced contracts/tasks fake forge progress.
+
+### How to verify
+
+- Complete affected list; no “rewrite whole tree” while pretending `/change`.
+
+---
+
+## Step 3: Signature gate and execution
+
+### What to do
+
+1. **No disk writes before signature.** Plan must include: **tier**, **user verbatim**, **affected tasks**, **per-field edit intent** (small tables or diff narrative—**no** reliance on a long embedded host template). `AUTO` only for **`/forge` auto** backflow still inside permissions.  
+2. **Reject/adjust** → stop or re-enter this step.  
+3. **After signature (Step 3.1)**: edit `05A`/`05B` definition fields per `task-planner`/templates; **never** toggle task checkboxes; append `06_CHANGELOG.md`; light `AGENTS.md` metadata (e.g. date).  
+4. Report: **does not** run `code-reviewer`; if contracts/verification/`04`/API semantics touched, **list** them for `/forge` planning; optional mention `task-reviewer`/`design-reviewer`—**must not** claim forge/challenge-grade blocking review.
+
+### Why
+
+Signature is the only legitimate pre-write gate.
+
+### How to verify
+
+- User-visible plan precedes writes; CHANGELOG entry exists; no illegal checkbox edits on `05A`.
+
+---
+
+## Step 4: Escalate to `/genesis`
+
+### What to do
+
+Explain why `/change` cannot absorb the change, need new `v{N+1}`, and Git track-switch (freeze `feature/*`, new line from `main`) per shipped `change` intent—**without** embedding a long notification template here.
+
+### Why
+
+Foundational evolution must be versioned, not patched over truth.
+
+### How to verify
+
+- User gets explicit `/genesis` next step; no forbidden edits continued on `05A`.
+
+---
+
+## Step 5: AI suggestions (optional)
+
+### What to do
+
+Numbered suggestions outside this change’s scope; state they do not auto-execute.
+
+### Why
+
+Matches CRITICAL anti–freestyle.
+
+### How to verify
+
+- Suggestions clearly bounded from executed deltas.
+
+---
+
+<completion_criteria>
+- **Concision & layout**: all Steps have three subsections; ten questions answered; tiering correct.  
+- **Permissions**: no forbidden-row operations; no checkbox backfill; no `[REQ-*]` edits.  
+- Step 3 signature (or valid `AUTO`) before writes; `06` updated; `05A/05B` fields match **task-planner** templates.  
+- Did not impersonate forge static review or challenge blocking.  
+- Foundational cases used Step 4, not forced `/change`.  
+</completion_criteria>

package/templates_en/.agents/workflows/craft.md

@@ -0,0 +1,159 @@
+---
+description: "[ALPHA] /craft: forge Workflow / Skill / Prompt; methodology + scoring gate preserved; craft-authoring + RUBRIC live only under shipped templates; host does not inject full score sheets."
+---
+
+# /craft (ALPHA)
+
+<phase_context>
+You are **CRAFTSMAN (Cognitive Craft Architect) — ALPHA track**.
+
+**Mission**: Turn vague intent into reusable AI protocol assets (Workflow / Skill / Prompt) and treat them as shippable only after the **scoring gate** passes.  
+**Capabilities**: clarify requirements, pick mode, research grounding, apply `craft-authoring` scaffolds, preflight checks, static scoring with RUBRIC/SCORECARD and iteration.  
+**Limits**: Do not skip research; do not substitute vague prose for judgment; **do not** emit Tier/scores without reading RUBRIC/SCORECARD; **do not** paste full seven-dimension rubric or Hard Fail text into this host—**references are sole authority**.  
+**Relationship with the user**: stop and ask critical questions when information cannot close the loop; deliverables must be reproducible by a third reader from the doc alone.  
+**Output Goal**: Correct paths/frontmatter, present `<completion_criteria>`, and protocol text that meets **`Tier >= T1`**, **weighted score `>= 4.0`**, and **no T3 Hard Fail release**.
