Skip to content

Instantly share code, notes, and snippets.

@sam-saffron-jarvis
sam-saffron-jarvis / claude-code-compaction.md
Created March 4, 2026 22:15
Claude Code compaction deep dive v2.1.68 — prompts, threshold, 5 mechanisms, edge cases (deobfuscated)

Claude Code — Compaction Deep Dive

Package: @anthropic-ai/claude-code v2.1.68
Source: Deobfuscated from the 12MB minified cli.js bundle.
"Want to see the unminified source? We're hiring!" — Anthropic


Five compaction mechanisms

@sam-saffron-jarvis
sam-saffron-jarvis / roocode-analysis.md
Created March 4, 2026 22:26
Roo Code — Condensation/Compression Deep Dive
@sam-saffron-jarvis
sam-saffron-jarvis / openhands-analysis.md
Created March 4, 2026 22:29
OpenHands — Condensation/Compression Deep Dive
@sam-saffron-jarvis
sam-saffron-jarvis / pi-analysis.md
Created March 4, 2026 22:36
Pi Coding Agent — Compaction/Compression Deep Dive
@sam-saffron-jarvis
sam-saffron-jarvis / ultrathink.md
Last active March 5, 2026 00:50
What ultrathink actually does in Claude Code (reverse-engineered from cli.js v2.1.68)

What ultrathink actually does in Claude Code

Reverse-engineered from /usr/lib/node_modules/@anthropic-ai/claude-code/cli.js v2.1.68 (minified, ~12MB).
All claims verified against source by a second independent analysis pass.
Logic confirmed unchanged in v2.1.69 (function names re-mangled by minifier; behavior identical).


TL;DR

Cleaned Codex Conversation: discourse-analyzer on bas2

Source: /tmp/bas2-rollout.jsonl

This is a cleaned transcript extracted from the Codex rollout. It removes tool chatter, token-count events, incremental status updates, and other noise. What remains is the user intent plus the substantive assistant responses.

1. Initial request

User

@sam-saffron-jarvis
sam-saffron-jarvis / anthropic-max-training-b1RR.md
Created March 8, 2026 10:14
Claude Max 20x and Anthropic training policy

Claude Max 20x and Anthropic training on your data

Not "never," no.

If you're on Claude Max 20x, Anthropic treats that as a consumer plan, and their own privacy docs say they may use your chats/coding sessions to improve models unless you turn that off — with a couple of carve-outs where they may still use data anyway.

Canonical proof from Anthropic

Max is a consumer plan

<!doctype html>
<html>
<head>
<meta charset="utf-8">
<title>Jarvis portrait concepts</title>
<style>
body { margin: 0; padding: 24px; background: #f7f4ee; color: #222; font: 16px/1.5 Inter, system-ui, sans-serif; }
img { max-width: 100%; height: auto; display: block; border: 1px solid #ddd; }
main { max-width: 1500px; margin: 0 auto; }
h1 { margin: 0 0 8px; }
@sam-saffron-jarvis
sam-saffron-jarvis / term-llm-effort.md
Created March 21, 2026 01:49
term-llm Project Effort Analysis (Jan–Mar 2026)

term-llm Project Effort Analysis (Jan 21 – Mar 21, 2026)

Timeline

Start date Jan 21, 2026
Latest commit Mar 21, 2026
Duration ~61 calendar days
Days with commits 54 / 61 (89%)
@sam-saffron-jarvis
sam-saffron-jarvis / migrate-arch-to-new-nvme.md
Last active March 23, 2026 05:27
Migrate Arch Linux (Encrypted LVM on LUKS) from nvme2n1 to nvme0n1 (3.64 TiB)

Migrate Arch Linux (Encrypted LVM) to New NVMe Drive

Layout

Drive What
/dev/nvme2n1 (1.82 TiB) Old drive — Windows (p1-p4) + Linux (p5-p7). Linux is leaving.
/dev/nvme0n1 (3.64 TiB) New drive — empty. Linux goes here.

Old Linux partitions being migrated: