heymegabyte/agent-skills / 05-architecture-and-stack
architecture-and-stack
VerifiedCloudflare-first platform selection. Workers, D1, R2, KV, DO, Queues, Vectorize, Containers, Workflows, native Go CLIs and measured Rust/WASM acceleration. Default stack, override conditions, auth, data patterns and reliability.
05 — Architecture and Stack
Default stack: _kernel/standards.md#stack. Override conditions below.
Language and execution placement
- Keep TypeScript/Hono as the Cloudflare product control plane. Build new CLIs in native Go using
rules/terminal-styling.md; extend existingaid, not a new portfolio CLI. - Use Rust → WASM for bounded computation only when representative measurements show a significant end-to-end improvement. Profile with existing tools/scripts; include initialization, JS/WASM copies, memory, bundle size and correctness. Do not promise a speedup from the language alone.
- Keep I/O and Cloudflare bindings in the host adapter. Verify the exact crate/features/imports against the deployed Workers runtime; browser WASM, WASI components and native Rust are not interchangeable deployment targets. Batch calls across the JS/WASM boundary.
- Use native Rust/Go jobs for workloads needing processes, large memory, native libraries or prolonged computation. The control plane remains on Cloudflare; apply the idle-compute policy in
rules/cloudflare-hostable-supervisor.md. - Prefer native browser/CSS/Web Animations/View Transitions for presentation. Do not move animations to WASM without measured benefit, or make build execution wait for animation playback.
Approved integration scope (2026-10-27)
- Pursue mail-authentication diagnostics, routing/matching, structured diffing, bounded XLSX processing, restic/rclone-backed backup/transfer jobs, conditional Markdown acceleration and fast incremental previews where the product needs them. Reuse mature implementations; keep secrets/network access in the appropriate host.
- Exclude from this rollout: a new unified
mbCLI, standalone doctor/benchmark/deployment-planning CLIs, custom MIME or shared-redaction WASM, a shared Go-hosted WASM engine, DuckDB product querying, SVG-card generation, dedicated analytics kernels and a separate telemetry CLI. These are scope decisions, not a ban on existing upstream capabilities. - Defer event replay tooling. Reconnecting a live build stream to reconstruct its UI remains part of build correctness and does not authorize operational job replay.
- Resolve product identity and repository ownership from current source; do not assign a product role from a domain name alone. issuel.ink's purpose remains undecided; no earlier suggested product role is approved by implication.
Sources: Workers WebAssembly, workers-rs, Charm CLI guidance.
Cloudflare-first decision tree
Compute: Workers (default, every HTTP/cron/queue) · Pages (static-only marketing, rare) · Containers (non-JS runtimes: Playwright headful, ffmpeg, Python ML, build orchestration) · Sandbox SDK (generated/risky code before live promotion)
State: D1 (default relational, ≤10GB/db, Sessions API read-replicas, Time Travel 30-day PIT) · KV (eventually-consistent, cache/sessions/feature-flags) · R2 (object storage, lifecycle Standard→IA after 30d) · Durable Objects (coordination + strongly-consistent SQLite storage since Apr 2025, chat rooms/builder sessions/rate-limiting) · Hyperdrive (front external Postgres/MySQL) · Vectorize (semantic search/RAG, 5M dim/index, topK 100, 10 metadata indexes)
Async: Queues (best-effort, 5000 msg/sec, R2 event notifications) · Workflows v2 (deterministic, 50K concurrent, 300 creates/sec, 2M queued/workflow, step.do + step.sleep + step.waitForEvent) · Inngest (event-driven, better DX/observability)
AI: Workers AI (Llama 4 Scout 17B default — multimodal, free tier; GLM-4.7-flash tool-calling; 3.3 70B legacy) · AI Gateway (caching + rate-limit + fallback + logging for every LLM call) · Vectorize (embeddings + ANN search)
Override conditions (when CF isn't enough)
| Need | Fallback | Adapter |
|---|---|---|
| Advanced SQL (RLS, OLAP, partial indexes) | Neon Postgres via Hyperdrive | SqlPort |
| Redis primitives at scale (sorted sets, streams) | Upstash Redis | KvPort |
| Sub-millisecond global state | Upstash QStash | QueuePort |
| Specific provider (OpenAI assistants, Anthropic batch) | Direct API via AI Gateway | AiPort |
| Vector + SQL co-located | Neon pgvector | VectorPort |
Adapters live in libs/core/ports/. Product code imports port, never vendor SDK directly. See rules/cloudflare-hostable-supervisor.md.