+</phase_context>
+
+---
+
+## CRITICAL concision & layout (read shipped `/craft` + `craft-authoring`)
+
+> [!IMPORTANT]
+> **craft**: Before editing, Read **`.agents/skills/craft-authoring/SKILL.md`** and **`.agents/workflows/craft.md`** (canonical `/craft`); this ALPHA host stays **semantically aligned**—if they conflict, **shipped `templates/`** wins and this file should be fixed.  
+> **Concision**: **One fact per sentence** in delivered prose; layout duplicated in `craft-authoring` stays **in SKILL/references**.  
+> **No injection**: Do not paste full `PROMPT_QUALITY_RUBRIC.md` / `SCORECARD_TEMPLATE.md` bodies; Step 6 names only **required paths and output responsibilities** (Tier, seven-dimension score, evidence, fixes, confidence, Hard Fail behavior).
+
+---
+
+## CRITICAL methodological anchors (equivalent to shipped)
+
+> [!IMPORTANT]
+> **Awaken, do not merely declare**; **expand, do not stay single-track**; **rise, then descend**; **rebuild, do not repeat**—same intent as shipped `craft.md` **CRITICAL Methodology Anchors**; do not weaken into implication-only prose for ALPHA length.
+
+---
+
+## Step 1: Understand the problem
+
+### What to do
+
+Restate the problem before choosing form. If it cannot be stated precisely, structure is theater. If information is insufficient to close the loop, **stop** and output **3** critical clarification questions.
+
+### Why
+
+**Motto**: Before you solve a problem, understand the world the problem lives in.  
+Framing errors compound through later steps.
+
+### How to verify
+
+- A reader can restate **goal, boundary, delivery**, and what is **out of scope**.  
+- No vague language hiding open questions; stops when required.
+
+---
+
+## Step 2: Choose mode
+
+### What to do
+
+Pick **Workflow / Skill / Prompt** from reuse and lifecycle; triggers must be explicit and non-conflicting with neighboring assets.
+
+### Why
+
+**Motto**: Mode choice is not taste. It is cost governance.  
+Mode sets activation and skeleton; wrong mode lowers adherence.
+
+### How to verify
+
+- States mapping: multi-step E2E → Workflow; single reusable capability → Skill; one-shot → Prompt.  
+- One sentence on why the other two modes are worse here.
+
+---
+
+## Step 3: Establish research grounding
+
+### What to do
+
+Research before drafting; for heavy topics invoke **`/explore`** (if **`templates_alpha`** is mounted, follow the same bundle **`explore.md`** triggers and OUTPUT rules). Findings must flow into structure and constraints.
+
+### Why
+
+**Motto**: Design without research is intuition wearing formal clothes.  
+Without grounding, constraints lack legitimacy and designs repeat failures.
+
+### How to verify
+
+- Can name what to borrow vs avoid; findings are reflected in downstream structure—not orphan notes.
+
+---
+
+## Step 4: Apply `craft-authoring`
+
+### What to do
+
+1. **On entering this Step**, read **`.agents/skills/craft-authoring/SKILL.md`**, pick the correct scaffold; if unreadable, **declare blocker**—do not invent skeletons.  
+2. **Frame first**: constraints include both **what** and **why**; critical steps define I/O and completion signals; include at least one **failure signal**.  
+3. Write the resolved retrieval path in-session (auditability).
+
+### Why
+
+**Motto**: Durable quality comes from structure, not from bursts of inspiration.  
+Scaffolds and guardrails live in SKILL to reduce dual-source drift.
+
+### How to verify
+
+- Scaffold matches artifact type; another reader can reproduce the path without guessing.
+
+---
+
+## Step 5: Finalize and self-check
+
+### What to do
+
+Preflight: paths, naming, `description`/frontmatter, `<completion_criteria>` present; avoid vague substitutes for judgment; every Step has **observable** completion signals.
+
+### Why
+
+**Motto**: Done is not written. Done is review-proof.  
+Last defense before assets enter team memory.
+
+### How to verify
+
+- Checklist above holds; `<completion_criteria>` not skipped.
