Everything about pi, the terminal coding agent you can extend: sessions and the TUI, models and providers, the TypeScript extension API, skills, prompt templates, pi packages, compaction and token economics, and headless SDK/RPC modes — then the oh-my-pi (omp) fork, and how to decide between vanilla pi and omp.
What a coding harness is, where pi comes from (Mario Zechner, pi-mono, Earendil Inc.), installing and authenticating, the four areas of the TUI, and the token dashboard in the footer.
8 minSession storage as JSONL trees, /tree /fork /clone /resume, editor features (@ refs, ! vs !! commands), the message queue, and the token-saving power of ephemeral and forked sessions.
8 minThe providers system, custom providers and models in models.json (API dialects, cost fields, key resolution), thinking levels and budgets, and local inference with llama.cpp — the biggest token lever of all.
8 minThe extension API: locations, registerTool/registerCommand/registerShortcut, lifecycle events, ctx.ui, state persistence, and extension patterns — including extensions as a token-control plane (ctx.compact, ctx.getContextUsage, input rewriting).
10 minThe Agent Skills standard, SKILL.md structure and locations, on-demand loading, and prompt templates with arguments — plus why on-demand skills are the textbook token-saving pattern.
8 minDistribution for extensions, skills, templates, and themes via npm/git, package filtering and scoping, the pi.dev gallery, and why every installed resource is a standing token tax.
6 minHow auto-compaction works (reserveTokens, keepRecentTokens, iterative summaries, branch summarization), prompt-cache awareness, custom summarizers via extensions, and a ten-item token-saving checklist.
10 minOne-shot prompts, JSON event stream and RPC modes, the @earendil-works SDK, sessions as portable JSONL data — and why scripted, deterministic pi runs are a token-efficiency win.
6 minWhat omp is, how to install it, the feature tour (hashline edits, eval kernels, LSP/DAP, subagents, advisor, stream rules, memory, collab), model roles, and the token-efficiency claims behind the headline numbers.
10 minHead-to-head comparison, decision criteria for each harness, token-efficiency trade-offs (four-tool minimalism vs 31-tool batteries-included), and practical advice to run both.
8 minTwenty questions across all ten lessons with a scored, graded result panel. The final retrieval-practice pass.
10 min