Skip to content

Release v0.9.8

Latest
Latest

Choose a tag to compare

View all tags
@github-actions github-actions released this 22 Sep 01:28
· 31 commits to main since this release
v0.9.8
55f07f8
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Hey there! We've been busy making Autohand better. Here's what's new since v0.9.7:

Updates

  • Merge pull request #612 from autohandai/ahtraces
  • Finish Ink teardown before writing session summaries
  • Build the Autohand traces sub agent from its private component
  • Implement native Hermes trace discovery and exact model usage
  • Normalize native OpenClaw session traces
  • Normalize native Amp thread traces
  • Normalize native Antigravity transcript traces
  • Normalize native Copilot sessions and usage
  • Stabilize usage dashboard teardown on Linux
  • Normalize native Grok session traces and usage
  • Keep trace defaults implicit in terminal fixtures
  • Normalize native Droid sessions and settings
  • Bundle trace controls and consent guidance in installers
  • Clarify trace consent and cloud deletion boundaries
  • Normalize native Kimi wire traces and usage
  • Require versioned consent for trace monitoring
  • Normalize native Pi sessions for trace usage
  • Join Cline session records and preserve model-level usage
  • Constrain trace discovery to native session stores
  • Show Free monthly quota and stop ambiguous Autohand AI timeout replays
  • Document source-scoping gate for agent trace rollout
  • Harden native trace parsing and token accounting
  • Restore queue-first input and selected-message steering
  • Cover additional native trace stores and stabilize skill suggestions
  • Align local native compilation with release packaging
  • Add local-first agent trace monitoring and discovery work maps
  • Stabilize usage dashboard Tuistory terminal geometry
  • Make loop grouping tests independent of git timing and drop any-casts
  • Report complete streamed tool calls as tool_calls and unify finish-reason mapping
  • Bound every ReAct loop recovery and report unfinished todos honestly
  • Keep usage heatmap month labels aligned with activity cells
  • Merge the rebuilt resume session picker
  • Normalize incomplete responses across remaining providers
  • Refactoring our reactloop and reflection to speed up the kv_Cache reads a bit more faster as we prepare for blueprint arrival
  • Let the hooks terminal assertions tolerate padded row numbers
  • Fixing a cosmetic slash menu to show below status line
  • Pin the picker's row reservation and make its clamp exact
  • Close out the resume picker review: sentinel timestamps, scoped search, and layering
  • Stop double-compensating the session picker's row-number width
  • Fix session picker row overflow and prove the resume picker in a real terminal
  • Terminate title and tab-colour escapes with ST instead of BEL
  • Make the resume picker's title fallback lazy and close review gaps
  • Adjusting terminal title to be fun
  • Wire /resume to the grouped session picker rows
  • Order resume picker pages by last activity instead of creation time
  • Adding more rich details for ghostty and iterm2 for status bar
  • Thread filterable through showModal, fix stale filter cursor on backspace, and unify Escape detection in the select modal
  • Add opt-in type-to-search filtering to the shared select modal
  • Remove redundant single-digit padding from the hooks event browser label
  • Right-align modal row numbers to a shared width
  • Group session picker rows by local calendar day and drop reveal-row padding
  • Implement session picker row builder with width-aligned columns
  • Adding details for modify install to say hello prompt
  • Clear the task panel whenever a turn ends, not only when it succeeds
  • Offer to start Autohand with a first message once the installer finishes
  • Keep every debug line behind AUTOHAND_DEBUG
  • Keep a failed turn's summary honest after slash commands and shard CI across runners
  • Document fetch body release and synchronous exit handlers in the resource lifecycle rules
  • Release unread fetch bodies, hung probes, and exit-time resources
  • Shut down relay, PTY, streaming, and ACP resources that outlived their session
  • Release per-turn resources that were retained for the life of the session
  • Expose the effective subagent registry to SDK clients
  • Honor process provider selection without persisting credentials
  • Fixing a few test and ui runtime errors
  • Let the stdin steer test run with the default Enter behaviour
  • Repair the four CI failures left by the branch integration
  • Keep the Ink composer wiring tests green after Enter started steering by default
  • Delete the superseded ContextManager and the unreferenced barrel modules
  • Remove dead exports and consolidate duplicated helpers across the CLI
  • Keep goal panels live across terminal processes
  • Recover offline goals through their original conversations
  • Persist goal stop reasons and resumable progress checkpoints
  • Require reported completion evidence for approved goal criteria
  • Preserve damaged goal snapshots and offer explicit backup recovery
  • Persist goal completion before resolving the next queued template
  • Allow fresh goals after budget exhaustion without losing history
  • Unify goal activation across commands and agent tools
  • Attribute turn usage to its original goal and session
  • Preserve goal mutations across concurrent sessions
  • Record monitored proof failures and unchanged upstream evidence
  • Record evidence and disposition for the September issue investigation
  • Bound command output and preserve readable memory under storage pressure (#556, #546)
  • Preserve provider rejection semantics and stop retrying retired models (#544, #545, #558)
  • Keep model settings owned by the selected provider
  • Reject unsafe Windows startup directories before filesystem work (#564, #563, #548, #547)
  • Classify completed replies without treating quoted examples as promises (#565, #562, #557)
  • Preserve native tool exchanges when cropping context (#567, #555, #550)
  • Prevent sync completion from redrawing a closed terminal screen
  • Enable authenticated local peer messaging and coordinated command admission
  • Never fail a handoff because the Web conversation moved on
  • Return an imported session to its own Web conversation
  • Wait for confirmed terminal shutdown in interactive tests
  • Honor provider overrides when resuming native transfers
  • Making ui a little less fresh refresh
  • Route the last three hand-rolled usage objects through the shared normalizer
  • Report what a context compaction cost and which skill spans survived it
  • Preserve cache token usage across turns and session telemetry
  • Report every top-level CLI command, and what each tool result cost
  • Document Moa timeout recovery and validation limits
  • Answer a bare /handoff with the surfaces it can hand this session to
  • Give the idle tip rotation test room for the slower cadence
  • Report goal transitions and skill spans from the CLI
  • Adding more ignore files
  • Rotate composer tips while idle and keep every turn's summary on that row
  • Add run budgets for model requests, tokens, and wall time shared with sub-agents
  • Validate command-mode answers against a caller-supplied JSON Schema with --output-schema
  • Add named config profiles and --set overrides that apply to one run only
  • Add --ephemeral runs that keep the session in memory and out of history
  • Back off session recovery after a provider outage instead of re-sending the turn within seconds
  • Finish the cloud streaming port: resolve the committed test conflict and document the inference endpoint
  • Introducing autohand doctor for inspecting flaky config files and restore broken json syntax, adding more control for sub agents with access to skills via autohand skill registry and more built-in sub agents and control for configure references too
  • Add lifecycle hook authoring, :alias messaging, steering, session titles, built-in analysis skills, and startup budgets
  • Run session-end hooks for ACP clients and emit the autoresearch finalize events
  • Install ah as a short command alias alongside autohand-code and agent

Full comparison: v0.9.7...v0.9.8

Get it

Quickest way:

curl -fsSL https://autohand.ai/install.sh | sh

Via npm or bun:

npm install -g autohand-cli

Via Homebrew:

brew install autohandai/code/autohand-code

Or grab a binary below for your platform.

Platform Architecture Binary
macOS Apple Silicon autohand-macos-arm64
macOS Intel autohand-macos-x64
Linux x64 autohand-linux-x64
Linux ARM64 autohand-linux-arm64
Windows x64 autohand-windows-x64.exe
Assets 24
Loading