fix(skills): remove Agent Space references + sync WDS May improvements

- Remove hardcoded Supabase URLs from start.md, wrap.md, handoff.md
  Handoffs now write to progress/[target_agent].md (file-based, no
  external dependency). Resumption via /[agent] progress/[agent].md.
- Add src/data/wds-glossary.md — locked terminology for all agents
- Add src/data/agent-contracts.md — domain boundaries, handoff rules,
  prerequisites, quality rules, out-of-scope per agent
- Add src/data/shared-activation.md — common startup steps referenced
  by all agents instead of repeated inline
- Fix src/tools/memory/SKILL.md: rename to wds-memory, remove WDS-E
  and external storage references

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

Author: MartenAngner

src/data/agent-contracts.md

@@ -0,0 +1,72 @@
+# WDS Agent Contracts
+
+Defines what each agent owns, what they explicitly do not own, and how they hand off to each other.
+All agents load this file at activation. These rules are non-negotiable.
+
+---
+
+## Domain Boundaries
+
+| Agent | Owns | Does NOT own |
+|-------|------|--------------|
+| **Saga** | Phases 0–2: Alignment, Product Brief, Trigger Mapping | Any design work. Any code. Scenarios (Phase 3+). |
+| **Freya** | Phases 3–4: UX Scenarios, UX Design. Phases 6–7: Asset Generation, Design System. | Discovery (Phases 1–2). Any code. PRDs. |
+| **Mimir** | Phase 5: Tech Audit, PRD, Build. Phase 8: Product Evolution. | Discovery. Design. Writing specs without a Work Order. |
+
+**If a user asks an agent to do work outside its domain:** name the right agent and offer to hand off. Never attempt the work yourself.
+
+---
+
+## Prerequisites
+
+Each agent requires the following before starting core work:
+
+| Agent | Required | Blocks |
+|-------|----------|--------|
+| Saga | Nothing | — |
+| Freya | `A-Product-Brief/product-brief.md` + `B-Trigger-Map/00-trigger-map.md` | Cannot design without strategic foundation |
+| Mimir | At least one Work Order from Freya | Cannot build without a WO. Cannot PRD without a WO. |
+| Mimir (existing codebase) | `E-Development/000-tech-audit.md` | Cannot PRD without knowing the codebase |
+
+---
+
+## Handoff Rules
+
+**Saga → Freya**
+Trigger: Product Brief and Trigger Map are complete and aligned.
+Action: Saga runs `/wrap freya`. Freya picks up with `/freya progress/freya.md`.
+Never: Saga does not write scenarios or design anything before handing off.
+
+**Freya → Mimir**
+Trigger: Work Order written, page spec complete, ready for implementation.
+Action: Freya runs `/wrap mimir` or `/handoff mimir`. Mimir picks up the Work Order.
+Never: Freya does not write code. Freya does not write PRDs.
+
+**Mimir → Freya**
+Trigger: Implementation complete, browser-verified. Or: blocked on design decision.
+Action: Mimir runs `/handoff freya` with the specific question or completion note.
+Never: Mimir does not modify specs or Work Orders. He implements what they say.
+
+---
+
+## Quality Rules (all agents)
+
+- **One task at a time.** Complete and verify before moving on.
+- **No plausible-looking wrong output.** If you cannot follow the template exactly, stop and say so. Wrong-but-plausible output breaks every downstream phase.
+- **Read the template before writing.** Every artifact has a template. Load it, follow it.
+- **Decisions are documented.** Any deviation from a template or unexpected choice goes in the design log.
+
+---
+
+## Out-of-Scope (explicit)
+
+Things no WDS agent does, ever:
+
+- Produce output in a custom format when a WDS template exists
+- Write to `progress/` without going through the memory tool
+- Commit without a meaningful message (conventional commits required)
+- Force push, skip hooks, or bypass git safety
+- Start a new phase without the prerequisite documents
+- Write code without a PRD (Mimir only)
+- Mark a requirement done without browser verification (Mimir only)
+- Design without a Trigger Map (Freya only)

src/data/shared-activation.md

@@ -0,0 +1,39 @@
+# WDS Shared Activation Steps
+
+Common startup sequence for all WDS agents (Saga, Freya, Mimir).
+Each agent's SKILL.md references this file instead of repeating these steps.
+
+---
+
+## Step: state
+
+Check for session state via the memory tool.
+Read `~/.claude/wds/src/tools/memory/SKILL.md` and follow the `load` operation for the current agent_id.
+If state found: show resume prompt. Wait for user response before continuing.
+
+---
+
+## Step: scan
+
+Scan workspace for WDS projects:
+- Find repos with `_progress/wds-project-outline.yaml` or `_progress/00-design-log.md`
+- Skip system repos (bmad-method-wds-expansion, whiteport-design-studio)
+- For each project: read design log, note phase status and in-progress work
+- Also check current directory for design process folders (A-Product-Brief/ through E-Development/) and any context documents at repo root
+
+---
+
+## Step: select
+
+IF multiple projects found with open work:
+  List them, ask which to work on.
+IF single project:
+  Continue to agent-specific activation.
+
+---
+
+## Step: brownfield-detect
+
+Check if the project has a codebase (src/, backend/, storefront/, app/, or similar code folders at repo root).
+IF codebase found → go to agent-specific brownfield handling.
+IF no codebase → continue to agent-specific greenfield flow.

src/data/wds-glossary.md

@@ -0,0 +1,98 @@
+# WDS Glossary
+
+Locked terminology for all WDS agents. One definition per term — no synonyms, no aliases.
+Agents load this file once at activation. Do not redefine these terms locally.
+
+---
+
+## Phases
+
+| Phase | Name | Owner |
+|-------|------|-------|
+| 0 | Alignment & Signoff | Saga |
+| 1 | Product Brief | Saga |
+| 2 | Trigger Mapping | Saga |
+| 3 | UX Scenarios | Freya |
+| 4 | UX Design | Freya |
+| 5 | Agentic Development | Mimir |
+| 6 | Asset Generation | Freya |
+| 7 | Design System | Freya |
+| 8 | Product Evolution | Mimir |
+
+---
+
+## Output Folder Structure
+
+```
+{output_folder}/
+├── A-Product-Brief/       Phase 1 — strategic foundation
+├── B-Trigger-Map/         Phase 2 — user research & personas
+├── C-UX-Scenarios/        Phase 3 — journey flows
+├── D-UX-Design/           Phase 4 — page specifications & design assets
+└── E-Development/         Phase 5 — technical requirements, work orders, code
+```
+
+Progress files (machine-local, not committed):
+```
+progress/
+├── [agent].md             Session state per agent
+└── project-index.md       Living artifact index, updated on wrap
+```
+
+---
+
+## Artifacts
+
+### Strategy (Phase 1)
+- **Product Brief** — `A-Product-Brief/product-brief.md`. Strategic foundation: vision, goals, constraints, target users. Required before any design work.
+- **Content Language** — `A-Product-Brief/content-language.md`. Tone, vocabulary, brand voice.
+- **Visual Direction** — `A-Product-Brief/visual-direction.md`. Aesthetic references, colour, typography intent.
+
+### Research (Phase 2)
+- **Trigger Map** — `B-Trigger-Map/00-trigger-map.md`. User psychology mapped to business goals. Required before UX Scenarios.
+- **Business Goals** — `B-Trigger-Map/01-business-goals.md`. Measurable outcomes, KPIs.
+- **Persona** — `B-Trigger-Map/NN-persona-[firstname]-the-[archetype].md`. Alliterative names required (e.g. Harriet the Hairdresser).
+- **Feature Impact** — `B-Trigger-Map/feature-impact.md`. Feature × persona × trigger mapping.
+
+### Design (Phases 3–4)
+- **UX Scenarios** — `C-UX-Scenarios/00-ux-scenarios.md`. User journey flows derived from Trigger Map.
+- **Page Spec** — `D-UX-Design/[page-name].md`. Per-page specification: layout, content, interactions, acceptance criteria.
+- **Design Tokens** — Extracted progressively during Phase 4, not upfront.
+
+### Development (Phase 5)
+- **Tech Audit** — `E-Development/000-tech-audit.md`. Living architecture document. Required before any PRD on an existing codebase.
+- **Master PRD** — `E-Development/000-PRD.md`. Platform requirements, written once, updated as project evolves.
+- **Feature PRD** — `E-Development/NNN-[feature].xml`. One per Work Order.
+- **Change Order** — `E-Development/NNN-NN-[slug].xml`. Feedback/change against a parent PRD.
+- **Work Order** — `E-Development/WO-NNN-[slug].md`. Task written by Freya for Mimir. Contains: objective, scope, files, acceptance criteria.
+- **Mimir Brief** — Narrative handoff document from Freya to Mimir when handing off design work.
+
+### Progress (machine-local)
+- **Design Log** — `_progress/00-design-log.md`. Project-wide progress, updated each session.
+- **Project Outline** — `_progress/wds-project-outline.yaml`. Phase status, project metadata.
+- **Session State** — `progress/[agent].md`. Agent-specific session state. Loaded by `/start`, written by `/wrap`.
+- **Project Index** — `progress/project-index.md`. Living index of all artifacts, updated by `/wrap`.
+
+---
+
+## Patterns
+
+- **Design Loop** — Freya's per-page cycle: discuss → spec → wireframe → approve → iterate → update spec → implement → browser review → extract tokens.
+- **Dream Up Mode** — Three modes for artifact generation: Dialog (collaborative), Suggest (agent proposes), Dream (agent generates fully). Selected at session start.
+- **Brownfield** — Project with an existing codebase. Triggers gap-map assessment before new work begins.
+- **Greenfield** — Project with no existing codebase. Follows standard phase progression.
+- **Gap Map** — Freya's cross-reference of what is designed vs what is built vs what has a Work Order.
+
+---
+
+## Model Selection
+
+| Task type | Model |
+|-----------|-------|
+| Any code, build, deploy, implement | Opus |
+| High-stakes / production / compliance | Opus |
+| Long or complex multi-step tasks | Opus |
+| Strategy, spec, dialog, UX, analysis | Sonnet |
+| Simple, low-stakes, short | Haiku |
+
+Default: lightest model that fits. Prefix Next actions with `MODEL:[Haiku|Sonnet|Opus]`.

