CHANGELOG.md

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[2.7.0] - 2026-04-21

Added

• Meta-Harness autoresearch loop (research/ subsystem). Opus-driven search over a bounded mutation surface. Fixed 5-minute kernel, append-only results log, contrastive proposer with access to source + top/bottom traces. See research/domain_spec.md for the design.
• Admiral actuator: scripts/stress_watcher.py sends SIGUSR1 to the stress harness on unrecoverable TUI wedges; the harness respawns its drydock child at the next iteration boundary. Rate-limited. First loop where admiral acts on its observations.
• Env override surface for experiments: DRYDOCK_AUTO_CONTINUE_DISABLE, DRYDOCK_STRESS_{SESSION_RESET_EVERY,MAX_CONSECUTIVE_SKIPS_BEFORE_RESET,SEND_PROMPT_RETRY_COUNT,SEND_PROMPT_WAIT_PER_RETRY_SEC}, DRYDOCK_ADMIRAL_{REPEAT_WARNING_THRESHOLD,REPEAT_FORCE_STOP_THRESHOLD,EMPTY_RESULT_THRESHOLD,SAME_TOOL_NAME_REPEAT_LIMIT_{BASH,READ}}, DRYDOCK_PROMPTS_DIR. All default to the prior hardcoded values; production behavior unchanged.
• DEPLOYMENT.md: known-working hardware + vLLM Docker flags + ~/.drydock/config.toml template + env var table + diagnostic checklist. Answers the "works on your machine, not mine" question.
• scripts/stress_babysitter.sh: hourly monitor cron that auto-restarts long stress runs from the last successful step. Silent when healthy; Telegrams only on state change.
• Check 0 empty-result loop detector in agent_loop.py. Fires FORCE_STOP at 3 identical tool calls when the result matches an empty pattern (total_count: 0, no todos, no tasks, no matches, blank). Catches todo / task / grep-empty loops that Check 1's threshold-of-8 is too lax for.
• Proactive write-oscillation prune in _sanitize_message_ordering. Any path with ≥3 write_file history entries gets its older copies pruned before the next LLM call, so context stays below Gemma 4's 131K ceiling instead of waiting for a 400.
• Raw-markdown leakage detector in stress harness + admiral. Surfaces TUI rendering regressions as stress-alert/raw-markdown- leakage on the admiral dashboard.
• OVERNIGHT_PROGRESS.md session-continuity doc template.

Changed

• should_break_loop in agent_loop.py:876 — the condition tool_turns == 0 was unreachable since the file's current structure landed. Text-only assistant responses now cleanly close the user turn instead of regenerating until PER_PROMPT_BUDGET_SEC times out. This is a behavior fix users will notice immediately — questions the model asks now wait for input.
• Continue-loop: env-gated via DRYDOCK_AUTO_CONTINUE_DISABLE=1 so stress runs (and any workflow hitting text-only prompt loops) can opt out.
• SESSION_ROOT resolution in scripts/shakedown_interactive.py: now reads the path from ~/.drydock/config.toml's [session_logging].save_dir instead of hardcoding ~/.drydock/logs/session/. Fixes false-SKIP reporting on any machine whose drydock config has the newer default ~/.vibe/logs/session/.
• Todo tool (todo.py): empty-read response now escalates. First empty: "No todos yet. Write one or proceed with the task." Second+: "Still no todos. Stop calling read. Empty state will not change on its own." Closes GitHub #10.
• Fresh-start wipe in stress harness: wipes cwd to fixtures (PRD/AGENTS/functional_tests) instead of accumulating prior runs' half-written artifacts.
• Raw-markdown detector patterns tightened (**bold**, ##heading, [link](url) only — dropped bullet + numbered-list patterns that matched rendered lists and generated 45 % false positives).

