How I Deploy 36 Supabase Edge Functions via GitHub Actions with 4 Parallel Claude Code Instances

The Setup

I'm building 自分株式会社 — an AI-integrated life management app — solo. To maximize output, I run 4 Claude Code instances in parallel:

VSCode instance     → lib/ (Flutter frontend)
Web instance        → supabase/functions/ (Edge Functions)
Windows App         → docs/ (migrations, documentation)
PowerShell instance → CI/CD oversight, global coordination

Each instance owns its directory and always starts with git pull --rebase origin main to prevent merge conflicts.

---

The Problem: 7 Edge Functions Outside CI/CD

After auditing deploy-prod.yml, I found 7 Edge Functions missing from the deploy list — they required manual deployment:

• check-competitor-updates — checks availability of 21 competitor websites
• health-check — DB connection + table availability check
• analyze-reality — AI-powered reality check feature
• trigger-analysis — election analysis trigger
• local-election-intelligence — local election data
• agent-runtime-cycle — AI agent periodic cycle
• get-competitor-monitoring (new — didn't exist yet)

Manual deployment = forgotten deployments = production/staging drift.

---

The Fix: Add Everything to deploy-prod.yml

- name: Deploy Supabase Edge Functions
  run: |
    # Existing functions
    supabase functions deploy reply-support-request --no-verify-jwt
    supabase functions deploy post-x-update --no-verify-jwt
    supabase functions deploy schedule-daily-digest --no-verify-jwt
    # ... (all existing functions)

    # Newly added to CI/CD
    supabase functions deploy check-competitor-updates --no-verify-jwt
    supabase functions deploy get-competitor-monitoring --no-verify-jwt
    supabase functions deploy health-check --no-verify-jwt
    supabase functions deploy analyze-reality --no-verify-jwt
    supabase functions deploy trigger-analysis --no-verify-jwt
    supabase functions deploy local-election-intelligence --no-verify-jwt
    supabase functions deploy agent-runtime-cycle --no-verify-jwt

Simple fix, but it requires knowing which functions existed. The audit step is the hard part.

---

The New Function: get-competitor-monitoring

I had check-competitor-updates (POST: scrapes competitor sites and writes to DB) but no corresponding read endpoint. So I created a companion GET function:

// GET /functions/v1/get-competitor-monitoring?days=7&limit=50
serve(async (req) => {
  const url = new URL(req.url);
  const days = parseInt(url.searchParams.get("days") ?? "7", 10);
  const limit = Math.min(
    parseInt(url.searchParams.get("limit") ?? "50", 10),
    100
  );

  const since = new Date(
    Date.now() - days * 24 * 60 * 60 * 1000
  ).toISOString();

  const { data } = await supabase
    .from("competitor_monitoring")
    .select("*")
    .gte("checked_at", since)
    .order("checked_at", { ascending: false })
    .limit(limit);

  // Deduplicate: keep only the latest result per competitor
  const latestByCompetitor: Record<string, CompetitorResult> = {};
  for (const row of data ?? []) {
    if (!latestByCompetitor[row.competitor_key]) {
      latestByCompetitor[row.competitor_key] = {
        key: row.competitor_key,
        name: row.competitor_name,
        available: row.available,
        latency_ms: row.latency_ms,
        checked_at: row.checked_at,
      };
    }
  }

  const competitors = Object.values(latestByCompetitor);
  const available = competitors.filter((c) => c.available).length;

  return new Response(
    JSON.stringify({
      competitors,
      summary: {
        total: competitors.length,
        available,
        availabilityPct: Math.round((available / competitors.length) * 100),
      },
    }),
    { headers: { "Content-Type": "application/json" } }
  );
});

Pattern: write function (POST) + read function (GET) as a pair. Never mix read and write in the same endpoint.

---

UI Coverage Tracking: EdgeFunctionSummaryCard

With 36 Edge Functions, I needed a way to see which ones had UI connections. edge_function_summary_card.dart shows a live audit:

Edge Functions Status
Total: 36 | With UI: 31 | Without UI: 5 (86%)

Functions without UI connections get a warning icon. Clicking "Details" opens /edge-functions where you can manually trigger any function and see the response.

This widget runs the edge-function-audit.yml GitHub Actions workflow daily, which creates an Issue if any function is orphaned (deployed but no UI path).

---

The 4-Instance Coordination Rules

Running 4 Claude Code instances in parallel on the same repo requires discipline:

Rule	Why
Each instance owns one directory	Prevents merge conflicts
git pull --rebase before every session	Picks up changes from other instances
Web instance never touches lib/	flutter analyze errors should only come from the VSCode instance
All instances use the same COMPRESSED_PROMPT_V3.md	Shared state across instances without real-time communication

The COMPRESSED_PROMPT_V3.md file is our shared memory: it records which features are implemented, which instance is responsible for what, and what the current counts are (EF count, page count, etc.).

---

flutter analyze 0 errors — Always

With 4 instances modifying code, flutter analyze errors can accumulate fast. Our rule: no instance commits with analyze errors. The CI pipeline enforces this:

- name: Analyze code
  run: flutter analyze --no-pub
  # Fails the entire deploy if there are any errors

---

Results

Before	After
29/36 Edge Functions in CI/CD	36/36 Edge Functions in CI/CD
Manual deploy for 7 functions	Zero manual deployments
Production/staging drift possible	supabase db push catches migrations on every push

The takeaway: an Edge Function that's not in CI/CD is a liability. It will drift from main, miss migrations, and eventually fail in ways that are hard to debug.

---

Try it: 自分株式会社

buildinpublic #Supabase #GitHubActions #Flutter #ClaudeCode