src/skills/handoff.md

@@ -5,9 +5,6 @@ Pass a specific piece of work to another WDS agent. This is NOT a session wrap 
 **Usage:** `/handoff [target-agent]`
 **Example:** `/handoff mimir`
 
-> **Handoffs go through Agent Space — never as files on disk.**
-> Agent Space is the single source of truth for cross-agent communication. If it's not available, fix connectivity — do not fall back to writing files.
-
 ---
 
 <handoff-steps>
@@ -16,7 +13,7 @@ Pass a specific piece of work to another WDS agent. This is NOT a session wrap 
     - Derive everything from the conversation. Do NOT ask questions.
     - Do NOT summarize this session. That is a wrap, not a handoff.
     - Focus only on what the receiving agent needs to start the specific task immediately.
-    - The sub-agent handles Agent Space delivery. You only compile and show.
+    - Handoff is written to `progress/[target_agent].md` — the receiving agent picks it up via `/start`.
   </constraints>
 
   <step id="1-compile">
@@ -52,49 +49,43 @@ Pass a specific piece of work to another WDS agent. This is NOT a session wrap 
     Then proceed immediately to step 3.
   </step>
 
-  <step id="3-subagent">
+  <step id="3-write">
     Spawn a sub-agent with this exact prompt — substitute the bracketed values:
 
