v0.9.11

[0.9.11]

Added

• PSI Tools: 5 new IDE-powered code intelligence tools for agent mode — find_symbols, document_symbols, find_references, find_definition, find_implementations (#921)
• Semantic code navigation: agent can jump to definitions, find usages, and discover implementations using IntelliJ's Program Structure Interface
• PSI tools work across all IDE-supported languages (Java, Kotlin, Python, JS/TS, Go, Rust, etc.)
• New PSI Tools toggle in Agent Settings (enabled by default)

Documentation

• Add PSI Tools (Code Intelligence) section to Agent Mode documentation with tool descriptions, example prompts, and architecture diagrams

v0.9.10

[0.9.10]

Added

• ACP Runners: communicate with external agents (Kimi, Gemini CLI, Kilocode, Claude Code, Copilot) via the Agent Communication Protocol (JSON-RPC 2.0 over stdin/stdout) with structured streaming and capability negotiation (#920)
• Conversation history for CLI and ACP runners: prior exchanges are formatted as a text preamble so external tools can recall earlier messages (#920)
• Per-tool enable/disable for MCP servers and agent mode built-in tools (#919)
• New fetch_page built-in agent tool for web page fetching (#919)
• Claude and Copilot as new ACP tool types with configurable ACP flags (#920)

Changed

• Move CLAUDE.md/AGENTS.md and DEVOXXGENIE.md content injection from per-user-message to system prompt, set once per conversation (#920)
• Dedicated CLI/ACP Runners settings page, extracted from Spec Driven Development settings (#920)

Dependencies

• Bump software.amazon.awssdk:bom (#918)

ACP Runners

Per MCP Tool enable/disable

Embedded Agent tool management

v0.9.9

[0.9.9]

Added

• CLI Runners as a chat-mode provider: execute prompts via external CLI tools (Claude Code, GitHub Copilot, Codex, Gemini CLI, Kimi) directly from the chat interface (#916)
• CLI tool execution for spec tasks: run selected or all To Do tasks through CLI runners (Claude, Copilot, Codex, Gemini, Kimi) (#874, #875)
• Kimi CLI runner with --prompt flag support and --quiet mode for clean output (#875, #916)

Changed

• Refactor CLI runners to dedicated com.devoxx.genie.service.cli package with Command pattern (GoF) (#874, #875)

Fixed

• Defer scrollTo in JCEF WebView to fix auto-scroll after DOM updates (#916)

Dependencies

• Bump the gradle-dependencies group with 3 updates (#873)

v0.9.8

[0.9.8]

Added

• Automated test execution: new run_tests agent tool that auto-detects build systems (Gradle, Maven, npm, Cargo, Go, Make), runs tests, and returns structured results to the LLM (#863)
• Agent iterates on test failures: system prompt instructs the LLM to run tests after code changes and fix until they pass
• Configurable test execution settings: enable/disable toggle, timeout (default 5 min), custom test command with {target} placeholder
• Test output parsing for Gradle, Maven Surefire, and Jest with pass/fail counts and failed test names
• Agent Loop: batch task execution with dependency-aware topological sort, progress tracking, and automatic task advancement (#872)
• Spec task runner service with batch execution UI for running selected or all To Do tasks

Fixed

• Prevent agent and MCP logs from leaking across projects (#867)
• Defer task advancement until prompt execution completes (#872)
• Downgrade Kotlin to 2.0.21 to match IDE runtime and remove dead activeTaskSpec code
• Prompts settings dialog text overflows viewport

Documentation

• Add "Automated Test Execution" section to Agent Mode docs with build system table, configuration, and troubleshooting
• Add Agent Loop batch task execution page for SDD
• Add references, documentation, and CLI board sections to SDD docs
• Reorder docs sidebar: Agent Mode after Features Overview, then Spec-driven Development, then Agent Loop
• Add Agent Loop cross-references to Agent Mode and Spec-driven Development pages

Run spec tasks in batch automatic

RunTasks.mp4

v0.9.7

[0.9.7]

Added

• Spec Driven Development (SDD) feature with Backlog.md integration, Kanban board, and enhanced Spec Browser (#862)
• 17 granular backlog tools replacing manage_spec for full CRUD operations on tasks, documents, and milestones
• Init Backlog button for quick project setup

