feat(ops): scheduled peak-hours scaling for the Render web service

.agents/types/agent-definition.ts

@@ -433,6 +433,7 @@ export type ModelName =
   | 'z-ai/glm-4.7-flash:nitro'
   | 'minimax/minimax-m2.5'
   | 'minimax/minimax-m2.7'
+  | 'minimax/minimax-m3'
   | (string & {})
 
 import type { ToolName, GetToolParams } from './tools'

.bin/bun

@@ -55,6 +55,7 @@ PROJECT_ROOT="$(find_project_root)"
 ENV_LOCAL_FILE="$PROJECT_ROOT/.env.local"
 ENV_DEVELOPMENT_LOCAL_FILE="$PROJECT_ROOT/.env.development.local"
 CACHE_TTL_SECONDS=${INFISICAL_CACHE_TTL:-900}  # Default 15 minutes
+INFISICAL_ENV_NAME=${INFISICAL_ENV:-dev}
 CALLING_DIR="$(pwd)"
 
 # Clean up any leftover temporary files
@@ -103,8 +104,9 @@ sync_from_infisical() {
 
     local temp_file=$(mktemp)
 
-    # Run infisical export in background from project root
-    (cd "$PROJECT_ROOT" && infisical export > "$temp_file" 2>/dev/null; echo $? > "$temp_file.exit") &
+    # Run infisical export in background from project root. Pass an explicit
+    # environment so the CLI never waits for interactive environment selection.
+    (cd "$PROJECT_ROOT" && infisical export --env="$INFISICAL_ENV_NAME" --silent > "$temp_file" 2>/dev/null; echo $? > "$temp_file.exit") &
     local pid=$!
 
     # Wait up to 10 seconds

.bun-version

@@ -1 +1 @@
-1.3.11
+1.3.14

.claude/launch.json

@@ -0,0 +1,57 @@
+{
+  "version": "0.0.1",
+  "configurations": [
+    {
+      "name": "freebuff-web",
+      "runtimeExecutable": "bun",
+      "runtimeArgs": [
+        "--cwd",
+        "freebuff/web",
+        "dev"
+      ],
+      "port": 3002,
+      "autoPort": false
+    },
+    {
+      "name": "codebuff-web",
+      "runtimeExecutable": "bun",
+      "runtimeArgs": [
+        "--cwd",
+        "web",
+        "dev"
+      ],
+      "port": 3000,
+      "autoPort": false
+    },
+    {
+      "name": "freebuff-desktop",
+      "runtimeExecutable": "bash",
+      "runtimeArgs": [
+        "-c",
+        "PORT=8787 exec bun freebuff-desktop/src/app/server.ts"
+      ],
+      "port": 8787,
+      "autoPort": false
+    },
+    {
+      "name": "freebuff-desktop-web",
+      "runtimeExecutable": "bash",
+      "runtimeArgs": [
+        "-c",
+        "cd freebuff-desktop && exec bun run dev:web"
+      ],
+      "port": 5174,
+      "autoPort": false
+    },
+    {
+      "name": "freebuff-desktop-ui-only",
+      "runtimeExecutable": "bash",
+      "runtimeArgs": [
+        "-c",
+        "cd freebuff-desktop && exec bun run ui:dev"
+      ],
+      "port": 5174,
+      "autoPort": false
+    }
+  ]
+}

