0.5.0 RELEASED (SUBAGENTS FINALLY)!!!

[acoliver]

Release v0.5.0

The main feature of this release is support for subagents. Try them by creating one with /subagent and telling the agent to use it.

I can't thank our contributors enough, especially @e2720pjk, who has been the hero of this release with numerous PRs for both stability and features!! Shout out to @renecannao who added an edit tool call that outperforms anything we had before. And special thanks to @MinWooPark-dotcom who fixed something that really annoyed me (not being able to see the whole command when in yolo mode).

0.5.0 Release Video

Installation

Install or upgrade LLxprt Code using npm:

npm install -g @vybestack/llxprt-code

Or use directly with npx:

npx @vybestack/llxprt-code

What's New

• Subagents - create and use subagents with different profiles (model/provider configs)
• Provider Aliases - Supports providers like Synthetic.new out of the box without configuration, create your own custom ones.
• Fuzz Match and Range edit tool call - improved edits from reasoning and non-reasoning models.
• Added configurable history retention limits to control command history storage (memory leak prevention)
• TUI links for OAuth
• Fixed keyboard input hangs when waiting for keyboard input
• Added support for function keys in Kitty terminal protocol
• Fixed Shift+Tab keybinding when not in Kitty mode
• Added Ctrl+R shortcut to show currently running command
• Fixed IME interference with Ctrl shortcuts
• Enhanced streaming error handling with compression retry
• Fixed AbortSignal handling and improved error resilience in pipeline
• Fixed tool output limits on stored responses
• Fixed profile keyfile authentication leak
• Added three new line-based file editing tools
• Added clickable OAuth links and clipboard support
• Fixed /set schema completions
• Improved escape sequence handling in fuzzy replacement
• Fixed cross-platform ripgrep path resolution
• Enhanced subagent error messages with actionable guidance
• Implemented improved tool call processing with pipeline mode
• Fixed OpenAI tool payload serialization
• Fixed CLI authentication arguments to work with --provider flag
• Fixed context limit override honoring in profiles
• Improved client disposal and memory management
• Fixed orphaned tool responses after compression
• Fixed ReDoS vulnerability in number parsing regex
• Added multi-model/multi-profile subagent support
• Preserved user memory context when switching profiles
• Fixed OAuth runtime cache handling for Qwen provider
• Enhanced complexity analyzer task parsing
• Added OAuth safety by default with includeOAuth

Thanks

Huge thanks to the following contributors for their pull requests in this release:

• @MinWooPark-dotcom
• @e2720pjk
• @renecannao

All Changes

• chore(release): v0.5.0 (no version changes) (669264b)
• fix: handle release when version is already applied and format YAML to CI standards (805a2ff)
• Merge pull request Fix Pipeline ToolCall Issues #518 #556 #593 from e2720pjk/issue518 (81797ce)
• fix: consolidate CodeRabbit review fixes and documentation improvements (8497ce8)
• Merge pull request Issue565: Ctrl+C IME interference handling and ASCII normalization #592 from e2720pjk/issue565 (70d9b1e)
• feat: enhance streaming error handling with compression retry (06b85ba)
• fix: enhance AbortSignal handling and error resilience in pipeline (f475b10)
• docs: correct pipeline documentation to reflect actual 75% implementation status (c0eeaf3)
• feat: complete Pipeline ToolCall Fixes implementation (ec76714)
• docs: update pipeline toolcall fixes implementation status (1ab0fa4)
• Merge pull request Fix tool output limiter handling (#589) #591 from vybestack/issue589 (25b9065)
• Merge pull request feat: add ctrl+r to display running shell command #577 from MinWooPark-dotcom/feat/show-full-command (a9ad588)
• refactor: apply CodeRabbit review feedback for KeypressContext (71745c5)
• fix: enforce tool output limits on stored responses (Tool output limiter should trim stored tool_response data #589) (6d1b42b)
• Merge branch 'main' into feat/show-full-command (e6c0737)
• fix: normalize uppercase ASCII codes in CSI-u sequences (0e0eb9a)
• fix: add IME interference handling for Ctrl shortcuts (b776c8c)
• Delete .github/workflows/pr-review.yml (1362242)
• Merge pull request Kitty protocol cherry-pick fixes and compatibility updates #590 from e2720pjk/20251116-kitty-cherry-pick-fixes (8d0bbf7)
• feat: complete Kitty protocol implementation with comprehensive fixes (525aac4)
• Merge pull request Fix /set schema completions (#587) #588 from vybestack/issue587 (4a5967b)
• Merge branch 'main' into feat/show-full-command (3e6ccd4)
• Fix /set schema completions (The configurations are nested repeatedly. #587) (c273970)
• Merge pull request Limit CLI history retention via settings #585 #586 from vybestack/issue585 fixes CLI history retains streamed subagent output until OOM #585 (6a5d39b)
• fix(core): add validation guard to skip instrumentation for complex syntax (1995f64)
• refactor(core): export stripShellMarkers utility (a01f2fc)
• refactor(cli): use shared stripShellMarkers; document heuristic fallback (ab96e56)
• refactor(core): use shared stripShellMarkers; document parsing limitations (1e82743)
• refactor(core): extract shell marker stripping to shared utility (6760d79)
• Add configurable history retention limits CLI history retains streamed subagent output until OOM #585 (35c58d3)
• Merge pull request Fix tool message sequence validation to prevent API errors #583 from e2720pjk/fix/tool-message-sequence-validation (944ccc6)
• Fix multi-tool sequence validation in OpenAIProvider (d30f2d1)
• Fix tool message sequence validation to prevent API errors (10da7ba)
• Merge pull request Fix profile keyfile auth leak #581 from vybestack/issue579 (e36fbaf)
• Fix profile keyfile auth leak (Profile save stores auth key even when keyfile is used #579) (0c893c5)
• Merge pull request Fix alias provider auth + subagent profile loading #578 from vybestack/issue572_573_574 (d586cde)
• Clean up auth lookups and model fallback after review (82e0cca)
• Fix profile load GCP env lookup after upstream merge (remove timestamp from default prompt (give maybe this was started at...) #572 status page shows text that doesn't obey theme #573 loading a profile prints the default model for the provider #574) (74a90b6)
• Merge branch 'main' into issue572_573_574 (c304961)
• Include runtime + prompt updates for alias-aware profiles (remove timestamp from default prompt (give maybe this was started at...) #572 status page shows text that doesn't obey theme #573 loading a profile prints the default model for the provider #574) (2663634)
• Fix Synthetic alias auth + subagent profile loading (remove timestamp from default prompt (give maybe this was started at...) #572 status page shows text that doesn't obey theme #573 loading a profile prints the default model for the provider #574) (15deb94)
• chore(cli): minor type alignment for running-command UI (eb8e160)
• feat(cli): add ctrl+r to show running command (f6307c6)
• feat(core): add POSIX markers and strip before summarize (ec48387)
• Merge pull request Fix Gemini subagent auto prompt (#568 #570) #575 from vybestack/issue568_570 (0bd0142)
• Fix shift+tab keybinding when not in kitty mode (#12552) (428f3d5)
• Add kitty support for function keys. (#12415) (e71e6b4)
• fix(ux) keyboard input hangs while waiting for keyboard input. (#10121) (5b9662a)
• Fix Gemini subagent auto prompt + profile env handling (/subagent doesn't work if gemini is the active provider #568 empty qwen profile doesn't work in subagents. #570) (b1a3fc2)
• Merge pull request Fix provider alias resolution in bundle environments (--provider xAI, --provider Fireworks) #571 from e2720pjk/fix/provider-aliases (a0f8630)
• refactor: simplify alias directory resolution logic (632d65a)
• fix: handle different directory structures for provider aliases in bundle vs development (4968bef)
• Merge pull request Fix js-yaml dependabot alert #567 from vybestack/security/linter-formatter (e887a8d)
• fix: enforce patched js-yaml and prompt parsing (8039b78)
• Merge pull request Complete Pipeline simplification #566 from e2720pjk/issue518 autonomos operation is broken in the recent nightly #518 (cf6b109)
• Provide GH_TOKEN to doc-only filter (8439c9e)
• Merge branch 'main' into issue518 (426c318)
• Treat missing PR file list as non-doc change (02b620a)
• Merge branch 'main' into issue518 (c7e6605)
• Let PR review finish without failing CI (a2e9107)
• Merge branch 'main' into issue518 (c3c21f0)
• Make PR review CI poll non-blocking (32c2668)
• test: add comprehensive integration tests for ToolCallPipeline (8a7af94)
• refactor(openai): Simplify ToolCallPipeline to focus on normalization (fa08ea7)
• coderabbit badge (bc5f571)
• Merge pull request Luther: Luther: llxprt exits with a 0 in non-interactive mode when talking to an api and failing #564 from vybestack/luther-issue560 (89268a0)
• Luther: address Luther: llxprt exits with a 0 in non-interactive mode when talking to an api and failing #560 (3c9e833)
• Merge pull request Phase 1 & 2: ToolCall Pipeline Fixes for #518 #556 #562 from e2720pjk/issue518 (5a75514)
• Fetch remediation PR branch correctly (5cde5bb)
• Fix LLxprt CLI package name (8225975)
• Merge branch 'main' into issue518 (b8baf26)
• Use nightly llxprt CLI in Luther (0b7fcb6)
• Include issue/PR comments in Luther prompt (13a0606)
• Resolve actionlint shellcheck findings (8a0e712)
• Fix actionlint warnings in pr-review workflow (fa46727)
• Fix actionlint findings in CI reviewer and e2e workflow (cdb7842)
• Slow PR reviewer CI polling to 5-minute intervals (533ede6)