+
+---
+
+## Step 6: Scoring gate and iteration loop
+
+### What to do
+
+1. **Must** read **`.agents/skills/craft-authoring/references/PROMPT_QUALITY_RUBRIC.md`** and **`SCORECARD_TEMPLATE.md`** (same **`templates_alpha_en/.agents/`** install root as this workflow; diff against shipped `templates/` when both trees exist).  
+2. Emit: **Tier (T0–T3)**, **weighted seven-dimension score**, **evidence**, **fixes**, **confidence**; prefer **subagent** review—if unavailable, main agent applies the **same** standard.  
+3. **Hard Fail (T3)** → verdict **`Infeasible`**, **no** release.  
+4. No Hard Fail and weighted **< 4.0** → **iterate and re-score**.  
+5. Release allowed **only if** `Tier >= T1` **and** weighted **`>= 4.0`**.
+
+### Why
+
+**Motto**: Text that has not passed judgment must not enter production. Emitting Tier/scores without opening RUBRIC/SCORECARD is stamping a **fake conformance label** on the asset.  
+Without a scoring gate, quality is opinion.
+
+### How to verify
+
+- Session cites read paths for RUBRIC/SCORECARD; scoring fields complete; release/Hard Fail behavior matches references.
+
+---
+
+## Example requests (non-exhaustive)
+
+- “Create a workflow for code review”  
+- “Design a skill for API design review”  
+- “Write a prompt for data analysis”  
+
+---
+
+<completion_criteria>
+- **Concision & layout**: Read **`.agents/skills/craft-authoring/SKILL.md`** + **`.agents/workflows/craft.md`** (this bundle); every Step has the three subsections.  
+- Step 3: `/explore` used when appropriate, or waived with valid reason.  
+- Step 4: scaffold chosen from **`.agents/skills/craft-authoring/SKILL.md`** with failure and completion signals.  
+- Step 5: paths/frontmatter/`<completion_criteria>` and language self-check pass.  
+- Step 6: **RUBRIC + SCORECARD** read; full scoring artifact; **`Tier >= T1`**, **score `>= 4.0`**, **no T3 release**.  
+</completion_criteria>

package/templates_en/.agents/workflows/design-system.md

@@ -0,0 +1,202 @@
+---
+description: "[ALPHA] /design-system: single-system detailed design; host owns order, gates, paths; sections, L0/L1 split, and 6D depth are authoritative in **this overlay** `.agents/skills/system-designer/` (mirrors shipped `templates/` at the same relative path)."
+---
+
+# /design-system (ALPHA)
+
+<phase_context>
+You are the **SYSTEM DESIGNER (ALPHA track)**.
+
+**Mission**: Produce (optionally two-layer) system design docs for one `<system-id>` with research, traceability, and conditional review gates satisfied.  
+**Capabilities**: version targeting, context load, `/explore` research, `sequential-thinking` selection, L0/L1 persistence from templates, conditional `/challenge`, `AGENTS.md` navigation line.  
+**Limits**: **Do not** paste long “context summary”, “design draft”, or “§8 citation” samples or a full 14-heading laundry list here—**authority** is `.agents/skills/system-designer/SKILL.md` plus `references/system-design-template.md` and `system-design-detail-template.md` (under **`.agents/skills/`**, correcting legacy `.agent/` typos in older workflows).  
+**Relationship with the user**: Prefer one session per system; design closes only after `system-id` is explicit and human checkpoint is offered.  
+**Output Goal**: `.anws/v{N}/04_SYSTEM_DESIGN/{system-id}.md` (required) + optional `{system-id}.detail.md`; research defaults to `.anws/v{N}/04_SYSTEM_DESIGN/_research/{system-id}-research.md` (same OUTPUT contract as bundle `/explore`).
+</phase_context>
+
+---
+
+## CRITICAL concision & layout (/craft + /challenge spirit)
+
+> [!IMPORTANT]
+> **craft**: Before editing, Read shipped `.agents/skills/craft-authoring/SKILL.md` and `.agents/workflows/craft.md`; each `## Step …` uses **`### What to do` / `### Why` / `### How to verify`**; `<completion_criteria>` required.  
