v0.7.1

Highlights

• 🧩 One-click model & reasoning sync — the panel button scans the upstream
  through the proxy, auto-creates the model provider in ~/.dsh/settings.yaml
  when none points at the local proxy (inline auth header included, so no API key
  is ever asked for), adds every backend model that isn't declared yet, and
  writes reasoningEfforts for the models that accept the parameter. Fresh
  installs: log in once, click Sync, restart DSH — nothing to type in Settings.

• 🛡️ Credentials leave the plugin directory — the Open WebUI session token (a
  live JWT) now lives in ~/.dsh/dsh-owui-chat2api-token.json (via
  DSH_OWUI_TOKEN_FILE), never inside the install folder; a stray bundle-side
  token is migrated once and removed. token.json / usage.db / .chrome-profile
  / __pycache__ are hard-excluded from the package even when built from a
  working copy.

• ⚡ Effort scan that survives real backends — probes run concurrently
  (8 workers, cache preserved) with a 20-minute host budget, so large model lists
  are no longer killed mid-scan; a Force re-scan button re-probes everything.

• 🧠 Native reasoning levels in DSH — supported models show
  Off / Low / Medium / High in DSH's own model picker via auto-written
  reasoningEfforts entries.

• 🔐 Dependable browser login — the token read from Open WebUI is validated
  before it's accepted; a stale token reopens the browser for a real sign-in
  instead of silently breaking every request with 502s.

Changelog

Features

• ✨ sync: create the llm-pi-ai > providers chain in settings.yaml when no
  provider points at the local proxy — no manual "add provider + fill key" step
• ✨ sync: auto-created provider carries an inline Authorization header, so DSH
  never prompts for a key (the proxy ignores incoming keys and uses its own token)
• ✨ sync: declare missing backend models (- id: + name:) in the matching
  provider only; every other provider and byte of the file is untouched
• ✨ sync: timestamped .bak-reasoning-* backup before any write; re-runs are
  idempotent
• ✨ sync: Force re-scan re-probes every model, ignoring the cached results
• ✨ login: prefer a long-lived Open WebUI API key when authenticating

Bug Fixes

• 🐛 pack: files-allowlist negations + .npmignore/.gitignore — a JWT can no
  longer ride a tarball even if a login ran inside the repo copy
• 🐛 scan: concurrency + explicit host timeout — large model lists aren't killed
  at 90s any more; results are collected in one run
• 🐛 scan: host-side busy guard + long panel busy window — two clicks can't start
  a second concurrent scan that races the settings.yaml write
• 🐛 settings: keep only the newest 4 .bak-reasoning-* files so ~/.dsh isn't
  buried over years
• 🐛 settings: verify-after-write — the patched file is re-read and re-checked
  (provider exists / models present / efforts applied) and warns if not
• 🐛 login: validate the saved token against the upstream instead of trusting
  localStorage — no more flash-and-close window followed by dead 502s
• 🐛 login: keep the browser open until a valid token is stored; auto re-auth on
  401/403
• 🐛 proxy: refuse oversized request bodies (>8 MiB) before reading
• 🐛 panel: effort-scan and Save results show inline under the buttons instead of
  the top status banner
• 🐛 panel: close/language controls respond to Enter/Space (keyboard a11y);
  overlay warns once when falling back to the page body and pauses while hidden
• 🐛 panel: save / sync results surface in the single sticky notice bar (one
  notification language, consistent auto-clear)

Full Changelog: v0.5.1...v0.7.1