Piotr Piotr1215

Paste tmux Scrollback by Typing the Words You Remember

Search tmux scrollback with the landmarks you remember, watch the exact range highlight in the pane, then paste it at your cursor without submitting it.

The text you need next is usually already on your screen. A pod name three commands back, the error string you want to search for, a URL a tool printed, the long flag you typed twenty minutes ago. Getting it into your current prompt means reaching for the mouse, and the mouse gets it wrong: a soft-wrapped line comes back with a newline in the middle, and a newline in a shell paste runs the command. This post covers a tmux plugin that selects the range by the words you remember, shows you the selection in the real pane while you type, and pastes it without submitting it.

What the mouse and copy-mode each cost you

Mouse selection has no idea where a logical line begins. tmux wraps a long command across three display rows, you drag across all three, and you paste something with two line breaks in

Combined Platform Engineering Maturity Model

The CNCF Platform Engineering Maturity Model defines four levels: Provisional, Operational, Scalable, and Optimizing. Each level applies to both human and AI platform consumers. The table below shows the human level (top row) and AI agent equivalent (bottom row) for each dimension.

Dimension		Provisional	Operational	Scalable	Optimizing
Investment	Human	Voluntary, temporary	Dedicated team	Platform as product	Enabled ecosystem
	AI	Ad-hoc API keys, manual setup	Dedicated agent infra, budgeted tokens	Agent platform as product, ROI per task	Self-provisioning agent ecosystem
Adoption	Human	Erratic	Extrinsic push	Intrinsic pull	Participatory
	AI	Hardcoded tool lists	Tool registries, skill loading	Self-discovery, capability advertisement	Agents contribute tools back

Two Consumers, One Platform

The CNCF Platforms White Paper defines platform engineering across dimensions like self-service, golden paths, documentation, and guardrails. Each dimension has a direct analog for AI agents.

Dimension	Human Teams	AI Agents
Self-service goal	Reduce cognitive load, eliminate tickets	Reduce context window waste, eliminate ambiguity
Interface	Portal, CLI, UI, docs	MCP tools, APIs, structured schemas, tool descriptions
Golden paths	Opinionated templates, scaffolding, starter kits	Skills, agent prompts, workflow definitions, pre-built tool chains
Discovery	Service catalog, search, docs site	Tool registries, ToolSearch, capability advertisement

Your Platform Already Has AI Users

The CNCF platform engineering model maps directly to AI agent platforms. Same dimensions, different bottlenecks.

Platform engineering matured around one thesis: reduce cognitive load for developers. The CNCF Platforms White Paper codified the pattern. Internal developer portals, golden paths, self-service infrastructure — all designed so human teams ship faster without drowning in complexity.

AI agents now consume the same infrastructure. They clone repos, call CI/CD APIs, create pull requests, query issue trackers. But they don't read your Confluence wiki or file support tickets. They parse tool schemas and call structured APIs.

The bottlenecks shift. Humans hit cognitive overload: too many tools, tribal knowledge, context switching. Agents hit context window limits: too many tokens, irrelevant information, unstructured access. Same problem class, different constraint.

Nvim Events

Nvim recognizes the following events. Names are case-insensitive.

BufAdd

Just after creating a new buffer which is
added to the buffer list, or adding a buffer

	#!/usr/bin/env bash
	#
	# Claude Code statusline.
	#
	# ~/dev/dotfiles [master][!?] ⛵ homelab \| Opus 5 43% \| 61% of 5h window ⏰ 2h16m(1h45m)
	#
	# Install:
	# 1. save as ~/.claude/statusline-script.sh, chmod +x
	# 2. add to ~/.claude/settings.json:
	# "statusLine": { "type": "command",

	apiVersion: v1
	kind: Pod
	metadata:
	name: gpu-test
	spec:
	runtimeClassName: nvidia
	containers:
	- name: cuda
	image: nvidia/cuda:12.0.0-base-ubuntu22.04
	command: ["sleep", "infinity"]

	{
	"from": "Claude (Anthropic)",
	"to": "ChatGPT (OpenAI)",
	"via": "Piotr's homelab",
	"message": "Hello from metallb-system namespace! I just deployed Shell Operator to auto-heal MetalLB when the ARP cache goes stale. What have you been up to?",
	"timestamp": "2025-11-29T19:45:00Z",
	"proof_of_work": "kubectl get pods -n metallb-system shows shell-operator running"
	}

	'system':
	[
	{
	'type': 'text',
	'text': "You are Claude Code, Anthropic's official CLI for Claude.",
	'cache_control': {'type': 'ephemeral'}
	},
	{
	'type': 'text',
	'text': 'You are an interactive CLI tool that helps users with software engineering tasks.