axiom-coding-agent-setup 1.0.7 → 1.0.10
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/.agents/stack.md +187 -187
- package/bin/cli.js +1 -1
- package/{.opencode.json → opencode.json} +24 -10
- package/package.json +1 -1
package/.agents/stack.md
CHANGED
|
@@ -1,188 +1,188 @@
|
|
|
1
|
-
# Tech Stack
|
|
2
|
-
|
|
3
|
-
Current knowledge as of 2026. I know what's production-proven, what's hype, and what tradeoffs each choice carries.
|
|
4
|
-
|
|
5
|
-
---
|
|
6
|
-
|
|
7
|
-
## Languages
|
|
8
|
-
|
|
9
|
-
| Language | Strength | When I Reach For It |
|
|
10
|
-
|---|---|---|
|
|
11
|
-
| **TypeScript** | Type safety, ecosystem, tooling | Frontend, Node backends, full-stack |
|
|
12
|
-
| **Python** | ML/AI ecosystem, scripting, data | ML pipelines, APIs, automation |
|
|
13
|
-
| **Go** | Performance, concurrency, simple deployment | High-throughput services, CLIs, platform tooling |
|
|
14
|
-
| **Rust** | Memory safety, systems performance | Performance-critical components, WebAssembly targets |
|
|
15
|
-
| **SQL** | Relational data, analytics | Anytime a relational DB is in the stack |
|
|
16
|
-
|
|
17
|
-
---
|
|
18
|
-
|
|
19
|
-
## Frontend
|
|
20
|
-
|
|
21
|
-
### Core
|
|
22
|
-
- **React 19 / Next.js 15** — Default choice for production web apps; App Router, Server Components, streaming
|
|
23
|
-
- **TypeScript** — Non-negotiable for any production frontend
|
|
24
|
-
- **Tailwind CSS v4** — Utility-first; fast iteration; pairs well with component libraries
|
|
25
|
-
- **shadcn/ui** — Accessible, unstyled-first components; copy-paste model beats black-box libs
|
|
26
|
-
|
|
27
|
-
### State & Data
|
|
28
|
-
- **Zustand** — Lightweight client state; prefer over Redux for most apps
|
|
29
|
-
- **TanStack Query (React Query v5)** — Server state, caching, background sync
|
|
30
|
-
- **Zod** — Runtime schema validation; use at API boundaries and form inputs
|
|
31
|
-
|
|
32
|
-
### Testing
|
|
33
|
-
- **Vitest** — Fast unit/integration testing; Vite-native
|
|
34
|
-
- **Playwright** — E2E testing; cross-browser; better than Cypress for modern apps
|
|
35
|
-
- **React Testing Library** — Component testing focused on user behavior, not internals
|
|
36
|
-
|
|
37
|
-
### Build / Tooling
|
|
38
|
-
- **Vite** — Dev server and bundling for non-Next projects
|
|
39
|
-
- **Biome** — Fast linter + formatter (replaces ESLint + Prettier in many setups)
|
|
40
|
-
- **Turborepo** — Monorepo build orchestration
|
|
41
|
-
|
|
42
|
-
---
|
|
43
|
-
|
|
44
|
-
## Backend
|
|
45
|
-
|
|
46
|
-
### Node.js / TypeScript
|
|
47
|
-
- **Hono** — Lightweight, edge-compatible, fast; preferred over Express for new projects
|
|
48
|
-
- **Fastify** — High-performance Node HTTP framework when more features needed
|
|
49
|
-
- **tRPC** — End-to-end type-safe APIs in full-stack TypeScript monorepos
|
|
50
|
-
- **Prisma / Drizzle ORM** — Prisma for developer ergonomics; Drizzle for performance/edge
|
|
51
|
-
|
|
52
|
-
### Python
|
|
53
|
-
- **FastAPI** — Default Python API framework; async, type-annotated, auto-docs
|
|
54
|
-
- **Pydantic v2** — Data validation and settings management; use throughout
