Claude Code automation recommendations (from claude-code-setup plugin)

[JoshuaVSherman]

Output from running the `claude-code-setup` plugin's `claude-automation-recommender` skill against this repo on 2026-05-20. Different stack from the frontend repos so the recommendations are largely different.

Codebase Profile

• Type: Node + Express 5 backend, TypeScript 5
• Framework: Express 5 + Apollo Server (via `@as-integrations/express5` — GraphQL)
• Database: MongoDB via Mongoose 9
• Auth: bcryptjs + jwt-simple (custom JWT, not passport)
• Security: helmet
• Email: nodemailer (recently migrated from SendGrid — PR fix(inquiry): swap SendGrid for nodemailer + Gmail SMTP (closes #752) #753, May 2026)
• HTTPS: express-sslify (production)
• Tests: Vitest
• CI: CircleCI (configured separately per repo conventions)
• Deploy: Heroku (webjamsalem app — auto-deploys from master)

---

🔌 MCP Servers

MongoDB MCP — direct DB queries during dev

Why: Mongoose schemas live in `src/model/*`. Querying production-shape data during dev usually requires either dropping into `mongo` shell or scripting throwaway code. An MCP server would let Claude query / inspect collections directly without leaving the chat. Especially useful for schema migrations and "is this user record correct" debugging.

Install:
```bash
claude mcp add @anthropic/mongodb-mcp
```

GitHub MCP — PR / issue / Actions ops

Why: Heavy gh CLI usage today (PRs, issues, status checks). MCP server gives Claude more direct, structured access — better at multi-step "find PR, check status, comment, merge" flows than spawning `gh` subprocesses.

```bash
claude mcp add @anthropic/github-mcp
```

---

🎯 Skills

`heroku-deploy-check` (custom, project-specific)

Why: Today's session showed real pain — the Heroku `webjamsalem` app auto-deploys from `master`, but you've hit issues (SendGrid credits exhausted #1040, then the SendGrid→nodemailer swap). A skill that, before merging to master, checks: dynos health, recent log errors, env-var sanity (without dumping secrets per memory rules), and the latest release status. Catches "this PR will break prod" before merge.

Create: `.claude/skills/heroku-deploy-check/SKILL.md`
Invocation: User-only (`/heroku-deploy-check`) — read-only, but performs API calls

`mongoose-schema-diff` (custom)

Why: Mongoose schema changes are risky — they can silently corrupt prod data if not handled with care (e.g. adding a required field with no default to a 50k-row collection). A skill that diffs the proposed schema against the current production-mirror, then warns about destructive changes, would prevent migration mishaps.

---

⚡ Hooks

PostToolUse: `npm run typecheck` on Edit/Write of `.ts` files

Why: TypeScript 5 strict mode. Type errors here can hit prod since Mongoose typings are easy to get wrong. `tsc --noEmit` is fast. Same hook recommendation as the frontend repos, but more critical here because runtime mistakes hit users not just devs.

```json
{
"hooks": {
"PostToolUse": [{
"matcher": "Edit|Write",
"hooks": [{
"type": "command",
"command": "cd /home/joshua/WebJamApps/web-jam-back && npm run typecheck 2>&1 | tail -10"
}]
}]
}
}
```

PreToolUse: block edits to `build/**` and `.env*`

Why: `build/` is generated by `tsc` — editing manually leads to silent drift. `.env` files contain secrets (per memory `feedback-never-dump-secrets-via-config` after the GMAIL_APP_PASSWORD leak 2026-05-18 — already burned once).

```json
{
"hooks": {
"PreToolUse": [{
"matcher": "Edit|Write",
"hooks": [{
"type": "command",
"command": "case "$CLAUDE_TOOL_INPUT_path" in build/|.env|/.env) echo 'BLOCKED: cannot edit generated/secret file'; exit 1;; esac"
}]
}]
}
}
```

---

🤖 Subagents

security-reviewer (high-value here)

Why: This backend handles auth (bcryptjs + jwt-simple), email submission (potential injection vectors), database writes (NoSQL injection risk in Mongoose if queries built carelessly). A subagent specifically prompted to review changes for: SQL/NoSQL injection patterns, secret handling, JWT verification logic, rate-limit gaps, helmet config drift, CSRF protections. Worth its weight on any backend repo handling user data.

Where: `.claude/agents/security-reviewer.md`

---

Implementation plan

• Add `MongoDB MCP` — useful immediately for inspecting production-shape data
• Type-check hook (matches the JaMmusic + CL recommendations)
• PreToolUse block on `build/` + `.env*` (high-value safety net)
• `security-reviewer` subagent (auth + DB + email = strong fit)
• Defer: `heroku-deploy-check` skill, `mongoose-schema-diff` skill (separate design work)

Source: `claude-code-setup` plugin v1.0.0, `claude-automation-recommender` skill, 2026-05-20 against the latest dev branch.

🤖 Generated with Claude Code