v0.3.1

Features

• Adapt MCP integration to the stateless SDK.

Bug Fixes

• Stabilize context compaction.
• Persist paste bodies.
• Keep cleared tool results recoverable.

Performance

• Refine compaction recovery.

v0.3.0

Features

• Add memory recall with frontmatter format.
• Introduce a TUI setup wizard.
• Add auto-dream memory consolidation for improved session management.
• Support OpenAI Responses API mode.
• Add agent trace spans for enhanced observability.
• Implement git worktree isolation for sessions and teammates, including tools to enter and exit worktrees with live session CWD.
• Backfill version information from build details.
• Introduce Agent Client Protocol (ACP) frontend and render write/edit operations as native ACP diffs.

Performance

• Improve static prompt caching.
• Unify interactions behind the gate.
• Optimize teammate subsystem extraction.
• Persist sessions more efficiently through the message commit barrier.
• Synchronize thinking process with model capabilities.
• Optimize passing provider extra parameters.

Refactor

• Split Session state for improved modularity.

v0.2.2

Features

• Add live agent fleet list.
• Introduce goal creation and live status updates.
• Support the /goal command.

Performance

• Optimize session startup.

v0.2.0

What's Changed

• perf: polish /diff UI
• fix: restored history render
• feat: make /diff an interactive overlay file list
• fix: escape glob metacharacters in snapshot excludes
• feat: add /diff preview and /redo to the undo stack
• feat: persist /undo stack across sessions and gc shadow repo
• feat: add /undo via shadow git snapshots
• chore: bump agentcore
• feat: tag LLM trace spans with session id for backend grouping
• feat: persist teammates and lazily resume them on message
• perf: update telemetry
• feat: add OpenTelemetry trace export for LLM calls
• refactor: structured hook decision protocol
• feat: team work-stealing on shared task list
• feat: share leader dynamic block with teammates
• perf: cache teammate system prompt with ephemeral cache_control
• style: polish transcript modal chrome and auto-follow scroll
• feat: replay teammate history for late observers
• feat: live teammate transcript modal via event hub
• feat: pre-create default team and auto-suffix duplicate teammate names
• feat: add Team with long-lived teammate agents

v0.1.3

chore: bump agentcore

v0.1.2

Features

• Update welcome page
• Restyle the tool's UI

Bug Fixes

• Fix TUI resize issues by optimizing scrollback cache and replay

Performance

• Refine retry status UI
• Prune stale models and expand provider coverage

v0.1.1

Features

• Add session memory-backed compaction for conversation history.
• Introduce settings for compact window size and ratio.
• Improve /model command by unifying model persistence.

Bug Fixes

• Fix provider and model format resolution in the /model command.

Performance

• Harden prompt caching mechanisms.
• Add observability for LLM calls.
• Create task directories only when needed (lazily).
• Optimize core tool loading.
• Extract info commands into a modal view.
• Unify the info panel and context window display.
• Improve task storage efficiency.
• Update task reminder mechanisms.
• Simplify TUI startup and reduce unused code.
• Refine internal runtime models and task handling.
• Improve TUI markdown rendering speed and accuracy.

Refactor

• Improve internal subagent tool classification.

v0.1.0

Features

• Add a plugin-first runtime.
• Implement plan mode with state management and approvals.
• Add path-based skill activation.
• Introduce task lifecycle hooks, including rollback.
• Unify background tasks using TaskRuntime.
• Add PostStopValidation hook for tasks.
• Redesign UI color palette and internationalize UI strings.

Bug Fixes

• Fix total token calculation.

Performance

• Improve runtime task reminders.
• Unify error display and provider configuration.
• Update bootstrap process and tighten configuration loading.
• Enhance task management behavior.
• Simplify TUI markdown rendering and refine UI colors.
• Update the context command.
• Optimize skill execution and persist usage ranking.
• Streamline approvals by building on the agentcore permission engine.
• Utilize a dedicated todo tool.

Refactor

• Adopt a new context engine with tool policy and file recovery.
• Use a context manager for compression.
• Unify 'todo' into the 'task' concept.
• Refine skill architecture.
• Harness runtime and add an inject primitive.

v0.0.4

Features

• Add support for approving operations outside the project's root directory.
• Introduce the btw command for additional interaction.
• Enhance skill management by allowing context forking and specifying allowed tools for skills.
• Display prompt suggestions after agent task completion.
• Improve output display with markdown rendering for agent responses and tool output, including a dedicated markdown view.
• Integrate Git context, automatic memory retention, and a new interactive shell mode.
• Show retry attempts and enhance overall UI rendering.

Performance

• Improve command palette user experience.
• Enhance the rendering and user experience of approval listings.
• Update the status bar for clearer information display.
• Improve performance of file finding operations by using glob patterns.

Refactor

• Refactor the approval engine, policies, and rules for a more unified and robust system.

v0.0.3

Features

• Add /tasks command with disk-based output.
• Use new MCP SDK.

Bug Fixes

• Repair message sequence.
• Deduplicate tools when restoring.

Other

• Update README.