• fix(openai): Correct tool call pipeline fragment accumulation and validation (d84e18b6)
  

• Force doc + test awareness in LLxprt review prompt (a844a5a)
• Bias LLxprt reviewer toward Needs Work when tests unclear (a4f9ff0)
• Detect Ready/Needs Work text in verdict parser (fdb69b5)
• Use nightly LLxprt CLI for PR reviews (80da120)
• Install dependencies before running LLxprt review (c7f4855)
• Use GET when querying CI runs (99cc7d6)
• Allow PR reviewer to query workflow runs (8eb11a5)
• Stabilize CI run lookup in PR reviewer (71ff865)
• Fix PR review workspace init order (4218868)
• Add LLxprt PR review automation and CI doc-only skips (189364b)
• chore: keep Luther from touching package-lock (85fd8a7)
• Merge pull request Fix: default to legacy tool pipeline #557 from vybestack/issue556 (9f39f50)
• fix: default to legacy tool pipeline (9b88a49)
• chore: harden e2e defaults and Luther commits (fc4f828)
• chore: uniquify luther branches (4977cc8)
• Merge pull request Luther: revise disabled tools page https://github.com/vybestack/llxprt-code/blob/main/docs/disabled-tools.md #550 from vybestack/luther-issue463 (9a070b5)
• chore: restore package-lock (d6e8237)
• Merge remote-tracking branch 'origin/main' into luther-issue463 (3945d64)
• fix: sanitize peer flags in lockfile (2565876)
• Delete package-lock.json (ed35a52)
• Luther: address revise disabled tools page https://github.com/vybestack/llxprt-code/blob/main/docs/disabled-tools.md #463 (158c201)
• fix: repair lockfile syntax (91bc79b)
• fix: remove stray peer flags (cea3027)
• Luther: address delete https://github.com/vybestack/llxprt-code/tree/main/docs/auth #468 (Luther: delete https://github.com/vybestack/llxprt-code/tree/main/docs/auth #548) (0e26e8c)
• Delete luther directory (f7a40c5)
• Luther: update https://github.com/vybestack/llxprt-code/blob/main/docs/sandbox.md (Luther: update https://github.com/vybestack/llxprt-code/blob/main/docs/sandbox.md #546) (110b549)
• Dont commit Luther logs (cd8a6bc)
• Stabilize Luther CI tests (Stabilize Luther CI tests #544) (67ca51b)
• Run CI test suite in Luther workflow Tests are failing for luther even though it didn't edit code #542 (Ensure Luther hits CI test suite #543) (b9bf318)
• Add --yolo to Luther agent run Luther needs --yolo #539 (Enable yolo mode for Luther #540) (560d3f4)
• Luther workflow refinements (Luther workflow refinements #538) (7eaef8c)
• Use LLXPRT_DEBUG for Luther logging LLxprt Luther Autonomous Agent Overall Plan #514 (Use LLXPRT_DEBUG for Luther logging #536) (6933628)
• Ensure Luther clears OK label (Ensure Luther clears OK label #535) (664bb7b)
• Add Luther workflow (Add Luther workflow #534) (7e00772)
• fix: rip out Gemini loop detector (Nightly build re-enables Gemini loop detector #530) (fix: rip out Gemini loop detector (#530) #531) (39d4e43)
• fix: honor profile context-limit overrides (investigate this error ✕ [API Error: Request would exceed the 121000 token context window even after compression (projected 127891 tokens including system prompt and a 65536 token completion budget). Reduce earlier history or lower maxOutputTokens.] #522) (fix: honor profile context-limit overrides (#522) #524) (281144e)
• Fix idle memory churn and dispose Gemini client (After llxprt is left idle, OOM (Out of Memory) reclamation will occur. #517) (Reduce idle memory churn and dispose Gemini client (#517) #523) (52ba91e)
• Fix server tools base URL leak (Web search and web fetch tools broken - API route not found errors #513) (Fix server tools base URL leak (#513) #515) (b436736)
• Fix orphaned tool responses after compression (Fix orphaned tool responses after compression #511) (dac4fbe)
• fix(core): Resolve ReDoS vulnerability in number parsing regex (fix(core): Resolve ReDoS vulnerability in number parsing regex #512) (8ff41f6)
