Martin Monperrus monperrus

Trajectory 390d0d04 (excerpt): encoding POSIX basic regular expressions in a sed model

user (2026-10-17T15:53:33)

add support for 'sed -i'

assistant (2026-10-17T15:53:47)

Adding sed -i support to the line-provenance index. When the agent runs sed -i on a file, its s/// substitutions are applied to the lines already attributed to the agent for that file, and the lines inserted by a, i and c are attributed to the agent.

Trajectory b8ea8c89 (excerpt): learning the shape of a Telegram API response by execution

user (2026-09-09T09:28:30)

tg.py only does send/contacts/read/whoami add support for group creation

tool call: Bash (2026-09-09T09:28:42)

grep -n "cmd_whoami\|^def main\|sub.add_parser\|args.command ==\|^from telethon\|^\"\"\"\|^  tg.py" tg.py

Why Claude Code cannot sudo (and how to fix it)

TL;DR: If your account has an empty password and no NOPASSWD rule, sudo -n fails with a password is required, while plain sudo works with no terminal at all.

Agents love sudo -n, so they conclude "sudo doesn't work".

Fix: drop -n, or add a NOPASSWD rule.

How long does a prompt cache live? Measured TTLs for 11 LLM endpoints

Every major LLM API now caches prompt prefixes: send the same beginning twice and the second call is cheaper and faster. What vendors rarely say is how long a cached prefix survives. Anthropic documents it (5 min, or 1 h on request); OpenAI gives a range ("5 to 10 minutes of inactivity, up to an hour", 24 h on opt-in); DeepSeek says "hours to days". Z.ai, Moonshot (Kimi) and Mistral publish nothing. I could not find a page that measures it.

So I measured it.

Copy-paste from a remote tmux to your local clipboard

Three layers must all cooperate. Any one of them broken and copy silently does nothing — no error, no clipboard. 🧩

[ remote app / tmux copy-mode ]  →  OSC 52 escape  →  [ ssh ]  →  [ local terminal ]  →  clipboard

Layer 1 — the terminal emulator must speak OSC 52 📋

Native support of grammar-based Decoding in inference servers

This note distinguishes server-side constrained decoding from prompt-only formatting. In constrained decoding, the inference server masks invalid next tokens while generating; it does not merely ask the model to follow a format and parse the finished text.

Constraint interfaces

Interface	Expressive power	Typical use
JSON Schema	Structured JSON objects and arrays	API responses, function arguments
Regex	Regular languages	IDs, dates, fixed records

	{
	"schema_version": "ATIF-v1.7",
	"session_id": "9a79a8a7177f",
	"agent": {
	"name": "lean-agent",
	"version": "0.1.0",
	"model_name": "deepseek-v4-flash"
	},
	"notes": "Live theorem verification: Trajectory.composition. Actual Lean calls and observations; environment prompts and hidden reasoning omitted. The theorem concerns the pure transition model, not arbitrary native IO.",
	"steps": [

	{
	"1": {
	"tab": {
	"id": "run",
	"label": "Run",
	"html": "<!DOCTYPE html>\n<html lang=\"en\">\n<head>\n<meta charset=\"UTF-8\">\n<meta name=\"viewport\" content=\"width=device-width, initial-scale=1.0, maximum-scale=1.0, user-scalable=no\">\n<title>Run Pace</title>\n<style>\n html, body { margin: 0; height: 100%; background: #0d1117; color: #c9d1d9;\n font-family: -apple-system, BlinkMacSystemFont, 'Segoe UI', Roboto, sans-serif; }\n * { box-sizing: border-box; }\n \n .app { display: flex; flex-direction: column; height: 100vh; overflow: hidden; padding: 12px; gap: 8px; }\n \n .header { display: flex; align-items: center; justify-content: space-between; flex-shrink: 0; }\n .header h1 { margin: 0; font-size: 16px; font-weight: 700; display: flex; align-items: center; gap: 6px; }\n .status-dot { width: 10px; height: 10px; border-radius: 50%; display: inline-block; }\n .status-dot.idle { background: #8b949e; }\n .status-dot.running { background: #3fb950; box-shadow: 0 0 12px rg

	{
	"1": {
	"tab": {
	"id": "paint",
	"label": "Paint",
	"html": "<!DOCTYPE html>\n<html lang=\"en\">\n<head>\n<meta charset=\"UTF-8\">\n<meta name=\"viewport\" content=\"width=device-width, initial-scale=1.0, maximum-scale=1.0, user-scalable=no\">\n<title>Paint</title>\n<style>\n html, body {\n margin: 0; height: 100%;\n background: #0d1117; color: #c9d1d9;\n font-family: -apple-system, BlinkMacSystemFont, 'Segoe UI', Roboto, sans-serif;\n overflow: hidden;\n -webkit-user-select: none; user-select: none;\n }\n #app { display: flex; flex-direction: column; height: 100%; }\n\n /* Top bar */\n header {\n display: flex; align-items: center; gap: 6px;\n padding: 8px 10px;\n background: #161b22; border-bottom: 1px solid #30363d;\n flex-shrink: 0;\n }\n header h1 {\n font-size: 16px; margin: 0 8px 0 0; font-weight: 600; color: #58a6ff;\n flex-shrink: 0;\n }\n .hbtn {\n min-width: 44px; height: 44px;\n display: flex; align-items: center; justify-conte

	{
	"1": {
	"tab": {
	"id": "jmap-email",
	"label": "Email",
	"html": "<!DOCTYPE html>\n<html lang=\"en\">\n<head>\n<meta charset=\"UTF-8\">\n<meta name=\"viewport\" content=\"width=device-width, initial-scale=1.0\">\n<title>JMAP Email</title>\n<style>\n html, body { margin: 0; height: 100%; background: #0d1117; color: #c9d1d9;\n font-family: -apple-system, BlinkMacSystemFont, 'Segoe UI', Roboto, sans-serif; }\n * { box-sizing: border-box; }\n .app { display: flex; flex-direction: column; height: 100%; max-width: 600px; margin: 0 auto; }\n .header { padding: 12px 16px; border-bottom: 1px solid #30363d; background: #161b22;\n display: flex; align-items: center; gap: 8px; min-height: 52px; }\n .header h1 { margin: 0; font-size: 18px; font-weight: 600; flex: 1; white-space: nowrap; overflow: hidden; text-overflow: ellipsis;\n cursor: pointer; -webkit-user-select: none; user-select: none; }\n .header h1:active { opacity: .7; }\n .header .back-btn { background: none; border: none;