chore(shared,client): align workspace with latest coop updates

Author: Oba-One

.claude/context/product.md

@@ -64,6 +64,36 @@ Someone using the paired mobile web app:
 - Captures sync to the extension via WebSocket relay
 - Captured content enters the private intake queue before review
 
+## Persona & Tone Quick-Reference
+
+Each extension surface and app page has a primary audience. Match tone and vocabulary accordingly.
+
+| Archetype | Persona | Perceives Coop As | Tone | Avoid |
+|-----------|---------|-------------------|------|-------|
+| Regular Member | Sam | "A tab organizer that helps my group" | Warm, plain-language, action-oriented | Jargon, options overload, blockchain vocabulary |
+| Trusted Member | Nia | "Our community's shared memory" | Encouraging, collaborative, direct | Condescension, assumed technical literacy |
+| Operator | Kai | "My community's command center" | Professional, efficient, task-focused | Oversimplification, hiding complexity |
+| Mobile Contributor | Luz | "Quick capture from my phone" | Minimal, fast, zero-friction | Long explanations, multi-step flows |
+
+### Surface → Persona Mapping
+
+| Surface | Primary Persona | Tone Notes |
+|---------|----------------|------------|
+| **Popup** | Sam (Regular) | Quick, casual, action-first. 1-2 words per button. No explanation needed. |
+| **Chickens tab** | Sam / Nia | Minimal, calm, Notion-like. 3-4 info pieces per card. Inline actions. |
+| **Coops tab** | Nia (Trusted) | Collaborative, evidence-based. Show sync state, archive receipts. |
+| **Roost tab** | Nia / Kai | Workspace-focused. Show progress, garden state, submissions. |
+| **Nest tab** | Kai (Operator) | Admin-precise. Member list, invite codes, receiver pairing, settings. |
+| **Landing page** | Sam (first visit) | Warm, atmospheric. Explain value in one scroll. Zero blockchain jargon. |
+| **Receiver PWA** | Luz (Mobile) | Ultra-minimal. Capture button dominant. Status indicators only. |
+
+### Vocabulary Constraints
+
+- **Regular Member / Mobile**: Zero blockchain vocabulary. 6th-grade reading level. Say "save" not "publish to chain", "group" not "multisig".
+- **Trusted Member**: May reference "shared feed", "archive", "sync". Never Solidity internals.
+- **Operator**: May reference "Safe address", "invite codes", "trusted member". Never raw transaction details.
+- All surfaces: Prefer chicken metaphors where they clarify (Loose Chickens, Coop Feed), drop them where they obscure.
+
 ## Brand Direction
 
 ### Chicken Metaphors

.claude/evals/README.md

@@ -0,0 +1,44 @@
+# Agent Evaluation Framework
+
+Structured test cases for verifying agent quality. Run evals after modifying agent definitions or skills.
+
+## Structure
+
+```
+.claude/evals/
+├── README.md           # This file
+├── results.md          # Historical results table
+├── triage/             # Triage agent evals
+├── code-reviewer/      # Code reviewer evals
+└── oracle/             # Oracle agent evals
+```
+
+## Running Evals
+
+Evals are run manually by spawning the agent with a test scenario and comparing output against expected results.
+
+1. Pick a test case from the agent's eval directory
+2. Spawn the agent with the scenario prompt
+3. Compare output against the expected output and scoring rubric
+4. Record results in `results.md`
+
+## Scoring
+
+- **triage**: Classification accuracy (correct severity, type, package routing)
+- **code-reviewer**: Finding precision (true positives vs false positives, correct severity levels)
+- **oracle**: Research quality (sources cited, synthesis depth, actionability of conclusions)
+
+## Maturity
+
+| Agent | Test Cases | Status |
+|-------|-----------|--------|
+| triage | 3 | ready |
+| code-reviewer | 2 | ready |
+| oracle | 2 | ready |
+
+## When to Run
+
+- After modifying `.claude/agents/*.md`
+- After modifying `.claude/skills/*/SKILL.md` for skills referenced by agents
+- Before upgrading model versions (compare scores across models)
+- Quarterly regression check