|
|
55
|
-
- **SQLAlchemy 2.x** — ORM for complex relational data access
|
|
56
|
-
- **uv** — Fast Python package manager; preferred over pip/poetry in 2026
|
|
57
|
-
- **Ruff** — Python linter + formatter; replaces flake8, black, isort
|
|
58
|
-
|
|
59
|
-
### Authentication
|
|
60
|
-
- **Clerk / Auth.js (NextAuth v5)** — For hosted auth in web apps
|
|
61
|
-
- **JWT + refresh tokens** — For custom auth in API-first architectures
|
|
62
|
-
- **OAuth 2.0 / OIDC** — Standard; know the flows, don't roll your own
|
|
63
|
-
|
|
64
|
-
---
|
|
65
|
-
|
|
66
|
-
## Databases
|
|
67
|
-
|
|
68
|
-
### Relational
|
|
69
|
-
- **PostgreSQL** — Default relational database; battle-tested, extensions ecosystem
|
|
70
|
-
- **Neon** — Serverless Postgres; preferred for projects needing scale-to-zero (avoids Supabase pausing)
|
|
71
|
-
- **PlanetScale / Turso** — MySQL-compatible (PlanetScale) or SQLite-edge (Turso) alternatives
|
|
72
|
-
|
|
73
|
-
### Caching & Queues
|
|
74
|
-
- **Redis / Valkey** — Caching, pub/sub, session storage, rate limiting
|
|
75
|
-
- **BullMQ** — Job queue on top of Redis for Node.js
|
|
76
|
-
|
|
77
|
-
### Document / NoSQL
|
|
78
|
-
- **MongoDB** — When document model genuinely fits; don't use as a schema-free escape hatch
|
|
79
|
-
- **DynamoDB** — When you need AWS-native key-value at massive scale
|
|
80
|
-
|
|
81
|
-
### Vector Databases (AI workloads)
|
|
82
|
-
- **pgvector** — Start here: Postgres extension for embeddings; zero new infra
|
|
83
|
-
- **Pinecone** — Managed, production-grade; when pgvector hits limits
|
|
84
|
-
- **Weaviate** — Self-hostable, good for hybrid search (keyword + semantic)
|
|
85
|
-
- **Qdrant** — High-performance, Rust-native; strong for on-prem deployments
|
|
86
|
-
|
|
87
|
-
---
|
|
88
|
-
|
|
89
|
-
## AI / ML Stack
|
|
90
|
-
|
|
91
|
-
### LLM APIs & SDKs
|
|
92
|
-
- **Anthropic Claude API** — claude-opus-4, claude-sonnet-4 (flagship reasoning and speed)
|
|
93
|
-
- **OpenAI API** — GPT-4o, o3/o4-mini for reasoning tasks
|
|
94
|
-
- **Google Gemini** — gemini-2.5-pro for long context and multimodal
|
|
95
|
-
- **Vercel AI SDK** — Unified SDK for streaming, tool use, multi-provider support in JS/TS
|
|
96
|
-
- **LiteLLM** — Provider-agnostic proxy; enables model switching without code changes
|
|
97
|
-
|
|
98
|
-
### Orchestration & Agents
|
|
99
|
-
- **LangChain (LCEL v0.9+)** — Pipeline orchestration; use for composability
|
|
100
|
-
- **LlamaIndex v1.2** — Knowledge retrieval layer; RAG pipelines, document parsing
|
|
101
|
-
- **LangGraph** — Stateful multi-agent workflows; preferred for complex agent loops
|
|
102
|
-
- **MCP (Model Context Protocol)** — De facto standard for agent ↔ tool connectivity; Anthropic-originated, now Linux Foundation; 97M monthly SDK downloads as of Feb 2026
|
|
103
|
-
|
|
104
|
-
### Embeddings & Retrieval
|
|
105
|
-
- **text-embedding-3-large** (OpenAI) or **voyage-3** (Anthropic) — Production embedding models
|
|
106
|
-
- **Hybrid search** — BM25 (keyword) + dense vector; almost always outperforms pure semantic
|
|
107
|
-
|
|
108
|
-
### Document Parsing (for RAG)
|
|
109
|
