Reference: Single-User Coding Pair Programmer
referencecoding-assistant
framework
Agent Framework
harness
Harness / IDE
memory
Memory
CC
claude-code-memory
retrieval
Retrieval / Vector Store
eval
Eval
observability
Observability / Tracing
sandbox
Sandbox / Code Execution
deployment
Deployment Target
LO
local
mcp_servers
MCP Servers
Architectural notes
Reference architecture for a single engineer using Claude Code as their daily pair programmer. Why these picks: - Claude Code provides the harness primitives (skills, hooks, subagents, plugins) out of the box; no custom orchestration needed. - Sonnet 4.6 as default; Opus 4.7 invoked via /fast or subagent for hard reasoning. - pgvector keeps embeddings next to other project data in Postgres. - promptfoo for CI-gated prompt regression tests. - E2B for sandboxed execution of agent-written code. Common gotchas: - Skill spam: too many skills load too much context; curate aggressively. - Hook misconfiguration can silently block tool calls.
Embed this stack
Show off your stack on your engineering blog or company page.
<iframe src="http://testserver/embed/stacks/coding-assistant-pair" width="100%" height="380" frameborder="0" style="border: 1px solid #334155; border-radius: 8px;"></iframe>