• Fix /clear to reset foreground agent context /clear doesn’t clear conversation history #509 (Ensure /clear fully resets foreground agent context (#509) #510) (c011640)
• Fix OpenAI tool payload serialization (OpenAI request builder emits orphaned tool outputs after tool errors #501) (Fix OpenAI tool payload serialization (#501) #507) (a3f0b62)
• Issue305：Implement improved tool call processing with pipeline mode and fix undefined tool names (Issue305：Implement improved tool call processing with pipeline mode and fix undefined tool names #505) (d110df6)
• Fix HistoryService token double counting (Token counting appears inflated for OAuth/authenticated Anthropic models (Claude Sonnet 4.5, Opus 4.1) #499) (Fix HistoryService token inflation (#499) #506) (758b391)
• Fix Bundle version ignores provider CLI arguments (--provider, --key, --baseurl, --model) #492: CLI auth args (--key/--keyfile/--baseurl) now work with --provider flag (Fix #492: CLI auth args (--key/--keyfile/--baseurl) now work with --provider flag #494) (557a0fe)
• feat: implement robust cross-platform ripgrep path resolution (fixes ripgrep binary missing at expected path in installed version #483) (feat: implement robust cross-platform ripgrep path resolution (fixes #483) #498) (404f6e0)
• fix: enhance SubagentOrchestrator error messages with actionable guidance (fixes Improve subagent error messages to suggest using list_subagents tool #487) (fix: enhance SubagentOrchestrator error messages with actionable guidance #496) (313455a)
• fix: update subagent tool name to self_emitvalue for CLI tool generation (Subagent tool naming validation error when using Anthropic provider with OAuth #482) (fix: Update subagent tool name to self_emitvalue for CLI tool generation (#482) #495) (5732d19)
• fix: add missing prompt files for range editor tools (range editor missing prompts #477) (fix: add missing prompt files for range editor tools (#477) #493) (d1d6d48)
• feat: add clickable OAuth links and clipboard support (Clickable OAuth links and clipboard fallback for Gemini/Anthropic/Qwen #481) (feat: add clickable OAuth links and clipboard support (#481) #484) (1a9e5d2)
• fix: restore --profile-load functionality and OAuth URL display (Regression: --profile-load flag broken in binary (works in scripts/start.js) #479, Claude auth regression #478) (fix: restore --profile-load functionality and OAuth URL display (#479, #478) #480) (83c0fb6)
• feat: enforce hidden todo reminders (Bug: Todo reminders leak into chat history and spam the user #475) (feat: enforce hidden todo reminders (#475) #476) (2c6eee1)
• fix: improve escape sequence handling in fuzzy replacement (Tab characters escaped to literal \t in edit tool #473) (fix: improve escape sequence handling in fuzzy replacement (#473) #474) (5a92587)
• feat: add three new line-based file editing tools (feat: add three new line-based file editing tools #446) (c2b44a7)
• Fix OAuth timing: Make includeOAuth safe by default (Fix OAuth timing: Make includeOAuth safe by default #472) (95fc225)
• Honor profile ephemerals in CLI bootstrap (Provider overrides lost after switching (#453 regression) #458) (Honor profile ephemerals in CLI bootstrap (#458) #461) (e148f3d)
• fix badges to one line (f05f0fb)
• docs: update provider messaging and replace screenshot (docs: update provider messaging and replace screenshot #460) (4b3853c)
• Preserve CLI/profile base URL and model overrides after provider switch (Preserve overrides when switching providers (#458) #459) (b1b6362)
• Fix provider switching to clear ephemerals for switching providers / profiles not clearing key/keyfile/profile and various ephemerals #453 (Fix provider switching ephemerals (#453) #457) (319bfef)