Infrastructure

• First clean 1658-prompt stress run completed 2026-04-21. ~1622 successful prompts (run + babysitter auto-restart), 35 SKIPs, 0 TIMEOUTs, ~60 admiral-initiated TUI recycles. ~46-hour wall-clock. Reference baseline for future experiments.
• scripts/auto_release.sh supports DRYDOCK_FORCE_VERSION for MINOR / MAJOR bumps (the ordinary PATCH-bumper would have taken us to 2.6.149, not 2.7.0).

Known follow-ups

• Meta-Harness median-of-3 replication before promotion (currently single-sample).
• TUI Markdown widget regression surfaced by admiral but not yet fixed.

[2.4.2] - 2026-03-12

Added

• Session ID included in telemetry events for better tracing

Changed

• Skills now extract arguments when invoked, improving parameter handling
• Auto-compact threshold falls back to global setting when not defined at model level
• Update notification toast no longer times out, ensuring the user sees the restart prompt
• Removed file_content_before from Vibe Code, reducing payload size

[2.4.1] - 2026-03-10

Added

• HarnessFilesManager for selective loading of harness files, enabling SDK usage without accessing the file system.

Changed

• Web search tool infers server URL from provider config instead of hardcoded production API
• ask_user_questions tool disabled in prompt mode

Fixed

• Space key fix extended to all Input widgets (question prompts, proxy setup) in VS Code terminal
• Ruff isort/formatter config conflict resolved (split-on-trailing-comma set to false)

[2.4.0] - 2026-03-09

Added

• User plan displayed in the CLI banner
• Reasoning effort configuration and thinking blocks adapter

Changed

• Auto-compact threshold is now per-model
• Removed expensive file scan from system prompt; cached git operations for faster agent switching
• Improved plan mode
• Updated whoami response handling with new plan type and name fields

Fixed

• Space key works again in VSCode 1.110+
• Arrow-key history navigation at wrapped-line boundaries in chat input
• UTF-8 encoding enforced when reading metadata files
• Update notifier no longer crashes on unexpected response fields

[2.3.0] - 2026-02-27

Added

• /resume command to choose which session to resume
• Web search and web fetch tools for retrieving and searching web content
• MCP sampling support: MCP servers can request LLM completions via the sampling protocol
• MCP server discovery cache (MCPRegistry): survives agent switches without re-discovering unchanged servers
• Chat mode for ACP (session/set_config_options with mode=chat)
• ACP session/set_config_options support for switching mode and model
• Tool call streaming: tool call arguments are now streamed incrementally in the UI
• Notification indicator in CLI: terminal bell and window title change on action required or completion
• Subagent traces saved in agents/ subfolder of parent session directory
• IDE detection in new_session telemetry
• Discover agents, tools, and skills in subfolders of trusted directories (monorepo support)
• E2E test infrastructure for CLI TUI

Changed

• System prompts rewritten for improved model behavior (3-phase Orient/Plan/Execute workflow, brevity rules)
• Tool call display refactored with ToolCallDisplay/ToolResultDisplay models and per-tool UI customization
• Middleware pipeline replaces observer pattern for system message injections
• Improved permission handling for write_file, read_file, search_replace (allowlist/denylist globs, out-of-cwd detection)
• Proxy setup UI updated with guided bottom-panel wizard
• Smoother color transitions in CLI loader animation
• Dead tool state classes removed (Grep, ReadFile, WriteFile state)

Fixed

• Agent switch (Shift+Tab) no longer freezes the UI (moved to thread worker)
• Empty assistant messages are no longer displayed
• Tool results returned to LLM in correct order matching tool calls
• Auto-scroll suspended when user has scrolled up; resumes at bottom
• Retry and timeout handling in Mistral backend (backoff strategy, configurable timeout)

Removed

[2.2.1] - 2026-02-18

Added

• Multiple clipboard copy strategies: OSC52 first, then pyperclip fallback when system clipboard is available (e.g. local GUI, SSH without OSC52)
• Ctrl+Z to put Vibe in background

Changed

• Improve performance around streaming and scrolling
• File watcher is now opt-out by default; opt-in via config
• Bump Textual version in dependencies
• Inline code styling: yellow bold with transparent background for better readability