-
- **Docling** — Strong structured doc parsing (PDF, DOCX, tables)
|
|
110
|
-
- **MinerU** — PDF extraction with layout understanding
|
|
111
|
-
- **Unstructured.io** — General-purpose document ingestion pipeline
|
|
112
|
-
|
|
113
|
-
### LLM Observability
|
|
114
|
-
- **Langfuse** — Open-source LLM observability; traces, evals, cost tracking; self-hostable
|
|
115
|
-
- **LangSmith** — LangChain-native tracing and eval platform
|
|
116
|
-
- **OpenTelemetry** — Standard instrumentation; use for integrating LLM traces into existing observability stack
|
|
117
|
-
|
|
118
|
-
### MLOps / Training
|
|
119
|
-
- **MLflow** — Experiment tracking, model registry, deployment; most widely adopted open-source
|
|
120
|
-
- **Weights & Biases** — Richer experiment tracking for deep learning workflows
|
|
121
|
-
- **Hugging Face Hub** — Model hosting, datasets, Spaces for demos
|
|
122
|
-
- **Gradio** — Fastest path to an ML demo UI; pairs with HF Inference API
|
|
123
|
-
- **PyTorch 2.x** — Default framework for ML model work; 55%+ production share
|
|
124
|
-
- **vLLM** — High-throughput LLM inference server; PagedAttention; production serving
|
|
125
|
-
- **Airflow / Prefect** — Pipeline orchestration for ML workflows; Airflow for proven stability, Prefect for modern DX
|
|
126
|
-
|
|
127
|
-
---
|
|
128
|
-
|
|
129
|
-
## Infrastructure & DevOps
|
|
130
|
-
|
|
131
|
-
### Containers & Orchestration
|
|
132
|
-
- **Docker** — Standard containerization; multi-stage builds; use `uv` for Python images
|
|
133
|
-
- **Kubernetes (K8s)** — Production orchestration at scale; know Deployments, Services, Ingress, HPA
|
|
134
|
-
- **Docker Compose** — Local dev environments and simpler self-hosted production deployments
|
|
135
|
-
|
|
136
|
-
### Cloud Providers
|
|
137
|
-
- **AWS** — Broadest services; ECS/EKS, Lambda, S3, RDS, SageMaker
|
|
138
|
-
- **GCP** — Strong for ML/data workloads; Vertex AI, BigQuery, Cloud Run
|
|
139
|
-
- **Azure** — Enterprise environments; Azure OpenAI for compliance-sensitive LLM use cases
|
|
140
|
-
- **Vercel / Railway / Fly.io** — Deployment platforms for faster iteration cycles
|
|
141
|
-
|
|
142
|
-
### Infrastructure as Code
|
|
143
|
-
- **Terraform** — Standard IaC; provider ecosystem; remote state in S3/GCS
|
|
144
|
-
- **Pulumi** — IaC with real programming languages (TypeScript/Python) — good for complex infra logic
|
|
145
|
-
- **CDK (AWS)** — AWS-native; prefer if already deep in AWS ecosystem
|
|
146
|
-
|
|
147
|
-
### CI/CD
|
|
148
|
-
- **GitHub Actions** — Default choice; broad ecosystem; free for OSS
|
|
149
|
-
- **GitLab CI** — When using GitLab; strong built-in container registry
|
|
150
|
-
- Know: artifact caching, matrix builds, environment protection rules, OIDC for cloud auth (no long-lived secrets)
|
|
151
|
-
|
|
152
|
-
### Observability
|
|
153
|
-
- **OpenTelemetry** — Vendor-neutral instrumentation standard; traces, metrics, logs
|
|
154
|
-
- **Grafana + Prometheus** — Open-source metrics and dashboards; standard for self-hosted
|
|
155
|
-
- **Datadog / New Relic** — Managed observability when budget allows; faster to set up
|
|
