hermes - 💡(How to fix) Fix [Bug]: sanitize_title removes the ESC anchor before the sequence is stripped — escape body survives in session titles

Summary

Session titles generated from a first message that contains terminal escape sequences (ANSI CSI / OSC) keep the body of the sequence as visible garbage — e.g. a pasted \x1b[31mRed\x1b[0m is stored as the title [31mRed[0m, and \x1b]11;rgb:2828/2c2c/3434\x07 as ]11;rgb:2828/2c2c/3434.

Repro

1. Start a new session via a chat gateway (observed via Telegram).
2. Make the first message contain escapes, e.g. \x1b[31mRed\x1b[0m or \x1b]11;rgb:2828/2c2c/3434\x07.
3. Auto-title generation runs on the first exchange.
4. The stored/displayed title contains the residue above.

Root cause

SessionDB.sanitize_title() removes control bytes with:

cleaned = re.sub(r'[\x00-\x08\x0b\x0c\x0e-\x1f\x7f]', '', title)

0x1b (ESC, within \x0e-\x1f) and 0x07 (BEL, within \x00-\x08) are deleted, but the printable sequence body ([31m, ]11;…) is ordinary ASCII and survives. The repo already ships tools/ansi_strip.py::strip_ansi (full ECMA-48, fast-path no-op on clean text) — used by terminal_tool / code_execution_tool / process_registry — which removes whole sequences, but it must run before the control-char strip, while the ESC anchor is still present.

Fix

from tools.ansi_strip import strip_ansi
...
cleaned = strip_ansi(title)                                  # ESC anchor still present
cleaned = re.sub(r'[\x00-\x08\x0b\x0c\x0e-\x1f\x7f]', '', cleaned)

Complementary, not redundant: strip_ansi removes escape-anchored sequences + C1 controls; the existing regex still removes lone C0 controls it doesn't touch. All-escape input collapses to '' → existing return None. The fast-path makes it a no-op for clean CJK/emoji/RTL titles.

Scope

Session titles are not FTS5-indexed (only message content is, and that is stored raw by design), so this is a title display/storage fix only.

I have a fix + regression tests ready and will open a PR.