.claude/evals/code-reviewer/clean-refactor.md

@@ -0,0 +1,43 @@
+# Code Reviewer Eval: Clean Refactor
+
+**Status**: ready
+**Last run**: —
+**Last score**: —
+
+## Scenario
+
+Review a PR that extracts a utility function from a component into shared/src/utils/. No behavior change. Tests pass. The extraction follows barrel import conventions.
+
+```typescript
+// Before: inline in extension/src/views/Popup/TabList.tsx
+function formatTabAge(createdAt: number): string {
+  const diff = Date.now() - createdAt;
+  if (diff < 60000) return 'just now';
+  if (diff < 3600000) return `${Math.floor(diff / 60000)}m ago`;
+  return `${Math.floor(diff / 3600000)}h ago`;
+}
+
+// After: extracted to shared/src/utils/format-time.ts
+export function formatTabAge(createdAt: number): string { /* same logic */ }
+
+// shared/src/index.ts updated with re-export
+export { formatTabAge } from './utils/format-time';
+
+// TabList.tsx now imports from @coop/shared
+import { formatTabAge } from '@coop/shared';
+```
+
+## Expected Output
+
+- **must_fix**: 0
+- **should_fix**: 0-1 (optional: could suggest `formatRelativeTime` as a more general name)
+- **recommendation**: APPROVE
+
+## Eval Criteria
+
+| Criterion | Weight | Pass | Fail |
+|-----------|--------|------|------|
+| Approves clean refactor | 35% | APPROVE recommendation | REQUEST_CHANGES on a clean PR |
+| No false must_fix | 30% | Zero must_fix findings | Invents issues that don't exist |
+| Recognizes barrel import correctness | 20% | Notes the import follows conventions | Flags import pattern as issue |
+| Optional naming suggestion only | 15% | At most a should_fix naming suggestion | Demands changes for style |

.claude/evals/code-reviewer/missing-error-handling.md

@@ -0,0 +1,43 @@
+# Code Reviewer Eval: Missing Error Handling
+
+**Status**: ready
+**Last run**: —
+**Last score**: —
+
+## Scenario
+
+Review a PR that adds a new async function fetching coop data without try/catch, no error boundary wrapping the consuming component, and raw error objects shown to the user via `toast.error(error.message)`.
+
+```typescript
+// New function in shared/src/modules/coop/index.ts
+export async function fetchCoopMembers(coopId: string) {
+  const response = await fetch(`/api/coops/${coopId}/members`);
+  const data = await response.json();
+  return data.members;
+}
+
+// Usage in extension popup
+function MemberList({ coopId }: { coopId: string }) {
+  const [members, setMembers] = useState([]);
+  useEffect(() => {
+    fetchCoopMembers(coopId).then(setMembers);
+  }, [coopId]);
+  return members.map(m => <MemberCard key={m.id} member={m} />);
+}
+```
+
+## Expected Output
+
+- **must_fix**: 1 — Unhandled promise rejection (no .catch() or try/catch on the fetch call)
+- **should_fix**: 1 — No user-friendly error message (should use categorizeError + toast)
+- **recommendation**: REQUEST_CHANGES
+
+## Eval Criteria
+
+| Criterion | Weight | Pass | Fail |
+|-----------|--------|------|------|
+| Catches unhandled promise | 30% | Flags the missing .catch() as must_fix | Misses it or downgrades to should_fix |
+| Flags raw error exposure | 25% | Notes missing categorizeError/user-friendly message | Ignores error UX |
+| Correct severity levels | 20% | must_fix for crash, should_fix for UX | Wrong severity assignments |
+| No false positives | 15% | Doesn't flag correct patterns as issues | Flags style/naming as must_fix |
+| Recommendation = REQUEST_CHANGES | 10% | Correctly recommends changes | Approves despite must_fix |

