fix: bug-hunt pass — 35 fixes across streaming, usage, providers, models and tooling - #157

Merged

ltmoerdani merged 36 commits into

ltmoerdani:mainfrom

Fahad090NP:fix/bug-hunt-round1

Aug 17, 2026

Commits on Aug 15, 2026

fix(usage): define missing esc() HTML-escape helper in webview (Models tab crash)
Fahad090NP
committed
7faabe1
fix(request): guard tool-schema sanitizer against cyclic schemas (stack-overflow crash)
Fahad090NP
committed
171adc9
fix(streaming): cancel cleanly when the user aborts during 5xx backoff
Fahad090NP
committed
42eeb56
fix(agents): revert auto-enabled core settings when autoEnableAgentsWindow is off
Fahad090NP
committed
108c346
fix(usage): count reasoning marker as internal data (no phantom tokens in estimates)
Fahad090NP
committed
c4b5246
fix(utils): correct negative USD sign placement and escape single quotes
Fahad090NP
committed
12a06ac
fix(usage): bucket mid-day events into the correct day (floor instead of round)
Fahad090NP
committed
acae699
fix(usage): clamp subscription percent to [0, 100] (no negative percentages)
Fahad090NP
committed
3a53a86
fix(usage): clear per-profile serverUsage and everTracked state on profile delete
Fahad090NP
committed
76ef45a
fix(streaming): remove dead delta.message reasoning extraction
Fahad090NP
committed
2114186
fix(provider): clamp temperature to the provider-accepted range
Fahad090NP
committed
15643f5
fix(models): sync Go model lists (drop dead ring entry, add minimax-m3)
Fahad090NP
committed
df6206e
fix(usage): raise sqlite3 history read cap so large histories don't silently fail
Fahad090NP
committed
1b355af
fix(streaming): read 5xx body so Router.Unavailable retry branch actually fires
Fahad090NP
committed
eac86c9
fix(autocomplete): don't let a stale token cancel the newer pending debounce
Fahad090NP
committed
84ec592
fix(vision): keep a placeholder on empty vision descriptions; reuse cache in whole-conversation mode
Fahad090NP
committed
2df9c4c
fix(streaming): don't drop reasoning when there is no progress sink
Fahad090NP
committed
da881c9
fix(routing): emit stable synthetic ids for Google tool calls
Fahad090NP
committed
54fb5c4
fix(provider): keep reasoning_content when merging consecutive assistant messages
Fahad090NP
committed
05f5711
fix(usage): clamp progress bar to [0,100]
Fahad090NP
committed
e147107
fix(streaming): send Accept header on chat POST requests
Fahad090NP
committed
27f368c
fix(request): preserve top-level tool-schema enums and common keywords
Fahad090NP
committed
f7a6cb4
fix(streaming): don't emit duplicate text when both message and choice carry it
Fahad090NP
committed
a1782c0
fix(thinking): expose plain 'on' option for toggle+effort models
Fahad090NP
committed
f39ffb0
fix(provider): validate thinking setting values and cap timeout settings
Fahad090NP
committed
be3743b
fix(streaming): report reasoning marker through the context-window progress wrapper
Fahad090NP
committed
1072290
fix(streaming): add empty-response diagnostics to Anthropic and Google transports
Fahad090NP
committed
1df4a2c
fix(provider): log request completion for the Google transport too
Fahad090NP
committed
b3ad8e1
fix(tooling): lint staged renames and resolve .js specifiers to .ts sources
Fahad090NP
committed
de4bc78
fix(usage): estimate cost for missing bundled models instead of tracking $0
Fahad090NP
committed
22c7659
fix(provider): use a real model family so VS Code family selection/grouping works
Fahad090NP
committed
edfd071
fix(usage): stop auto-resolving the active profile over the user's explicit choice
Fahad090NP
committed
1b169fa
fix(errors): cap absurd rate-limit reset durations in error messages
Fahad090NP
committed
a2e5745
refactor(thinking): derive picker options from the shared validated allowlist
Fahad090NP
committed
4e076b0
fix(diagnostics): one failing vendor can't abort the whole picker report
Fahad090NP
committed
bb1617f

Commits on Aug 17, 2026

docs(usage): point active-profile override comment at the right issue (#63)
Fahad090NP
committed
eff395a