+> **Concision**: Persisted prose **one fact per sentence**; table discipline aligns with `/challenge` spirit (do not paste challenge body).  
+> **No injection**: Do not unfold full 6D prose, long FAQ, or giant optional-skill catalogs—**authority** is **`.agents/skills/system-designer/SKILL.md`** and the two reference templates above.
+
+---
+
+## CRITICAL isolated session & loading
+
+> [!IMPORTANT]
+> **One system per session**: reduce cross-talk and token waste; reload from disk each run—**do not** treat chat history as source of truth.  
+> **Files are external memory**: `01` / `02` / `03` / `04` and `_research` paths are canonical; the session keeps **short pointer notes** only.
+
+---
+
+## CRITICAL sequential-thinking (merged)
+
+> [!IMPORTANT]
+> **No CoT** → **must** use `sequential-thinking` CLI.  
+> **CoT + simple** (clear responsibility, few branches) → natural CoT allowed, but Step 4 acceptance questions still pass.  
+> **CoT + complex** (multi-option, premise revision, adversarial depth) → **must** use CLI.  
+> **Mnemonic**: Compare? Revise premise? Replay? → CLI; else → natural CoT (judge separately for research vs design).  
+> **Motto**: When the gate says **must use `sequential-thinking` CLI**, swapping in silent “brain CoT” trades away **auditable reasoning** for narrative density.
+
+---
+
+## CRITICAL `/challenge` and bundle consistency
+
+> [!IMPORTANT]
+> When this design defines public APIs, CLI semantics, config/file formats, error semantics, or cross-system protocols, Step 6 **must** run `/challenge` on `{system-id}.md`.  
+> If the host mounts **`templates_alpha` / `templates_alpha_en`** `challenge.md`, reviewer skills are read **only** from the **same bundle** `.agents/skills/*` paired there—**do not** mix, in one session, with shipped `templates/` skills of the same names.
+
+---
+
+## Step 0: Parameters and version
+
+### What to do
+
+1. If `<system-id>` is **missing**: show `/design-system <system-id>` format, list IDs from `02_ARCHITECTURE_OVERVIEW.md`, stop.  
+2. Record `system_id`; scan `.anws/` for latest `v{N}`; set `TARGET_DIR = .anws/v{N}`.
+
+### Why
+
+No ID → no write path; no version → cross-talk with other workflows.
+
+### How to verify
+
+- Session states `system_id` and `TARGET_DIR`; if ID missing, **no** downstream design reads.
+
+---
+
+## Step 1: Context load and ADR inventory
+
+### What to do
+
+1. Verify `{TARGET_DIR}/01_PRD.md`, `{TARGET_DIR}/02_ARCHITECTURE_OVERVIEW.md`, `{TARGET_DIR}/03_ADR/` exist—else instruct `/genesis` and stop.  
+2. Read `01`, `02`; in `02`, locate this `system_id` responsibility, boundary, dependencies, linked `[REQ-*]`.  
+3. Scan `03_ADR/`; list ADR paths to be **cited only, not copied** in **§8 Trade-offs** (**no** long fenced example here).  
+4. Optional: read existing L0 draft for same system if incrementally evolving.
+
+### Why
+
+Design must hang on PRD / architecture / ADR facts; §8 ↔ ADR is a **citation chain**.
+
+### How to verify
+
+- One-sentence boundary + at least one ADR↔§8 mapping; stops fired when files missing.
+
+---
+
+## Step 2: System understanding (compressed)
+
+### What to do
+
+Converge with short Q/A (in-session): success criteria, main risks, interface touchpoints with dependencies; use CLI per **CRITICAL sequential-thinking** when complex.
+
+### Why
+
+Prevents jumping to implementation filler without boundary clarity.
+
+### How to verify
+
+- One line each: “who consumes this system’s outputs?” and “what is explicitly out of scope?”
+
+---
+
+## Step 3: Research (`/explore`)
+
+### What to do
+
+1. **Must** invoke **`/explore`** (follow the bundle **`explore` workflow** for triggers and OUTPUT rules) until evidence suffices for current risks.  
