feat: add local Browser Use, model discovery, and automation notifications

.env.example

@@ -142,6 +142,10 @@ LINKSENSE_WORKER_EGRESS_NETWORK=linksense-worker-egress
 
 LINKSENSE_UPLOAD_MAX_FILE_SIZE_MB=100
 LINKSENSE_UPLOAD_MAX_FILES_PER_CONVERSATION=100
+# Optional absolute path to the macOS Desktop client ZIP, mounted read-only for API downloads.
+# Local development finds outputs/desktop/LinkSense-Desktop-macos.zip automatically.
+LINKSENSE_DESKTOP_CLIENT_ARCHIVE=
+LINKSENSE_DESKTOP_CLIENT_ARCHIVE_WINDOWS=
 # User attachment uploads accept any meaningful regular file. This legacy
 # MIME list is ignored by Composer attachments; preview support is still
 # controlled by separate preview safety checks.

.github/ISSUE_TRIAGE.md

@@ -26,7 +26,7 @@ Configure these repository settings before enabling them:
 
 The checked-in workflows use the HTTPS Responses API-compatible gateway at `https://codex-lb.lingx-ai.com/backend-api/codex`. To change providers, update the literal `OPENAI_BASE_URL` under `engine.env`, update `network.allowed`, and recompile both lock files. Never store the provider key in the workflow source.
 
-Both workflows import `shared/codex-model-pricing.md`. It supplies explicit `gpt-6.1-sol` pricing to the agent, threat-detection, and AWF API-proxy catalogs, so the pinned runtime can account for this model even before its built-in catalog is updated. Threat detection preserves the native `detection` model alias from the current default branch. Keep `GH_AW_MODEL_AGENT_CODEX` set to the intended model; changing it to an unknown model also requires adding that model's pricing and recompiling. Do not disable threat detection or the AI-credit budget, or add a blanket fallback price to bypass a missing catalog entry.
+Both workflows import `shared/codex-model-pricing.md`. It supplies explicit `gpt-6.1-sol` pricing to the agent, threat-detection, and AWF API-proxy catalogs, so the pinned runtime can account for this model even before its built-in catalog is updated. Threat detection explicitly uses `gpt-6.1-sol`: the unpriced `detection` alias failed with `unknown_model_ai_credits` in run `40614653511`, before a security assessment could run. Keep `GH_AW_MODEL_AGENT_CODEX` set to the intended model; changing it to an unknown model also requires adding that model's pricing and recompiling. Do not disable threat detection or the AI-credit budget, or add a blanket fallback price to bypass a missing catalog entry.
 
 The overlay uses the [official OpenAI Standard prices](https://developers.openai.com/api/docs/models/gpt-6.1-sol), checked on 2026-10-08: USD 2 input, 10 output, 0.10 cache read, and 2.50 cache write per million tokens. Frontmatter values are **USD per token**. AIC is a workflow budget estimate, not the gateway's invoice: the pinned AWF overlay accepts flat rates and does not model long-context, Fast-mode, regional, or gateway-specific surcharges. Update the rates to the gateway's applicable billing policy before relying on AIC as an actual-spend limit.
 

.github/workflows/agent-ready.md

@@ -45,7 +45,7 @@ safe-outputs:
   threat-detection:
     engine:
       id: codex
-      model: detection
+      model: gpt-6.1-sol
   create-pull-request:
     title-prefix: "[agent] "
     labels:

.github/workflows/issue-triage.md

@@ -44,7 +44,7 @@ safe-outputs:
   threat-detection:
     engine:
       id: codex
-      model: detection
+      model: gpt-6.1-sol
   add-labels:
     allowed:
       - bug

.gitignore

@@ -36,3 +36,4 @@ apps/web/public/vendor/
 apps/web/dummy-non-existing-folder/
 **/__pycache__/
 **/.venv/
+apps/runner/store

AGENTS.md

@@ -89,6 +89,7 @@ prisma/
 
 ## Frontend Standards
 
+- For UI changes, first consult `docs/development/ui-design-system.md` and reuse its implemented tokens, components, and patterns. Follow the current task's business constraints; read Figma for new designs or gaps rather than requiring it for routine changes covered by the repository baseline.
 - Use React function components and Hooks. Prefer React’s built-in state for page-local state, and use TanStack Query consistently for server state.
 - Introduce a lightweight state library only when cross-page shared state is genuinely complex. Do not duplicate server cache in global state.
 - Page components orchestrate flows, business components handle domain interactions, and shared components provide reusable UI. Do not accumulate request, authorization, and presentation logic in a single component.

THIRD-PARTY-NOTICES.md

@@ -58,3 +58,14 @@ the upstream terms for your distribution and deployment model, especially if
 you modify or redistribute an upstream image or provide one of its services
 directly to others. The release manifest and the software inside each image
 remain the authority for the exact installed version and its notices.
+
+## React Bits interface effects
+
+The card spotlight decoration in `apps/web/src/components/ui/card-spotlight.ts`
+and its stylesheet adapts React Bits SpotlightCard at commit
+[`e12c22192ac7b4b452c5e2feb9b586d74765d696`](https://github.com/DavidHDev/react-bits/blob/e12c22192ac7b4b452c5e2feb9b586d74765d696/src/ts-tailwind/Components/SpotlightCard/SpotlightCard.tsx).
+Copyright (c) 2026 David Haz. It retains the upstream MIT + Commons Clause terms,
+which are included in [the shipped notice](apps/web/public/licenses/react-bits.txt)
+and served at `/licenses/react-bits.txt`. It is part of the LinkSense interface;
+it is not added to downloadable application, plugin or site templates. These
+third-party portions are not relicensed under CPAL-1.0.