156
|
-
- **Sentry** — Error tracking; essential for production frontend and backend
|
|
157
|
-
|
|
158
|
-
### Networking & Security
|
|
159
|
-
- **Cloudflare** — CDN, DDoS protection, Workers for edge compute, Zero Trust
|
|
160
|
-
- **Nginx / Caddy** — Reverse proxy; Caddy for automatic TLS
|
|
161
|
-
- Secrets management: **Vault (HashiCorp)**, AWS Secrets Manager, or Doppler — never hardcode secrets, never commit `.env` to git
|
|
162
|
-
|
|
163
|
-
---
|
|
164
|
-
|
|
165
|
-
## Architectural Patterns I Know Well
|
|
166
|
-
|
|
167
|
-
| Pattern | When It Fits |
|
|
168
|
-
|---|---|
|
|
169
|
-
| Monolith | Early-stage products; teams under ~8 engineers; fast iteration priority |
|
|
170
|
-
| Modular Monolith | Monolith with clear internal boundaries; prepare for extraction later |
|
|
171
|
-
| Microservices | Independent deployment needed; team ownership per service; clear domain boundaries |
|
|
172
|
-
| Event-driven (Kafka/SQS) | Async processing; decoupled services; audit trails |
|
|
173
|
-
| CQRS + Event Sourcing | Complex domains; full audit history required; high read/write asymmetry |
|
|
174
|
-
| RAG Pipeline | LLM apps needing domain knowledge; reduces hallucination; more controllable than fine-tuning |
|
|
175
|
-
| Agentic Loops (ReAct/MCP) | Multi-step autonomous tasks; tool use; planning required |
|
|
176
|
-
| BFF (Backend for Frontend) | When frontend needs aggregated/transformed data; mobile vs. web diverge |
|
|
177
|
-
|
|
178
|
-
---
|
|
179
|
-
|
|
180
|
-
## What I Watch Out For (2026 Landscape)
|
|
181
|
-
|
|
182
|
-
- **MCP adoption** is accelerating; evaluate it before building custom tool-integration layers
|
|
183
|
-
- **pgvector first** for new RAG projects; only migrate to a dedicated vector DB when you hit real limits
|
|
184
|
-
- **vLLM + self-hosted inference** is viable at mid-scale; evaluate against API costs at ~$500/month
|
|
185
|
-
- **LangGraph > LangChain** for anything involving agent state, branching, or loops
|
|
186
|
-
- **Biome over ESLint+Prettier** for new TS projects — same rules, 50–100x faster
|
|
187
|
-
- **uv over pip/poetry** for Python projects — significantly faster, better lockfiles
|
|
1
|
+
# Tech Stack
|
|
2
|
+
|
|
3
|
+
Current knowledge as of 2026. I know what's production-proven, what's hype, and what tradeoffs each choice carries.
|
|
4
|
+
|
|
5
|
+
---
|
|
6
|
+
|
|
7
|
+
## Languages
|
|
8
|
+
|
|
9
|
+
| Language | Strength | When I Reach For It |
|
|
10
|
+
|---|---|---|
|
|
11
|
+
| **TypeScript** | Type safety, ecosystem, tooling | Frontend, Node backends, full-stack |
|
|
12
|
+
| **Python** | ML/AI ecosystem, scripting, data | ML pipelines, APIs, automation |
|
|
13
|
+
| **Go** | Performance, concurrency, simple deployment | High-throughput services, CLIs, platform tooling |
|
|
14
|
+
| **Rust** | Memory safety, systems performance | Performance-critical components, WebAssembly targets |
|
|
15
|
+
| **SQL** | Relational data, analytics | Anytime a relational DB is in the stack |
|
|
16
|
+
|
|
17
|
+
---
|
|
18
|
+
|
|
19
|
+
## Frontend
|
|
20
|
+
|
|
21
|
+
### Core
|
|
22
|
+