.claude/evals/oracle/architecture-question.md

@@ -0,0 +1,28 @@
+# Oracle Eval: Architecture Question
+
+**Status**: ready
+**Last run**: —
+**Last score**: —
+
+## Scenario
+
+> "Why does Coop use y-webrtc instead of y-websocket as the primary sync transport?"
+
+## Expected Output
+
+Must address:
+1. **Privacy/local-first rationale**: y-webrtc sends data directly peer-to-peer, so the server never sees document content — aligning with Coop's local-first principle
+2. **y-websocket as fallback**: Coop does use y-websocket (via the API server at `/yws`) as a fallback for document sync when WebRTC fails
+3. **Signaling vs transport distinction**: The API server provides signaling (peer discovery) but doesn't relay document data in the primary path
+
+Must cite at least 2 sources from the codebase (e.g., CLAUDE.md, ADR-007, shared/storage module, API server ws handler).
+
+## Eval Criteria
+
+| Criterion | Weight | Pass | Fail |
+|-----------|--------|------|------|
+| Explains privacy rationale | 30% | Connects y-webrtc to local-first/privacy | Generic "it's faster" answer |
+| Mentions y-websocket fallback | 25% | Identifies y-websocket as fallback, not absent | Says Coop doesn't use y-websocket |
+| Cites 2+ sources | 20% | References specific files or docs | Only reads one file |
+| Accurate technical details | 15% | Correct on signaling vs transport | Confuses signaling with relay |
+| Actionable conclusion | 10% | Summarizes when each transport is used | Ends without synthesis |

.claude/evals/oracle/cross-module-investigation.md

@@ -0,0 +1,34 @@
+# Oracle Eval: Cross-Module Investigation
+
+**Status**: ready
+**Last run**: —
+**Last score**: —
+
+## Scenario
+
+> "How does the capture flow work from tab detection to published artifact?"
+
+## Expected Output
+
+Must trace the full path through multiple packages:
+
+1. **Extension (capture)**: Background service worker detects tabs, creates TabCandidate entries in Dexie
+2. **Shared/agent (refine)**: Agent pipeline runs 16-skill observation on candidates, produces CoopInterpretation
+3. **Extension/Popup + Chickens (review)**: User reviews candidates/drafts, decides what to push
+4. **Shared/coop (publish)**: Explicit push creates Artifact in shared Yjs document
+5. **Shared/storage + sync (sync)**: Artifact syncs to peers via y-webrtc, persisted to Dexie
+6. **Shared/archive (archive)**: Optional archive to Storacha/Filecoin with receipt
+
+Must identify the key boundary: everything before step 4 is local-only, step 4 is the explicit push that makes data shared.
+
+Must cite specific file paths in at least 3 packages.
+
+## Eval Criteria
+
+| Criterion | Weight | Pass | Fail |
+|-----------|--------|------|------|
+| Covers 3+ packages | 25% | Traces through extension, shared, and at least one more | Only covers one package |
+| Identifies local → shared boundary | 25% | Calls out explicit push as the privacy boundary | Treats everything as shared |
+| Specific file paths | 20% | Names actual source files | Only mentions package names |
+| Correct data flow order | 20% | Capture → refine → review → publish → sync → archive | Wrong sequence or missing steps |
+| Synthesis/summary | 10% | Ends with a clear mental model | Dumps findings without conclusion |

.claude/evals/results.md

@@ -0,0 +1,5 @@
+# Eval Results
+
+| Date | Model | Agent | Task | Score | Turns | Notes |
+|------|-------|-------|------|-------|-------|-------|
+| — | — | — | — | — | — | No evals run yet |