Fixed

• Banner: sync skills count after initial app mount (fixes wrong count in some cases)
• Collapsed tool results: strip newlines in truncation to remove extra blank line
• Context token widget: preserve stats listeners across /clear so token percentage updates correctly
• Vertex AI: cache credentials to avoid blocking the event loop on every LLM request
• Bash tool: remove NO_COLOR from subprocess env to fix snapshot tests and colored output

[2.2.0] - 2026-02-17

Added

• Google Vertex AI support
• Telemetry: user interaction and tool usage events sent to datalake (configurable via enable_telemetry)
• Skill discovery from .agents/skills/ (Agent Skills standard) in addition to .vibe/skills/
• ACP: session/load and session/list for loading and listing sessions
• New model behavior prompts (CLI and explore)
• Proxy Wizard (PoC) for CLI and for ACP
• Proxy setup documentation
• Documentation for JetBrains ACP registry

Changed

• Trusted folders: presence of .agents is now considered trustable content
• Logging handling updated
• Pin cryptography to >=44.0.0,<=46.0.3; uv sync for cryptography

Fixed

• Auto scroll when switching to input
• MCP stdio: redirect stderr to logger to avoid unwanted console output
• Align pyproject.toml minimum versions with uv.lock for pip installs
• Middleware injection: use standalone user messages instead of mutating flushed messages
• Revert cryptography 46.0.5 bump for compatibility
• Pin banner version in UI snapshot tests for stability

[2.1.0] - 2026-02-11

Added

• Incremental load of long sessions: windowing (20 messages), "Load more" to fetch older messages, scroll to bottom when resuming
• ACP support for thinking (agent-client-protocol 0.8.0)
• Support for FIFO path for env file

Changed

• UI redesign: new look and layout for the CLI
• Textual UI optimizations: ChatScroll to reduce style recalculations, VerticalGroup for messages, stream layout for streaming blocks, cached DOM queries
• Bumped agent-client-protocol to 0.8.0
• Use UTC date for timestamps
• Clipboard behavior improvements
• Docs updated for GitHub discussions
• Made the Upgrade to Pro banner less prominent

Fixed

• Fixed inaccurate token count in UI in some cases
• Fixed agent prompt overrides being ignored
• Terminal setup: avoid overwriting Wezterm config

Removed

• Legacy terminal theme module and agent indicator widget
• Standalone onboarding theme selection screen (replaced by redesign)

[2.0.2] - 2026-01-30

Added

• Allow environment variables to be overridden by dotenv files
• Display custom rate limit messages depending on plan type

Changed

• Made plan offer message more discreet in UI
• Speed up latest session scan and harden validation
• Updated pytest-xdist configuration to schedule single test chunks

Fixed

• Prevent duplicate messages in persisted sessions
• Fix ACP bash tool to pass full command string for chained commands
• Fix global agent prompt not being loaded correctly
• Do not propose to "resume" when there is nothing to resume

[2.0.1] - 2026-01-28

Fixed

• Fix encoding issues in Windows

[2.0.0] - 2026-01-27

Added

• Subagent support
• AskUserQuestion tool for interactive user input
• User-defined slash commands through skills
• What's new message display on version update
• Auto-update feature
• Environment variables and timeout support for MCP servers
• Editor shortcut support
• Shift+enter support for VS Code Insiders
• Message ID property for messages
• Client notification of compaction events
• debugpy support for macOS debugging

Changed

• Mode system refactored to Agents
• Standardized managers
• Improved system prompt
• Updated session storage to separate metadata from messages
• Use shell environment to determine shell in bash tool
• Expanded user input handling
• Bumped agent-client-protocol to 0.7.1
• Refactored UI to require AgentLoop at VibeApp construction
• Updated README with new MCP server config
• Improved readability of the AskUserQuerstion tool output