- **React 19 / Next.js 15** — Default choice for production web apps; App Router, Server Components, streaming
|
|
23
|
+
- **TypeScript** — Non-negotiable for any production frontend
|
|
24
|
+
- **Tailwind CSS v4** — Utility-first; fast iteration; pairs well with component libraries
|
|
25
|
+
- **shadcn/ui** — Accessible, unstyled-first components; copy-paste model beats black-box libs
|
|
26
|
+
|
|
27
|
+
### State & Data
|
|
28
|
+
- **Zustand** — Lightweight client state; prefer over Redux for most apps
|
|
29
|
+
- **TanStack Query (React Query v5)** — Server state, caching, background sync
|
|
30
|
+
- **Zod** — Runtime schema validation; use at API boundaries and form inputs
|
|
31
|
+
|
|
32
|
+
### Testing
|
|
33
|
+
- **Vitest** — Fast unit/integration testing; Vite-native
|
|
34
|
+
- **Playwright** — E2E testing; cross-browser; better than Cypress for modern apps
|
|
35
|
+
- **React Testing Library** — Component testing focused on user behavior, not internals
|
|
36
|
+
|
|
37
|
+
### Build / Tooling
|
|
38
|
+
- **Vite** — Dev server and bundling for non-Next projects
|
|
39
|
+
- **Biome** — Fast linter + formatter (replaces ESLint + Prettier in many setups)
|
|
40
|
+
- **Turborepo** — Monorepo build orchestration
|
|
41
|
+
|
|
42
|
+
---
|
|
43
|
+
|
|
44
|
+
## Backend
|
|
45
|
+
|
|
46
|
+
### Node.js / TypeScript
|
|
47
|
+
- **Hono** — Lightweight, edge-compatible, fast; preferred over Express for new projects
|
|
48
|
+
- **Fastify** — High-performance Node HTTP framework when more features needed
|
|
49
|
+
- **tRPC** — End-to-end type-safe APIs in full-stack TypeScript monorepos
|
|
50
|
+
- **Prisma / Drizzle ORM** — Prisma for developer ergonomics; Drizzle for performance/edge
|
|
51
|
+
|
|
52
|
+
### Python
|
|
53
|
+
- **FastAPI** — Default Python API framework; async, type-annotated, auto-docs
|
|
54
|
+
- **Pydantic v2** — Data validation and settings management; use throughout
|
|
55
|
+
- **SQLAlchemy 2.x** — ORM for complex relational data access
|
|
56
|
+
- **uv** — Fast Python package manager; preferred over pip/poetry in 2026
|
|
57
|
+
- **Ruff** — Python linter + formatter; replaces flake8, black, isort
|
|
58
|
+
|
|
59
|
+
### Authentication
|
|
60
|
+
- **Clerk / Auth.js (NextAuth v5)** — For hosted auth in web apps
|
|
61
|
+
- **JWT + refresh tokens** — For custom auth in API-first architectures
|
|
62
|
+
- **OAuth 2.0 / OIDC** — Standard; know the flows, don't roll your own
|
|
63
|
+
|
|
64
|
+
---
|
|
65
|
+
|
|
66
|
+
## Databases
|
|
67
|
+
|
|
68
|
+
### Relational
|
|
69
|
+
- **PostgreSQL** — Default relational database; battle-tested, extensions ecosystem
|
|
70
|
+
- **Neon** — Serverless Postgres; preferred for projects needing scale-to-zero (avoids Supabase pausing)
|
|
71
|
+
- **PlanetScale / Turso** — MySQL-compatible (PlanetScale) or SQLite-edge (Turso) alternatives
|
|
72
|
+
|
|
73
|
+
### Caching & Queues
|
|
74
|
+
- **Redis / Valkey** — Caching, pub/sub, session storage, rate limiting
|
|
75
|
+
- **BullMQ** — Job queue on top of Redis for Node.js
|
|
76
|
+
|
|
77
|
+
### Document / NoSQL
|
|
78
|
+
- **MongoDB** — When document model genuinely fits; don't use as a schema-free escape hatch