Auth (default Clerk M2M JWT)
- Clerk — M2M JWT (free, networkless verification), passkeys, OAuth, magic links; Better Auth when Clerk pricing doesn't fit (rare) (free tier: 50K MRU since 2026-02)
- Hash API keys at rest. Audit log every sensitive action.
- Tenant isolation: every table carries
org_id, every query filters by it (404 on mismatch, never 403)
Data patterns
D1
[[d1_databases]]
binding = "DB"
database_name = "myapp"
wrangler typesagainstcompatibility_date+ bindings (preferred over hand-maintained Env interface)- Drizzle v1 RQBv2 + Zod for query + validation; batch via
db.batch([...])(no transactions in D1) - Sessions API:
db.withSession(bookmark)· Time Travel:wrangler d1 time-travel restore
R2: per-extension content-type on upload · lifecycle Standard→IA after 30d · event notifications → Queues at 5000 msg/sec for thumbnailing/indexing · versioning for asset rollback
Durable Objects: one DO per stateful entity · SQLite-backed, 10GB per DO · direct stub env.MY_DO.getByName(name) · alarm misfires → idempotent handler
Reliability
- Workers CPU 10ms free / 50ms paid default (configurable 5min); wall time 30s paid
ctx.waitUntil()for async post-response work;ctx.passThroughOnException()for graceful degradation- WebSocket + JSRPC payload up to 32 MiB
Cost discipline
- Workers free tier: 100k req/day; Workers Paid: $5/mo (10M req + 30M CPU-ms) + $0.30/M extra req + $0.02/M extra CPU-ms
- D1 free: 5M rows-read + 100K rows-written/day — exceeding either hard-fails every query until 00:00 UTC reset (since 2026-09); 500MB/DB, 5GB account total
- D1 on Workers Paid: 5GB + 25B rows-read + 50M rows-written/mo; then $0.75/GB-mo + $0.001/M rows-read + $1/M rows-written; 10GB/DB hard cap; no egress; read replication included (verified 2026-10-04 against developers.cloudflare.com — Workers/D1/R2 rates all unchanged)
- R2: 10GB free, $0.015/GB-mo, $0/egress · Workers AI Llama 4 Scout FREE · AI Gateway free
- Solo SaaS <$100k/mo MRR stays 10-100× cheaper than AWS-equivalent on CF
Default config (wrangler.jsonc)
{
"name": "myapp",
"main": "src/worker/index.ts",
"compatibility_date": "2026-04-15",
"compatibility_flags": ["nodejs_compat"],
"observability": { "enabled": true },
"secrets_required": ["CLERK_SECRET_KEY", "RESEND_API_KEY"],
"d1_databases": [{ "binding": "DB", "database_name": "myapp" }],
"kv_namespaces": [{ "binding": "CACHE", "id": "..." }],
"r2_buckets": [{ "binding": "BUCKET", "bucket_name": "myapp-assets" }],
"ai": { "binding": "AI" }
}
Decision template (use for every architecture call)
- Can CF primitive do this? → Use it.
- Does this need adapter for portability? → Adapter only if real business case.
- Cost projection at 10× current scale → Still affordable?
- Failure mode → Graceful degradation defined?
- Migration path → If we have to leave CF, what does it cost?
See submodules: cloudflare-primitives.md, data-patterns.md, reliability.md, auth-patterns.md.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack." | Stars: 196k | Updated | Claude CodeCodex | Verified | Install |
| ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read. | Stars: 193k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report. | Stars: 193k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt. | Stars: 193k | Updated | Claude CodeCodex | Verified | Install |