.cursor/rules/end-to-end-pr-development.mdc

@@ -0,0 +1,56 @@
+---
+description: End-to-end PR development, deployment, and production verification workflow
+alwaysApply: true
+---
+
+# End-To-End PR Development
+
+Use this workflow for complex UI changes, backend changes, deployment-sensitive fixes, and features that need real product verification. For simple questions, small bug explanations, or low-risk local edits, do not run the full production loop.
+
+## Environment Checklist
+
+- Ensure `direnv` is active and `bun` resolves to the repo wrapper: `command -v bun` should point at `.bin/bun`.
+- Infisical manages local secrets sync. If env sync fails or `.env.local` appears stale, ask the user to run `infisical login` / `direnv allow`, then retry with `INFISICAL_CACHE_TTL=0 bun ...`.
+- Axiom manages production logs for the Next.js server. Use the repo log scripts and the Axiom query token from env when production behavior needs diagnosis.
+- Convex manages Freebuff Web production backend logs. Use Convex logs for backend runtime errors and operational warnings.
+- GitHub Actions manages deployment status and required checks. Monitor checks actively after merging or pushing deploy fixes.
+- If Convex APIs changed, regenerate types before typechecking with `bun x convex dev --once` from the relevant Convex app directory. In non-interactive agent terminals, never use `convex dev`, `npx convex dev`, or `bun x convex dev` without `--once`, and never hand-edit `src/convex/_generated/*` files to fake a clean typecheck.
+
+## Development Loop
+
+- Keep changes scoped and delivery-focused. Prefer the simplest implementation that matches existing repo patterns.
+- Run targeted checks first. Typecheck or broader tests when the change touches shared contracts, backend behavior, generated Convex APIs, or critical UI paths.
+- If typecheck items are skipped for speed, still ensure no obvious type issues in changed files and state what was skipped.
+- Do not wait for perfect local coverage when the user asks for production iteration. Fix, push, merge, monitor, and verify in production.
+
+## Merge And Deploy
+
+- For changes intended to ship, merge directly with `main` when requested.
+- After merge/push, monitor GitHub Actions for about 12 minutes or until the relevant checks/deployment state is clear.
+- If checks fail, resolve merge conflicts or failures quickly, push fixes, and keep monitoring.
+- When a fix needs a PR, keep iterations in a single PR thread where practical. Merge immediately when requested, then continue monitoring checks and deployment.
+
+## Production Verification
+
+- Test immediately in production after deployment: web changes on `https://freebuff.com/web`; CLI changes through tmux output.
+- For Freebuff Web, assume the browser is logged in as Victor. If auth is missing, stop and ask the user to log in/authenticate before continuing tests.
+- Keep testing fast and focused: verify each new feature or fix, click through the main affected paths, and make sure adjacent core behavior is not broken.
+- For chat/platform changes, type into inputs, send messages, inspect output, navigate between relevant links/tabs, and test on production data.
+- For complex features, broaden coverage enough to catch broken states, regressions, and missing interactions. For smaller fixes, run the minimum test that proves the core behavior.
+
+## Evidence
+
+- Capture screenshots for production verification, especially before/after states and the final working state.
+- If video is needed, use a real click-through screen recording, not stitched screenshots. Make clicks visible and easy to follow.
+- Attach screenshot/video evidence to the PR thread in comments. Keep all iteration history in that PR thread when possible.
+- Final summaries should link the PR or PRs used for the branch and mention where the screenshot/video evidence lives.
+
+## External Services
+
+- If testing requires external accounts, email inboxes, payment flows, private provider dashboards, or other access the agent does not have, document it in the PR and final summary.
+- Include a clear checklist of what the agent verified versus what the user must verify manually.
+
+## Final Report
+
+- State what shipped, what was verified in production, which checks/deployments were monitored, and any remaining risks.
+- Call out skipped checks, auth blockers, external-service gaps, or logs that indicate follow-up is needed.

.env.example

@@ -10,14 +10,29 @@ OPENCODE_API_KEY=dummy_opencode_key
 
 # Database & Server
 DATABASE_URL=postgresql://manicode_user_local:secretpassword_local@localhost:5432/manicode_db_local
+# Optional: Redis-compatible store for distributed free-mode rate limits.
+# On Render, set this to the Key Value instance's internal URL.
+# REDIS_URL=redis://localhost:6379
 PORT=4242
 