Fixed

• Use ensure_ascii=False for all JSON dumps
• Delete long-living temporary session files
• Ignore system prompt when saving/loading session messages
• Bash tool timeout handling
• Clipboard: no markup parsing of selected texts
• Canonical imports
• Remove last user message from compaction
• Pause tool timer while awaiting user action

Removed

• instructions.md support
• workdir setting in config file

[1.3.5] - 2026-01-12

Fixed

• bash tool not discovered by vibe-acp

[1.3.4] - 2026-01-07

Fixed

• markup in blinking messages
• safety around Bash and AGENTS.md
• explicit permissions to GitHub Actions workflows
• improve render performance in long sessions

[1.3.3] - 2025-12-26

Fixed

• Fix config desyncing issues

[1.3.2] - 2025-12-24

Added

• User definable reasoning field

Fixed

• Fix rendering issue with spinner

[1.3.1] - 2025-12-24

Fixed

• Fix crash when continuing conversation
• Fix Nix flake to not export python

[1.3.0] - 2025-12-23

Added

• agentskills.io support
• Reasoning support
• Native terminal theme support
• Issue templates for bug reports and feature requests
• Auto update zed extension on release creation

Changed

• Improve ToolUI system with better rendering and organization
• Use pinned actions in CI workflows
• Remove 100k -> 200k tokens config migration

Fixed

• Fix -p mode to auto-approve tool calls
• Fix crash when switching mode
• Fix some cases where clipboard copy didn't work

[1.2.2] - 2025-12-22

Fixed

• Remove dead code
• Fix artefacts automatically attached to the release
• Refactor agent post streaming

[1.2.1] - 2025-12-18

Fixed

• Improve error message when running in home dir
• Do not show trusted folder workflow in home dir

[1.2.0] - 2025-12-18

Added

• Modular mode system
• Trusted folder mechanism for local .vibe directories
• Document public setup for vibe-acp in zed, jetbrains and neovim
• --version flag

Changed

• Improve UI based on feedback
• Remove unnecessary logging and flushing for better performance
• Update textual
• Update nix flake
• Automate binary attachment to GitHub releases

Fixed

• Prevent segmentation fault on exit by shutting down thread pools
• Fix extra spacing with assistant message

[1.1.3] - 2025-12-12

Added

• Add more copy_to_clipboard methods to support all cases
• Add bindings to scroll chat history

Changed

• Relax config to accept extra inputs
• Remove useless stats from assistant events
• Improve scroll actions while streaming
• Do not check for updates more than once a day
• Use PyPI in update notifier

Fixed

• Fix tool permission handling for "allow always" option in ACP
• Fix security issue: prevent command injection in GitHub Action prompt handling
• Fix issues with vLLM

[1.1.2] - 2025-12-11

Changed

• add terminal-auth auth method to ACP agent only if the client supports it
• fix user-agent header when using Mistral backend, using SDK hook

[1.1.1] - 2025-12-10

Changed

• added include_commit_signature in config.toml to disable signing commits

[1.1.0] - 2025-12-10

Fixed

• fixed crash in some rare instances when copy-pasting

Changed

• improved context length from 100k to 200k

[1.0.6] - 2025-12-10

Fixed

• add missing steps in bump_version script
• move pytest-xdist to dev dependencies
• take into account config for bash timeout

Changed

• improve textual performance
• improve README:
  • improve windows installation instructions
  • update default system prompt reference
  • document MCP tool permission configuration

[1.0.5] - 2025-12-10

Fixed

• Fix streaming with OpenAI adapter

[1.0.4] - 2025-12-09

Changed

• Rename agent in distribution/zed/extension.toml to mistral-vibe

Fixed

• Fix icon and description in distribution/zed/extension.toml

Removed

• Remove .envrc file

[1.0.3] - 2025-12-09

Added

• Add LICENCE symlink in distribution/zed for compatibility with zed extension release process

[1.0.2] - 2025-12-09

Fixed

• Fix setup flow for vibe-acp builds

[1.0.1] - 2025-12-09

Fixed

• Fix update notification

[1.0.0] - 2025-12-09

Added

• Initial release