fix(orgs): enforce root-cause-first org validation

.cursor/rules/root-cause-first.mdc

@@ -0,0 +1,21 @@
+---
+description: Require source-level fixes for malformed or polluted domain data
+alwaysApply: true
+---
+
+# Root-Cause-First Bug Fixing
+
+When malformed, stale, duplicated, or unexpected data appears in the UI:
+
+1. Do not start with a UI filter, hidden row, fallback label, or string-pattern special case.
+2. Determine whether the value is persisted/remote domain data or only a presentation defect.
+3. Inspect the authoritative source and trace all transformations and writers to the earliest boundary that created the invalid state.
+4. Fix the invariant at that boundary: input parsing, API/RPC ingestion, persistence, sync reconciliation, or canonical state projection.
+5. Add a regression test at the producing boundary. A render/selector test alone is insufficient.
+6. Handle historical pollution separately: inventory dependencies, confirm destructive cleanup, clean narrowly, and read back the authoritative source.
+7. UI filtering is allowed only for an explicit product requirement or as defense-in-depth after the source fix. Never use it as the sole fix for invalid upstream data.
+8. Do not change adjacent valid behavior without an explicit request.
+
+Before declaring success, report: authoritative source, root cause, producing write path, source-level invariant, historical remediation, and verification evidence.
+
+Any UI predicate that hides malformed data must cite an explicit product requirement. If the value violates the domain model, reject the UI-only patch until the producing path is fixed and regression-tested.

AGENTS.md

@@ -32,6 +32,24 @@ If the skill block isn't already prefetched in your context, read its `SKILL.md`
 
 ---
 
+## Root-Cause-First Bug Fixing
+
+When malformed, stale, duplicated, or unexpected data appears in the UI:
+
+1. **Do not start with a UI filter, hidden row, fallback label, or string-pattern special case.** First determine whether the value is real persisted/remote domain data or only a presentation defect.
+2. Identify the authoritative source, inspect the actual stored payload, and trace every transformation and writer back to the earliest boundary that created the invalid state.
+3. Fix the invariant at that authoritative boundary: user-input parsing, API/RPC ingestion, persistence write, sync reconciliation, or canonical state projection.
+4. Add a regression test at the producing boundary proving the invalid state can no longer be created. A selector/render test alone is not sufficient.
+5. Treat historical pollution separately: inventory dependent data, get confirmation before destructive cleanup, perform the narrowest cleanup, then read back the authoritative source.
+6. UI filtering is allowed only when exclusion is an explicit product requirement or defense-in-depth **after** the source fix. It must never be the sole fix for invalid upstream data.
+7. Do not change adjacent valid behavior unless the user explicitly requests it.
+
+Before declaring the issue fixed, report the authoritative source, root cause, producing write path, source-level invariant, historical remediation, and verification evidence.
+
+Review gate: any UI predicate introduced to hide malformed data must cite an explicit product requirement. If the value violates the domain model, reject the UI-only patch until the producing path is fixed and covered by a regression test.
+
+---
+
 ## Default Delivery Flow
 
 ### Touching `*.tsx` files (UI work)

CLAUDE.md

@@ -32,6 +32,24 @@ If the skill block isn't already prefetched in your context, read its `SKILL.md`
 
 ---
 
+## Root-Cause-First Bug Fixing
+
+When malformed, stale, duplicated, or unexpected data appears in the UI:
+
+1. **Do not start with a UI filter, hidden row, fallback label, or string-pattern special case.** First determine whether the value is real persisted/remote domain data or only a presentation defect.
+2. Identify the authoritative source, inspect the actual stored payload, and trace every transformation and writer back to the earliest boundary that created the invalid state.
+3. Fix the invariant at that authoritative boundary: user-input parsing, API/RPC ingestion, persistence write, sync reconciliation, or canonical state projection.
+4. Add a regression test at the producing boundary proving the invalid state can no longer be created. A selector/render test alone is not sufficient.
+5. Treat historical pollution separately: inventory dependent data, get confirmation before destructive cleanup, perform the narrowest cleanup, then read back the authoritative source.
+6. UI filtering is allowed only when exclusion is an explicit product requirement or defense-in-depth **after** the source fix. It must never be the sole fix for invalid upstream data.
+7. Do not change adjacent valid behavior unless the user explicitly requests it.
+
+Before declaring the issue fixed, report the authoritative source, root cause, producing write path, source-level invariant, historical remediation, and verification evidence.
+
+Review gate: any UI predicate introduced to hide malformed data must cite an explicit product requirement. If the value violates the domain model, reject the UI-only patch until the producing path is fixed and covered by a regression test.
+
+---
+
 ## Default Delivery Flow
 
 ### Touching `*.tsx` files (UI work)

src-tauri/crates/project-management/src/projects/io/orgs.rs

@@ -54,6 +54,9 @@ pub fn create_project_org(request: &CreateProjectOrgRequest) -> Result<ProjectOr
     if name.is_empty() {
         return Err("Org name is required".to_string());
     }
+    if name.contains("://") {
+        return Err("Org name must be a name, not a URL".to_string());
+    }
 
     let slug = normalize_slug(name);
     if slug.is_empty() {
@@ -392,6 +395,27 @@ mod tests {
         assert!(orgs.iter().any(|entry| entry.id == org.id));
     }
 
+    #[test]
+    fn create_project_org_rejects_urls_as_names() {
+        let _sandbox = test_env::sandbox();
+
+        for name in [
+            "orgii://cloud/join?invite=abc",
+            "https://example.com/team",
+            "ssh://git@example.com/team",
+        ] {
+            let error = create_project_org(&CreateProjectOrgRequest {
+                name: name.to_string(),
+                id: None,
+            })
+            .expect_err("URL must not be persisted as an org name");
+
+            assert_eq!(error, "Org name must be a name, not a URL");
+        }
+
+        assert_eq!(read_project_orgs().expect("read orgs").len(), 1);
+    }
+
     #[test]
     fn create_project_org_uniquifies_colliding_slug_and_key() {
         let _sandbox = test_env::sandbox();