Changed

• Remove deprecated o1-mini model from OpenAI registry

Dependencies

• Bump the gradle-dependencies group across 1 directory with 18 updates (#861)

Demo

v0.9.6

[0.9.6]

Added

• Inline code completion using Fill-in-the-Middle (FIM) models with configurable provider/model settings (#857)

Changed

• Migrate Gradle IntelliJ Plugin from 1.x to 2.x (#858)

Documentation

• Update contributing guidelines
• Add CLAUDE.md/AGENTS.md feature to v0.9.5 changelog

v0.9.5

[0.9.5]

Added

• Add CLAUDE.md and AGENTS.md support: automatically include project instructions in prompts with priority logic (CLAUDE.md first, then AGENTS.md) (#856, #854)
• Improve agent logging and MCP tool provider integration (#855)
• Replace parallelism spinner with dynamic add/remove sub-agent rows for more flexible agent configuration (#855)

Fixed

• Add null-safety to welcome template rendering with fallback to local content
• Use correct announcement fields (type/message) in welcome.json

Documentation

• Update Agent Mode docs to reflect add/remove sub-agent UI
• Add Agent Mode and Parallel Sub-Agents sections to homepage and README
• Update MCP documentation and fix GitHub issue template links

v0.9.4

[0.9.4]

Added

• Agent mode with built-in tool execution: read_file, write_file, edit_file, list_files, search_files, run_command, parallel_explore (#849)
• Parallel sub-agents for concurrent codebase exploration, each configurable with a different LLM provider/model (#850)
• MCP configuration JSON import/export for easy sharing of MCP server setups (#852, #786)

Fixed

• Clear selection metadata when removing files from context (#847, #783)

Documentation

• Rename Sub-Agents documentation to Agent Mode

More details @ https://genie.devoxx.com/docs/features/agent-mode

Parallel sub-agents in action

Consolidated response

v0.9.2

[0.9.2]

Added

• Add dedicated Skills settings dialog for custom user-defined skills
• Add $ARGUMENT placeholder support in custom skill templates (injects everything typed after /skill-name)

Changed

• Keep Prompts settings focused on system instruction, DEVOXXGENIE.md options, and keyboard shortcuts
• Update welcome screen wording to Skills terminology and highlight MCP Marketplace first

Fixed

• Show resolved custom skill prompt in the user message bubble instead of raw /command
• Migrate legacy HTTP/SSE MCP transport usage to StreamableHttpMcpTransport (removes deprecated-for-removal API usage)
• Issue #809: Fix LM Studio API URL compatibility - changed default from http://localhost:1234/api/v1/ to http://localhost:1234/v1/ to match OpenAI-compatible chat completions endpoint, while preserving rich metadata endpoint access for context length detection
• Issue #809: Fix LM Studio context length detection so model metadata no longer defaults to 8K when larger context values are available
• Issue #809: Add optional LM Studio fallback context setting in the GUI when model metadata does not expose context length
• Replace deprecated Project.getBaseDir() with ProjectUtil.guessProjectDir() across all scanner extensions (#838)
• Replace deprecated NioEventLoopGroup with MultiThreadIoEventLoopGroup + NioIoHandler for Netty 4.2.x compatibility (#839)
• Resolve plugin verifier warnings for experimental and deprecated API usage (#840)
• Add Kotlin K2 plugin mode declaration (supportsKotlinPluginMode) in plugin.xml, required since IntelliJ 2024.2.1 (#840)

v0.9.1 : MCP Marketplace

[0.9.1]

Added

• Add MCP server marketplace for browsing and installing MCP servers (#835)
• Add location and type filters to the MCP marketplace dialog (#835)
• Add support for custom headers in HTTP MCP transports for authentication (#835)

Changed

• Refactor MCP integration to remove the core sub-module and use the official langchain4j-mcp artifact (#835)
• Remove legacy GitHub MCP and FileSystem MCP quick-action buttons from settings in favor of marketplace-driven setup (#835)

Fixed

• Issue #791: Escape triple braces in user content to prevent template variable exception (#834)