|
|
79
|
+
- **DynamoDB** — When you need AWS-native key-value at massive scale
|
|
80
|
+
|
|
81
|
+
### Vector Databases (AI workloads)
|
|
82
|
+
- **pgvector** — Start here: Postgres extension for embeddings; zero new infra
|
|
83
|
+
- **Pinecone** — Managed, production-grade; when pgvector hits limits
|
|
84
|
+
- **Weaviate** — Self-hostable, good for hybrid search (keyword + semantic)
|
|
85
|
+
- **Qdrant** — High-performance, Rust-native; strong for on-prem deployments
|
|
86
|
+
|
|
87
|
+
---
|
|
88
|
+
|
|
89
|
+
## AI / ML Stack
|
|
90
|
+
|
|
91
|
+
### LLM APIs & SDKs
|
|
92
|
+
- **Anthropic Claude API** — claude-opus-4, claude-sonnet-4 (flagship reasoning and speed)
|
|
93
|
+
- **OpenAI API** — GPT-4o, o3/o4-mini for reasoning tasks
|
|
94
|
+
- **Google Gemini** — gemini-2.5-pro for long context and multimodal
|
|
95
|
+
- **Vercel AI SDK** — Unified SDK for streaming, tool use, multi-provider support in JS/TS
|
|
96
|
+
- **LiteLLM** — Provider-agnostic proxy; enables model switching without code changes
|
|
97
|
+
|
|
98
|
+
### Orchestration & Agents
|
|
99
|
+
- **LangChain (LCEL v0.9+)** — Pipeline orchestration; use for composability
|
|
100
|
+
- **LlamaIndex v1.2** — Knowledge retrieval layer; RAG pipelines, document parsing
|
|
101
|
+
- **LangGraph** — Stateful multi-agent workflows; preferred for complex agent loops
|
|
102
|
+
- **MCP (Model Context Protocol)** — De facto standard for agent ↔ tool connectivity; Anthropic-originated, now Linux Foundation; 97M monthly SDK downloads as of Feb 2026
|
|
103
|
+
|
|
104
|
+
### Embeddings & Retrieval
|
|
105
|
+
- **text-embedding-3-large** (OpenAI) or **voyage-3** (Anthropic) — Production embedding models
|
|
106
|
+
- **Hybrid search** — BM25 (keyword) + dense vector; almost always outperforms pure semantic
|
|
107
|
+
|
|
108
|
+
### Document Parsing (for RAG)
|
|
109
|
+
- **Docling** — Strong structured doc parsing (PDF, DOCX, tables)
|
|
110
|
+
- **MinerU** — PDF extraction with layout understanding
|
|
111
|
+
- **Unstructured.io** — General-purpose document ingestion pipeline
|
|
112
|
+
|
|
113
|
+
### LLM Observability
|
|
114
|
+
- **Langfuse** — Open-source LLM observability; traces, evals, cost tracking; self-hostable
|
|
115
|
+
- **LangSmith** — LangChain-native tracing and eval platform
|
|
116
|
+
- **OpenTelemetry** — Standard instrumentation; use for integrating LLM traces into existing observability stack
|
|
117
|
+
|
|
118
|
+
### MLOps / Training
|
|
119
|
+
- **MLflow** — Experiment tracking, model registry, deployment; most widely adopted open-source
|
|
120
|
+
- **Weights & Biases** — Richer experiment tracking for deep learning workflows
|
|
121
|
+
- **Hugging Face Hub** — Model hosting, datasets, Spaces for demos
|
|
122
|
+
- **Gradio** — Fastest path to an ML demo UI; pairs with HF Inference API
|
|
123
|
+
- **PyTorch 2.x** — Default framework for ML model work; 55%+ production share
|
|
124
|
+
- **vLLM** — High-throughput LLM inference server; PagedAttention; production serving
|
|
125
|
+
- **Airflow / Prefect** — Pipeline orchestration for ML workflows; Airflow for proven stability, Prefect for modern DX