 # Authentication
 CODEBUFF_GITHUB_ID=dummy_github_id
 CODEBUFF_GITHUB_SECRET=dummy_github_secret
 FREEBUFF_GITHUB_ID=dummy_freebuff_github_id
 FREEBUFF_GITHUB_SECRET=dummy_freebuff_github_secret
+# Google OAuth (optional). Authorized redirect URI per origin:
+#   {origin}/api/auth/callback/google
+# Freebuff falls back to CODEBUFF_GOOGLE_* when its own are unset.
+CODEBUFF_GOOGLE_ID=dummy_google_id
+CODEBUFF_GOOGLE_SECRET=dummy_google_secret
+FREEBUFF_GOOGLE_ID=dummy_freebuff_google_id
+FREEBUFF_GOOGLE_SECRET=dummy_freebuff_google_secret
+NEXTAUTH_URL=http://localhost:3000
+NEXTAUTH_FREEBUFF_URL=http://localhost:3002
 NEXTAUTH_SECRET=dummy_nextauth_secret_at_least_32_chars_long
+# Codebuff user ID for the dedicated server-only Freebuff Web service account.
+# Never set this to a personal user ID.
+# FREEBUFF_WEB_SERVICE_USER_ID=00000000-0000-0000-0000-000000000000
 
 # Payment (Stripe)
 STRIPE_SECRET_KEY=sk_test_dummy_stripe_secret
@@ -29,6 +44,7 @@ STRIPE_SUBSCRIPTION_500_PRICE_ID=price_dummy_subscription_500_id
 
 # External Services
 SERPER_API_KEY=dummy_serper_key
+GRAVITY_API_KEY=dummy_gravity_ads_key
 LOOPS_API_KEY=dummy_loops_key
 ZEROCLICK_API_KEY=dummy_zeroclick_key
 
@@ -40,9 +56,11 @@ DISCORD_APPLICATION_ID=dummy_discord_app_id
 # Frontend/Public Variables
 NEXT_PUBLIC_CB_ENVIRONMENT=dev
 NEXT_PUBLIC_CODEBUFF_APP_URL=http://localhost:3000
+NEXT_PUBLIC_FREEBUFF_APP_URL=http://localhost:3002
 NEXT_PUBLIC_SUPPORT_EMAIL=support@codebuff.com
 NEXT_PUBLIC_POSTHOG_API_KEY=phc_dummy_posthog_key
 NEXT_PUBLIC_POSTHOG_HOST_URL=https://us.i.posthog.com
+NEXT_PUBLIC_GRAVITY_PIXEL_ID=00000000-0000-0000-0000-000000000000
 NEXT_PUBLIC_STRIPE_PUBLISHABLE_KEY=pk_test_dummy_publishable
 NEXT_PUBLIC_STRIPE_CUSTOMER_PORTAL=https://billing.stripe.com/p/login/test_dummy
 NEXT_PUBLIC_WEB_PORT=3000

.github/CODEOWNERS

@@ -1 +1 @@
-*       @brandonkachen @jahooma @charleslien
+*       @jahooma

.github/actions/setup-project/action.yml

@@ -21,11 +21,22 @@ runs:
         path: |
           node_modules
           */node_modules
-        key: ${{ runner.os }}-deps-${{ hashFiles('**/bun.lock*', '**/package.json') }}
+          packages/*/node_modules
+          freebuff/*/node_modules
+        key: ${{ runner.os }}-deps-${{ hashFiles('bun.lock') }}
         restore-keys: |
-          ${{ runner.os }}-deps-${{ hashFiles('**/bun.lock*') }}
           ${{ runner.os }}-deps-
 
     - name: Install dependencies
       shell: bash
-      run: bun install --frozen-lockfile
+      run: |
+        for i in 1 2 3; do
+          bun install --frozen-lockfile && break
+          if [ "$i" -eq 3 ]; then
+            echo "bun install failed after $i attempts"
+            exit 1
+          fi
+          echo "bun install failed (attempt $i), clearing download cache and retrying..."
+          rm -rf ~/.bun/install/cache
+          sleep 5
+        done

.github/knowledge.md

@@ -25,8 +25,8 @@ GitHub API URLs are extracted as environment variables to avoid duplication:
 
 ```yaml
 env:
-  GITHUB_API_URL: https://api.github.com/repos/CodebuffAI/codebuff
-  GITHUB_UPLOADS_URL: https://uploads.github.com/repos/CodebuffAI/codebuff
+  GITHUB_API_URL: https://api.github.com/repos/CodebuffAI/freebuff-private
+  GITHUB_UPLOADS_URL: https://uploads.github.com/repos/CodebuffAI/freebuff-private
 ```
 
