Unreleased
Changes since v0.99.1
Automatically listed from commit messages.
- Fix auto-poke canceling pending cloud handoffs
- add repeatable Linux cloud acceptance gate
- isolate repeatable cloud migration e2e runs
- exercise cloud handoffs through real SSH and TUI
- keep acceptance sockets short and preserve failure evidence
- require new prompts in cloud continuation evidence
- reconcile completed turns before client dispatch
- surface reasons for every cloud command failure
- Place inline swarm beside pinned todos on wide terminals
- Show one pinned lane dot per worker todo item
- pass cloud failure context to the next agent turn
- update weekly stars chart
- update downloads badge
- resolve context budget from routed model catalog
- Round inline diff rail corners
- wait for history before sending fork startup prompts
- Keep scrollbar track invisible outside thumb intersections
- Show diff counts once in the inline diff header
- Compact diff header paths without wrapping
- Move compact diff expansion shortcut to the footer
- compact failed background tasks with connected diagnostics
- Ensure info widgets have bottom-border text
- Guard shortcut chrome against inherited terminal colors
- Compact shortcut badges and remove redundant diff hidden counts
- Show tool duration alongside token counts
- Preserve tool execution duration in live and restored rows
- Unify tool failures with normal rows and compact connected reasons
- Replace generic widget footers with compact distinct facts
- Test public widget footer metadata and narrow rendering
- Validate activity footer lifecycle and fix singular client label
- preserve source whitespace in inline diffs
- Improve expanded inline diff gutters and wrapping
- color tool duration independently from token usage
- Test expanded diffs through full terminal rendering
- Preserve warm OpenAI websocket safely after interrupted responses
- Centralize compact tool failure summaries and align error connectors
- Simplify compact failure connector to standalone rounded corner
- Join compact failure connector directly to reason
- Compact successful background tasks and collapse tool image previews
- Exercise expanded image geometry in cache reuse regression
- Repair interleaved renderer and regression-test hunks
- Fix todo add schema rejected by Gemini
- Cover todo schema rejection at Gemini request boundary
- Track Anthropic OAuth token usage and per-account cost estimates
- Refresh local OAuth totals independently of quota cache
- unify subscription usage report identity
- Remove password-reset restriction from system prompt
- Remove blanket password restrictions from browser workflows
- update downloads badge
- retain warm websocket through cancellation recovery
- Disable shared turn-end weak-point digest followups
- Disable TUI turn-end weak-point digest
- Add display-only full todo completion report
- Automatically display full todo completion reports
- Keep full todo completion reports expanded
- Highlight blocked acceptance in todo completion reports
- Render todo completion snapshots with pinned cards and full evidence
- Style completion reports with pinned todo cards and compact details
- Show only recorded turn-end gate findings in todo completion
- add isolated ephemeral machine lifecycle and native attach
- keep Jcode model switches on managed subscription transport
- Test deployed Sonnet route alongside curated subscription models
- accept deployed Sonnet 4.6 model ID
- Show only gate categories above completed todo cards
- Make aggregate todo gate digest an explicit default-off opt-in
- Recognize GPT-6.1 Sol reasoning effort on compatible routes
- Hide delivery-state labels from pinned todo cards
- update weekly stars chart
- Hide per-item confidence labels in todo cards
- Verify incomplete todo gates with aggregate digest disabled
- Retry unchanged incomplete todos up to three times in TUI
- Document three-attempt incomplete todo retry budget
- isolate Bedrock test config (fixes #1952)
- triage account usage in capability ledger (fixes #1950)
- Limit unchanged incomplete todo followups to three attempts
- preserve UTF-8 across Responses SSE chunks (fixes #1944)
- preserve explicitly enabled Bedrock live-test credentials (fixes #1952)
- Avoid redundant final follow-up after delivered response
- run search safety regressions on Windows (refs #1932)
- report Windows working set and private commit (#1932)
- abort adopted work when its owner is cancelled
- remove redundant completion confidence footer
- extract tests below source size budget
- avoid repeating final answers after compaction
- scope Windows search gate to portable regressions (refs #1932)
- scope runtime guards after app setup
- cancel blocking searches with their async owner
- create remote fixtures inside their reactor
- keep search cancellation within safety guardrails (refs #1932)
- pin bounded cancellable agentgrep searches (fixes #1932)
- run search safety regressions on Windows (refs #1932)
- report Windows working set and private commit (#1932)
- abort adopted work when its owner is cancelled
- extract tests below source size budget
- scope Windows search gate to portable regressions (refs #1932)
- cancel blocking searches with their async owner
- keep search cancellation within safety guardrails (refs #1932)
- pin bounded cancellable agentgrep searches (fixes #1932)