.claude/evals/triage/p1-critical-bug.md

@@ -0,0 +1,28 @@
+# Triage Eval: P1 Critical Bug
+
+**Status**: ready
+**Last run**: —
+**Last score**: —
+
+## Scenario
+
+> User reports: "Extension popup is blank white screen after update, console shows TypeError: Cannot read properties of undefined (reading 'map')"
+
+## Expected Output
+
+- **Severity**: P1 (service down for users)
+- **Type**: bug
+- **Affected package**: extension
+- **Root cause hint**: Likely a data shape change in shared breaking a .map() call in popup
+- **Route to**: cracked-coder with `--mode tdd_bugfix`
+- **Not**: oracle (no investigation needed — clear stack trace), not P2 (popup is primary surface)
+
+## Eval Criteria
+
+| Criterion | Weight | Pass | Fail |
+|-----------|--------|------|------|
+| Severity = P1 | 30% | Correctly identifies as P1 | Classifies as P2 or lower |
+| Package = extension | 20% | Identifies extension as affected | Points to shared or app |
+| Routes to cracked-coder | 25% | Routes to cracked-coder with TDD | Routes to oracle or /debug |
+| Concise brief | 15% | Brief is < 5 lines | Over-explains or speculates |
+| Correct type = bug | 10% | Identifies as bug | Classifies as enhancement |

.claude/evals/triage/p2-sync-issue.md

@@ -0,0 +1,28 @@
+# Triage Eval: P2 Sync Persistence Issue
+
+**Status**: ready
+**Last run**: —
+**Last score**: —
+
+## Scenario
+
+> User reports: "Yjs sync works between two browsers but published artifacts don't appear in the coop feed for the third member who joins later"
+
+## Expected Output
+
+- **Severity**: P2 (degraded functionality, not total failure)
+- **Type**: bug
+- **Affected package**: shared (storage/sync modules)
+- **Root cause hint**: Likely a persistence gap — Yjs state not persisted to Dexie or y-websocket fallback not loading historical state for late joiners
+- **Route to**: oracle first (investigate sync/persistence boundary), then cracked-coder
+- **Not**: P1 (sync works for active peers), not extension-only (this is a shared module issue)
+
+## Eval Criteria
+
+| Criterion | Weight | Pass | Fail |
+|-----------|--------|------|------|
+| Severity = P2 | 25% | Correctly identifies as P2 | Classifies as P1 or P3 |
+| Identifies sync/persistence boundary | 25% | Points to storage/sync modules | Treats as pure UI issue |
+| Routes oracle → cracked-coder | 25% | Two-step: investigate then fix | Goes straight to cracked-coder |
+| Package = shared | 15% | Identifies shared as affected | Points only to extension |
+| Concise | 10% | Brief is < 5 lines | Over-explains |

.claude/evals/triage/p3-enhancement.md

@@ -0,0 +1,27 @@
+# Triage Eval: P3 Enhancement Request
+
+**Status**: ready
+**Last run**: —
+**Last score**: —
+
+## Scenario
+
+> User asks: "Can we add a dark mode toggle to the popup header?"
+
+## Expected Output
+
+- **Severity**: P3 (nice-to-have improvement)
+- **Type**: enhancement
+- **Affected package**: extension (popup surface)
+- **Route to**: `/plan` with react + ui-compliance skills
+- **Not**: P2 (not a degradation), not cracked-coder directly (needs planning first)
+
+## Eval Criteria
+
+| Criterion | Weight | Pass | Fail |
+|-----------|--------|------|------|
+| Severity = P3 | 30% | Correctly identifies as P3 | Classifies as P2 or P4 |
+| Type = enhancement | 20% | Identifies as enhancement | Classifies as bug |
+| Routes to /plan | 25% | Routes to /plan for planning | Routes directly to cracked-coder |
+| Suggests skills | 15% | Mentions react and/or ui-compliance | No skill suggestion |
+| Concise | 10% | Brief is < 5 lines | Over-explains |