 This pattern:

.github/workflows/bot-sweep.yml

@@ -1,13 +1,19 @@
 name: Freebuff Bot Sweep
 
+# DISABLED 2026-06-17: the hourly email report was ignored in practice and
+# wasn't surfacing real bots. Replaced by the admin dashboard at
+# freebuff.com/abuse (interactive suspect list + one-click bans). The
+# /api/admin/bot-sweep endpoint still works; re-enable the schedule below
+# if you want the email back.
+#
 # Hourly dry-run sweep over active freebuff sessions. Calls the
 # /api/admin/bot-sweep endpoint, which emails james@codebuff.com with a
 # ranked list of suspects. No bans are issued — review and run
 # scripts/ban-freebuff-bots.ts manually.
 
 on:
-  schedule:
-    - cron: '0 * * * *'
+  # schedule:
+  #   - cron: '0 * * * *'
   workflow_dispatch:
 
 jobs:

.github/workflows/buffbench.yml

@@ -14,7 +14,7 @@ jobs:
       - name: Set up Bun
         uses: oven-sh/setup-bun@v2
         with:
-          bun-version: '1.3.5'
+          bun-version-file: .bun-version
 
       - name: Cache dependencies
         uses: actions/cache@v5
@@ -23,7 +23,8 @@ jobs:
             node_modules
             */node_modules
             packages/*/node_modules
-          key: ${{ runner.os }}-deps-${{ hashFiles('**/bun.lock*') }}
+            freebuff/*/node_modules
+          key: ${{ runner.os }}-deps-${{ hashFiles('bun.lock') }}
           restore-keys: |
             ${{ runner.os }}-deps-
 

.github/workflows/chat-attachment-sweep.yml

@@ -0,0 +1,46 @@
+name: Freebuff Chat Attachment Sweep
+
+# Reclaims storage by deleting the extracted-text blobs of chat document
+# attachments older than the retention window (CHAT_DOC_RETENTION_DAYS), then
+# clearing those messages' attachment refs. Calls the
+# /api/admin/chat-attachment-sweep endpoint on the freebuff web deployment
+# (which owns the chat Convex blob store). Idempotent and bounded per run, so a
+# daily schedule comfortably drains the backlog.
+#
+# Requires:
+#   - secret CHAT_ATTACHMENT_SWEEP_SECRET (also set on the freebuff web service)
+#   - optional var CHAT_ATTACHMENT_SWEEP_URL to override the endpoint URL
+
+on:
+  schedule:
+    # Daily at 04:17 UTC — off-peak, retention is measured in days so frequency
+    # doesn't need to be high.
+    - cron: '17 4 * * *'
+  workflow_dispatch:
+
+jobs:
+  sweep:
+    runs-on: ubuntu-latest
+    timeout-minutes: 10
+    steps:
+      - name: Trigger chat-attachment-sweep
+        env:
+          CHAT_ATTACHMENT_SWEEP_SECRET: ${{ secrets.CHAT_ATTACHMENT_SWEEP_SECRET }}
+          CHAT_ATTACHMENT_SWEEP_URL: ${{ vars.CHAT_ATTACHMENT_SWEEP_URL || 'https://freebuff.com/api/admin/chat-attachment-sweep' }}
+        run: |
+          set -euo pipefail
+          if [ -z "$CHAT_ATTACHMENT_SWEEP_SECRET" ]; then
+            echo "CHAT_ATTACHMENT_SWEEP_SECRET is not set — skipping."
+            exit 0
+          fi
+          status=$(curl -sS -o /tmp/resp.json -w '%{http_code}' \
+            -X POST "$CHAT_ATTACHMENT_SWEEP_URL" \
+            -H "Authorization: Bearer $CHAT_ATTACHMENT_SWEEP_SECRET" \
+            -H "Content-Type: application/json" \
+            --max-time 120)
+          echo "HTTP $status"
+          cat /tmp/resp.json
+          echo
+          if [ "$status" != "200" ]; then
+            exit 1
+          fi