• fix(gemini): preserve userMemory context when switching profiles ([Bug] Context memory is lost when switching profiles #409) (fix(gemini): preserve userMemory context when switching profiles (#409) #454) (b4c9296)
• Fix OAuth runtime cache handling for Qwen (Qwen Auth broken #436) (Fix OAuth logout regression for Qwen #451) (27a7080)
• chore(ci): log workflow dispatch PR resolution (8a02946)
• fix(ci): allow dispatch runs without pr input (f44786c)
• fix(ci): guard workflow dispatch PR input (ce72881)
• chore: update edit-fuzzy test copyright (7bd38b0)
• feat(ci): allow e2e dispatch to target PR (feat(ci): allow e2e dispatch to target PR #448) (9aa0442)
• Verify PRs Migrate stream validation error logic from gemini-cli #434 and Fix: E2E fail in simple-mcp-server.test.ts #440 are present and working (Test: Revert to e5d98ae4 to diagnose test failures #442 investigation) (Verify PRs #434 and #440 are present and tests pass #447) (d1fa6e2)
• feat(edit): Add fuzzy matching to edit tool (feat(edit): Add fuzzy matching to edit tool #439) (e8177e8)
• refactor(tools): centralize MCP tool name generation (refactor(tools): centralize MCP tool name generation #444) (272ea29)
• Test: Revert to e5d98ae to diagnose test failures (Test: Revert to e5d98ae4 to diagnose test failures #442) (cfbbdf7)
• test: update expected tool call name with server prefix in MCP integration test (Fix: E2E fail in simple-mcp-server.test.ts #440) (73e0b79)
• Migrate stream validation error logic from gemini-cli (Migrate stream validation error logic from gemini-cli #434) (3ce7cbe)
• Guard e2e workflow for fork PRs (Guard e2e workflow for fork PRs #435) (e5d98ae)
• Revise README for LLxprt Code to include subagent de-emphasize gemini-cli fork (44e635d)
• 202511021650security address codeql (202511021650security address codeql #432) (26bce31)
• ci: drop windows build tool prerequisites (removepreqswindows #433) (48f4649)
• Fix complexity analyzer regex backtracking (Fix complexity analyzer regex backtracking #430) (c825fa5)
• test: stabilize GEMINI_CLI env assertion (test: stabilize GEMINI_CLI env assertion #425) (25584d0)
• fix: normalize complexity analyzer task parsing (Fix CodeQL regex warning in complexity analyzer #428) (0b84ac6)
• Delete .completed/PLAN-20251029-STATELESS7 directory (1db0dc8)
• Delete 20251031subagentic2 directory (8901bf1)
• fix: harden complexity analyzer task parsing (Fix CodeQL regex warnings #427) (b43b6be)
• fix: expand profile auth keyfile for subagents (Fix profile auth keyfile expansion (#424) #426) (e5b51aa)
• Fix subagent auto prompt and diagnostics info (/subagent auto not working #419-arbitrary 60k token window being enforced and misprojected #422) (Fix subagent auto prompt and diagnostics info (#419-#422) #423) (6bb3db7)
• Harden tool parsing and sandbox execution (Harden tool parsing and sandbox execution #418) (4e10f9e)
• test: bootstrap runtime in token tracking integration suites (31ae8fd)
• test: ensure runtime config stub matches BaseProvider guard (f115237)
• fix: bootstrap provider runtime in integration tests (Fix integration runtime bootstrap #417) (61b21fe)
• Add multi-model / multi-profile subagents. This inital version is in-process synchronous subagents (Add multi-model / multi-profile subagents. This inital version is in-process synchronous subagents #416) (c9895f4)
• fix: Temporarily apply incremental-rendering patch to reduce flicker (fix: Temporarily apply incremental-rendering patch to reduce flicker #412) (bdc5d29)

---

Full Changelog: v0.4.8...v0.5.0

[minmax-algo]

@acoliver Thanks for the shout-out!
It was my very first open-source contribution, so seeing my name in the release notes really made my day.
I'm just glad I could fix that annoyance for everyone (it was bugging me too!).

Huge congrats on the v0.5.0 release!