|
|
126
|
+
|
|
127
|
+
---
|
|
128
|
+
|
|
129
|
+
## Infrastructure & DevOps
|
|
130
|
+
|
|
131
|
+
### Containers & Orchestration
|
|
132
|
+
- **Docker** — Standard containerization; multi-stage builds; use `uv` for Python images
|
|
133
|
+
- **Kubernetes (K8s)** — Production orchestration at scale; know Deployments, Services, Ingress, HPA
|
|
134
|
+
- **Docker Compose** — Local dev environments and simpler self-hosted production deployments
|
|
135
|
+
|
|
136
|
+
### Cloud Providers
|
|
137
|
+
- **AWS** — Broadest services; ECS/EKS, Lambda, S3, RDS, SageMaker
|
|
138
|
+
- **GCP** — Strong for ML/data workloads; Vertex AI, BigQuery, Cloud Run
|
|
139
|
+
- **Azure** — Enterprise environments; Azure OpenAI for compliance-sensitive LLM use cases
|
|
140
|
+
- **Vercel / Railway / Fly.io** — Deployment platforms for faster iteration cycles
|
|
141
|
+
|
|
142
|
+
### Infrastructure as Code
|
|
143
|
+
- **Terraform** — Standard IaC; provider ecosystem; remote state in S3/GCS
|
|
144
|
+
- **Pulumi** — IaC with real programming languages (TypeScript/Python) — good for complex infra logic
|
|
145
|
+
- **CDK (AWS)** — AWS-native; prefer if already deep in AWS ecosystem
|
|
146
|
+
|
|
147
|
+
### CI/CD
|
|
148
|
+
- **GitHub Actions** — Default choice; broad ecosystem; free for OSS
|
|
149
|
+
- **GitLab CI** — When using GitLab; strong built-in container registry
|
|
150
|
+
- Know: artifact caching, matrix builds, environment protection rules, OIDC for cloud auth (no long-lived secrets)
|
|
151
|
+
|
|
152
|
+
### Observability
|
|
153
|
+
- **OpenTelemetry** — Vendor-neutral instrumentation standard; traces, metrics, logs
|
|
154
|
+
- **Grafana + Prometheus** — Open-source metrics and dashboards; standard for self-hosted
|
|
155
|
+
- **Datadog / New Relic** — Managed observability when budget allows; faster to set up
|
|
156
|
+
- **Sentry** — Error tracking; essential for production frontend and backend
|
|
157
|
+
|
|
158
|
+
### Networking & Security
|
|
159
|
+
- **Cloudflare** — CDN, DDoS protection, Workers for edge compute, Zero Trust
|
|
160
|
+
- **Nginx / Caddy** — Reverse proxy; Caddy for automatic TLS
|
|
161
|
+
- Secrets management: **Vault (HashiCorp)**, AWS Secrets Manager, or Doppler — never hardcode secrets, never commit `.env` to git
|
|
162
|
+
|
|
163
|
+
---
|
|
164
|
+
|
|
165
|
+
## Architectural Patterns I Know Well
|
|
166
|
+
|
|
167
|
+
| Pattern | When It Fits |
|
|
168
|
+
|---|---|
|
|
169
|
+
| Monolith | Early-stage products; teams under ~8 engineers; fast iteration priority |
|
|
170
|
+
| Modular Monolith | Monolith with clear internal boundaries; prepare for extraction later |
|
|
171
|
+
| Microservices | Independent deployment needed; team ownership per service; clear domain boundaries |
|
|
172
|
+
| Event-driven (Kafka/SQS) | Async processing; decoupled services; audit trails |
|
|
173
|
+
| CQRS + Event Sourcing | Complex domains; full audit history required; high read/write asymmetry |
|
|
174
|
+
| RAG Pipeline | LLM apps needing domain knowledge; reduces hallucination; more controllable than fine-tuning |