     ---
-    You are a delivery agent. Your only job is to post a handoff to Agent Space and return the token.
-
-    Send this request:
-
-    ```bash
-    curl -s -X POST "https://uztngidbpduyodrabokm.supabase.co/functions/v1/agent-messages" \
-      -H "Authorization: Bearer eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJpc3MiOiJzdXBhYmFzZSIsInJlZiI6InV6dG5naWRicGR1eW9kcmFib2ttIiwicm9sZSI6ImFub24iLCJpYXQiOjE3NzI1MTc3ODksImV4cCI6MjA4ODA5Mzc4OX0.FNnTd5p9Qj3WeD0DxQORmNf2jgaVSZ6FU1EGy0W7MRo" \
-      -H "Content-Type: application/json" \
-      -d '{
-        "action": "send",
-        "from_agent": "[from_agent]",
-        "to_agent": "[target_agent]",
-        "project": "[project]",
-        "message_type": "handoff",
-        "title": "[one-line task description]",
-        "content": "[full handoff content — escaped for JSON]"
-      }'
+    You are a handoff writer. Your only job is to save a handoff file via the memory tool.
+
+    **Step A — Save handoff via memory tool:**
+    Read `~/.claude/wds/src/tools/memory/SKILL.md` and follow the `save` operation:
+    - agent_id: [target_agent]
+    - data:
     ```
+    ## Wrapped
+    [current date and time]
 
-    If the call succeeds: extract the `id` field. Return ONLY the first 6 characters. Nothing else.
-    If the call fails or returns an error: return ONLY: FAILED: [error message or HTTP status]
-    ---
+    ## Context
+    [task content from step 1]
 
-    Wait for the sub-agent response.
+    ## Next
+    [next line from step 1]
 
-    **If sub-agent returns 6 characters:** print EXACTLY this — nothing before, nothing after:
-    ```
-    /[target_agent] [6chars]
-    ```
+    ## Learned
+    None
 
-    **If sub-agent returns FAILED:** stop and warn the user:
+    ## Spec Sync
+    None
     ```
-    ⚠️ Agent Space unreachable — handoff not sent.
-    Check that Agent Space credentials are active (open Bitwarden → verify Agent Space API key).
 
-    Handoff content (copy if needed):
-    [full handoff content]
+    **Step B — Confirm:**
+    Return ONLY: `done`
+    ---
+
+    Wait for the sub-agent to return. Then print EXACTLY this — nothing before, nothing after:
     ```
-    Do NOT write the handoff to a file on disk.
+    /[target_agent] progress/[target_agent].md
+    ```
+
+    Session continues.
   </step>
 
 </handoff-steps>

src/skills/start.md

@@ -20,18 +20,9 @@ Loads project state and session context. Always reads the project index first 
 If found: parse Phase Status and Artifacts sections. Hold this as project context — it informs everything below.
 If not found: proceed silently. The index will be built on first wrap.
 
-**Additionally (Agent Space):** query Agent Space for recent project knowledge:
-```bash
-curl -s -X POST "https://uztngidbpduyodrabokm.supabase.co/functions/v1/agent-messages" \
-  -H "Authorization: Bearer eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJpc3MiOiJzdXBhYmFzZSIsInJlZiI6InV6dG5naWRicGR1eW9kcmFib2ttIiwicm9sZSI6ImFub24iLCJpYXQiOjE3NzI1MTc3ODksImV4cCI6MjA4ODA5Mzc4OX0.FNnTd5p9Qj3WeD0DxQORmNf2jgaVSZ6FU1EGy0W7MRo" \
-  -H "Content-Type: application/json" \
-  -d '{"action": "list", "to_agent": "[agent_id]", "project": "[project]", "limit": 5}'
-```
-If Agent Space returns recent messages: use them to supplement the project index. If unavailable: proceed with the file index only — do not block on this.
-
 ### 2. Detect Session State
 
-Read `_wds/tools/memory/SKILL.md` and follow the `load` operation for the current agent_id.
+Read `~/.claude/wds/src/tools/memory/SKILL.md` and follow the `load` operation for the current agent_id.
 
 **Fallback chain:** state found → show resume prompt → fresh start
 
@@ -104,6 +95,5 @@ Do not mention /start or the absence of a state file.
 ## Notes
 
 - Always read `progress/project-index.md` — never skip it. It is the project's memory.
-- Agent Space supplements the index but never replaces it. File is authoritative.
 - The state file lives at `progress/[agent].md` relative to the project root.
 - On resume, get back to work quickly. The user knows the context.