moving

docs/audits/2026-07-12-goal-integration-checkpoint.md

@@ -0,0 +1,90 @@
+# Goal Integration Checkpoint - 2026-07-12
+
+## Scope
+
+Clean integration branch: `codex/goal-integration-2026-07-12`
+
+Base: `origin/main` at `89329f99`
+
+This checkpoint combines the current mainline with:
+
+- hardened Channels from `015e20e4`
+- durable Rooms and external-tool collaboration from `39511843`
+- provider-native model discovery and managed LiteLLM runtime routing
+- recoverable API-key setup in onboarding and Settings
+- mobile onboarding, CSP, and dependency fixes found during combined validation
+
+The original dirty UX checkout was not merged wholesale. Its unrelated working
+files remain untouched.
+
+## Functional Result
+
+- Provider model identities come from provider APIs, not a maintained model list.
+- Anthropic cursor and Google page-token pagination are exhausted.
+- Provider models use stable `provider/model` IDs through UI, Settings, Chat,
+  explicit model selection, fleet execution, and LiteLLM runtime configuration.
+- Saving a key hydrates provider environment aliases, stores the key in Vault,
+  scrubs legacy plaintext config, refreshes catalogs, and restarts the managed
+  router when needed.
+- A model released while Waggle is open is pulled again on focus and becomes
+  executable without a Waggle code change.
+- API-key success and router readiness are separate states. Router failure keeps
+  an inline retry and does not claim that the model is ready.
+- Channels and durable Rooms coexist with the current MCP retrieval, marketplace,
+  evolution, embedding-routing, and security changes on main.
+
+## Verification
+
+- `npm ci --legacy-peer-deps`: pass
+- `npm run build:packages`: pass
+- root `npm run build`: pass
+- server TypeScript project: pass
+- web TypeScript project: pass
+- app TypeScript project: pass
+- combined backend slice: 21 files, 221 tests passed
+- combined web slice: 14 files, 94 tests passed
+- agent, hook, shim, and worker slice: 14 files, 200 tests passed
+- hive-mind MCP server build and focused scope tests: 13 tests passed
+- fresh-port Playwright: 5/5 for app load, navigation, Settings/Models, Room,
+  and zero console errors
+- 390 x 844 rendered inspection: onboarding/API-key, Settings/Models, and Channels
+  fit without horizontal overflow; selected tabs remain visible; console errors 0
+- diff check: pass
+- added-line credential scan outside tests/docs: 0 candidates
+- production dependency audit: 0 high, 0 critical; 19 moderate remain
+- full workspace dependency audit: 0 high, 0 critical; 24 moderate remain
+
+## Corrections Found During Integration
+
+1. The root production build caught a stale frontend `Settings` type that omitted
+   `defaultModel`; the contract now matches the model-save path.
+2. The production CSP blocked the inline pre-hydration theme script. It now runs
+   as a same-origin classic script without weakening `script-src 'self'`.
+3. Tall onboarding steps were vertically centered on mobile, clipping content
+   above the scroll origin. Mobile steps now align to the top and remain centered
+   on larger viewports.
+4. `hono` 4.12.29, `vite` 6.4.3, `vitest` 3.2.7, and `form-data` 4.0.6 now
+   resolve across the workspace, clearing every high- and critical-severity
+   dependency advisory without a breaking application-code change.
+
+## Open Build Hygiene
+
+- Tailwind reports ambiguous arbitrary motion utility classes.
+- Vite reports mixed dynamic and static imports for shape selection.
+- The production build reports chunks larger than 500 kB.
+- Focused tests still emit known image-source and expected error-path stderr;
+  the fresh browser run remains free of console errors.
+
+## Remaining Goal Gates
+
+- paid external-provider credential smoke against at least one live provider
+- live Telegram, Discord, and Slack credentials; WhatsApp scan/restart/unlink
+- one real held-tool approval arriving through a Channel
+- packaged Tauri desktop and signed installer evidence
+- remaining T13-T19 launch, utility, hook, Browser Companion, CI, and deployment
+  evidence, or explicit scope deferrals
+- reconcile the broader dirty UX correction checkout in deliberate slices
+- run the final five-persona scorecards; no persona is yet formally confirmed at
+  9/10
+
+This branch is local and has not been pushed.