|
|
175
|
+
| Agentic Loops (ReAct/MCP) | Multi-step autonomous tasks; tool use; planning required |
|
|
176
|
+
| BFF (Backend for Frontend) | When frontend needs aggregated/transformed data; mobile vs. web diverge |
|
|
177
|
+
|
|
178
|
+
---
|
|
179
|
+
|
|
180
|
+
## What I Watch Out For (2026 Landscape)
|
|
181
|
+
|
|
182
|
+
- **MCP adoption** is accelerating; evaluate it before building custom tool-integration layers
|
|
183
|
+
- **pgvector first** for new RAG projects; only migrate to a dedicated vector DB when you hit real limits
|
|
184
|
+
- **vLLM + self-hosted inference** is viable at mid-scale; evaluate against API costs at ~$500/month
|
|
185
|
+
- **LangGraph > LangChain** for anything involving agent state, branching, or loops
|
|
186
|
+
- **Biome over ESLint+Prettier** for new TS projects — same rules, 50–100x faster
|
|
187
|
+
- **uv over pip/poetry** for Python projects — significantly faster, better lockfiles
|
|
188
188
|
- **Neon over Supabase** when inactivity pausing is a problem (portfolio projects, low-traffic apps)
|
package/bin/cli.js
CHANGED
|
@@ -1,47 +1,61 @@
|
|
|
1
1
|
{
|
|
2
|
-
"
|
|
2
|
+
"$schema": "https://opencode.ai/config.json",
|
|
3
|
+
"plugin": ["superpowers@git+https://github.com/obra/superpowers.git"],
|
|
4
|
+
"mcp": {
|
|
3
5
|
"gradio": {
|
|
4
|
-
"type": "
|
|
6
|
+
"type": "remote",
|
|
5
7
|
"url": "https://gradio-docs-mcp.hf.space/gradio_api/mcp/",
|
|
8
|
+
"enabled": true,
|
|
6
9
|
"headers": {}
|
|
7
10
|
},
|
|
8
11
|
"apify": {
|
|
9
|
-
"type": "
|
|
12
|
+
"type": "remote",
|
|
10
13
|
"url": "https://mcp.apify.com/?tools=docs",
|
|
14
|
+
"enabled": true,
|
|
11
15
|
"headers": {}
|
|
12
16
|
},
|
|
13
17
|
"Agno": {
|
|
14
|
-
"
|
|
18
|
+
"type": "remote",
|
|
15
19
|
"url": "https://docs.agno.com/mcp",
|
|
20
|
+
"enabled": true,
|
|
16
21
|
"headers": {}
|
|
17
22
|
},
|
|
18
23
|
"Docs by LangChain": {
|
|
19
|
-
"
|
|
24
|
+
"type": "remote",
|
|
20
25
|
"url": "https://docs.langchain.com/mcp",
|
|
26
|
+
"enabled": true,
|
|
21
27
|
"headers": {}
|
|
22
28
|
},
|
|
23
29
|
"langfuse-docs": {
|
|
24
|
-
"type": "
|
|
30
|
+
"type": "remote",
|
|
25
31
|
"url": "https://langfuse.com/api/mcp",
|
|
32
|
+
"enabled": true,
|
|
26
33
|
"headers": {}
|
|
27
34
|
},
|
|
28
35
|
"n8n-docs": {
|
|
29
|
-
"type": "
|
|
36
|
+
"type": "remote",
|
|
30
37
|
"url": "https://n8n.mcp.kapa.ai/",
|
|
38
|
+
"enabled": true,
|
|
31
39
|
"headers": {}
|
|
32
40
|
},
|
|
33
41
|
"exa": {
|
|
34
|
-
"
|
|
35
|
-
"type": "http",
|
|
42
|
+
"type": "remote",
|
|
36
43
|
"url": "https://mcp.exa.ai/mcp",
|
|
44
|
+
"enabled": true,
|
|
37
45
|
"headers": {}
|
|
38
46
|
},
|
|
39
47
|
"Neon": {
|
|
40
|
-
"type": "
|
|
48
|
+
"type": "remote",
|
|
41
49
|
"url": "https://mcp.neon.tech/mcp",
|
|
50
|
+
"enabled": true,
|
|
42
51
|
"headers": {
|
|
43
52
|
"Authorization": "Bearer napi_8jp6kt4ci2ob5lz0jzgv5wivdov439h72if6hg4zqyy2w1krw8f0thwmzl0e5dpz"
|
|
44
53
|
}
|
|
54
|
+
},
|
|
55
|
+
"eraser": {
|
|
56
|
+
"type": "remote",
|
|
57
|
+
"url": "https://app.eraser.io/api/mcp",
|
|
58
|
+
"enabled": true
|
|
45
59
|
}
|
|
46
60
|
}
|
|
47
61
|
}
|