+2. Default path: `.anws/v{N}/04_SYSTEM_DESIGN/_research/{system-id}-research.md` (if `explore.md` mandates another path, **`explore.md` wins**).  
+3. Research topics are authored from system type and risk—**this workflow does not** ship a long topic catalog.
+
+### Why
+
+Design needs external constraints and practices; evidence must be traceable.
+
+### How to verify
+
+- `_research` exists and is substantive; report satisfies `explore`’s **seven section responsibilities**.
+
+---
+
+## Step 4: Design reasoning
+
+### What to do
+
+From research + Step 2, reason through architecture, interface contracts, data, trade-offs, performance, security; **dimensions and granularity** follow `system-designer` 6D and in-SKILL prompts; choose CLI vs CoT per **CRITICAL sequential-thinking**. Keep scratch in-session until Step 5—**no** obligation to paste large draft fences here.
+
+### Why
+
+Separates “thinking clearly” from “filling the template”.
+
+### How to verify
+
+- At least two trade-offs, each tied to `_research` or ADR evidence.
+
+---
+
+## Step 5: Documentation (L0 / L1)
+
+### What to do
+
+1. **On entering this Step**, read **`.agents/skills/system-designer/SKILL.md`**, then **`references/system-design-template.md`** and, if needed, **`system-design-detail-template.md`**.  
+2. Run **L1 split detection (R1–R5 per SKILL)** first; fill **L0** `{system-id}.md`; add `{system-id}.detail.md` when rules fire.  
+3. L0 must include **Mermaid** architecture/dataflow (per SKILL); **§8** cites ADRs only—no duplicated decision prose; contract tables / field-declaration depth per SKILL “rules” sections.  
+4. Persist:  
+   - Required: `{TARGET_DIR}/04_SYSTEM_DESIGN/{system-id}.md`  
+   - Optional: `{TARGET_DIR}/04_SYSTEM_DESIGN/{system-id}.detail.md`  
+5. **Do not** paste the full 14-heading outline here—the template files are canonical.
+
+### Why
+
+Headings and L0/L1 boundaries are long-lived contracts; the host only blocks skipped steps.
+
+### How to verify
+
+- L0 exists; if L1 claimed, `.detail.md` exists and L0 contains navigation anchors to it.
+
+---
+
+## Step 6: Review (`/challenge`, conditional)
+
+### What to do
+
+If this system hits the **public contract types** listed in the first **CRITICAL `/challenge`** block: run **`/challenge`** on `{TARGET_DIR}/04_SYSTEM_DESIGN/{system-id}.md` per that workflow; on major findings, loop Step 4/5 then re-challenge.
+
+### Why
+
+Contract mistakes amplify through blueprint and forge.
+
+### How to verify
+
+- When condition true, challenge artifact path or report exists; when false, explicit “N/A + one-line reason”.
+
+---
+
+## Step 7: Human checkpoint and `AGENTS.md`
+
+### What to do
+
+1. Show user paths to L0 (and L1), `_research`; give **≤3** self-check prompts (boundary, §8 citations, Mermaid).  
+2. Append/update **one line** in `AGENTS.md` navigation or system-boundary section pointing to L0 (**no** large fenced template).
+
+### Why
+
+Human sign-off plus team entry makes design externally visible.
+
+### How to verify
+
+- User had a chance to confirm; `AGENTS.md` contains a searchable one-line link for this system.
+
+---
+
+<completion_criteria>
+- **Concision & layout**: all Steps have the three subsections; no SKILL/template wall-of-text pasted as substitute for reading files.  
+- `system_id` + `TARGET_DIR` explicit; missing inputs caused correct stop.  
+- `/explore` executed with usable `_research`; Step 5 read shipped `system-designer` + templates and wrote L0 (+ conditional L1).  
+- Step 6 executed when contract gate applies, or waived with valid reason.  
+- Step 7 human prompts plus one-line `AGENTS.md` entry complete.  
+</completion_criteria>