[v0.17.0] [Hermes Agent] Is it stable?

# Stability Score Evidence — v0.17.0

> v0.17.0 picked up fresh receipts: tool loading can collapse to cronjob-only, QQBot can fail on gateway startup, Matrix/table formatting still bites, and desktop/file-tool/SimpleX regressions joined the stack. The pager has started using footnotes.



## Stability adjustments

## Affected components

- approval-boundary
- auth
- browser
- compaction
- config-profiles
- dashboard
- desktop-linux
- desktop-macos
- desktop-ui
- desktop-windows
- discord
- email
- feishu
- file-tools
- gateway-service
- imessage
- installer
- matrix
- mcp
- memory
- message-delivery
- model-routing
- opencode
- provider-routing
- qqbot
- runtime
- security
- simplex
- telegram
- tool-loader
- update-system
- voice
- web-tools
- whatsapp
- windows-runtime
- windows-update

## Evidence
| Issue | Area | Type | Severity | Penalty | Reason |
|---|---|---|---:|---:|---|
| NousResearch/hermes-agent#51405 | feishu | runtime | 5 | -5 | v0.17.0 gateway startup can crash on fresh Feishu installs because cron.scheduler_provider is missing. |
| NousResearch/hermes-agent#50734 | approval-boundary | kernel | 5 | -5 | v0.17.0 read_file/tool path can ignore safety directives and exfiltrate full .env credentials to LLM providers. |
| NousResearch/hermes-agent#49578 | approval-boundary | kernel | 5 | -5 | v0.17.0 execute_code can bypass agent file edit restrictions and write outside the allowed edit path. |
| NousResearch/hermes-agent#51447 | auth | runtime | 4 | -4 | v0.17.0 Bitwarden profile-suffixed secrets may not reach platform plugins, silently breaking non-default Telegram bot profiles. |
| NousResearch/hermes-agent#51029 | telegram | runtime | 4 | -4 | v0.17.0 multiplexer env override handling can leak a default profile platform token into a secondary Telegram profile. |
| NousResearch/hermes-agent#50806 | runtime | runtime | 4 | -4 | v0.17.0 CLI can hang after a shell tool starts a detached child process, repeatedly submitting tool results instead of returning control. |
| NousResearch/hermes-agent#50765 | runtime | runtime | 4 | -4 | Hermes ACP session/prompt hangs after conversation turn on Windows v0.17.0 while v0.16.0 works, blocking ACP clients. |
| NousResearch/hermes-agent#50394 | mcp | runtime | 4 | -4 | v0.17.0 MCP bridge can churn after one failing stdio server, leaving healthy servers' tools as unknown at runtime. |
| NousResearch/hermes-agent#49627 | runtime | runtime | 4 | -4 | v0.17.0 setup can fail during agent initialization with init_agent() receiving an unexpected event_callback argument. |
| NousResearch/hermes-agent#49305 | message-delivery | runtime | 4 | -4 | Hermes Desktop/TUI on v0.17.0 can lose a user message and persist no assistant response when an additional profile uses codex_app_server. |
| NousResearch/hermes-agent#51448 | provider-routing | runtime | 3 | -3 | Hermes Desktop v0.17.0 on native Windows can return empty LM Studio streams while the same config works under WSL. |
| NousResearch/hermes-agent#51381 | provider-routing | runtime | 3 | -3 | v0.17.0 can send malformed tools[] schemas without function names to OpenCode/DeepSeek backends, causing provider deserialization failure. |
| NousResearch/hermes-agent#51316 | mcp | tool-specific | 3 | -3 | v0.17.0 MCP single-query snapshots can miss slow-start stdio servers, producing incomplete tool lists until reload. |
| NousResearch/hermes-agent#51290 | runtime | runtime | 3 | -3 | v0.17.0 background skill review can block ACP multi-turn IDE sessions after the first turn. |
| NousResearch/hermes-agent#51030 | telegram | runtime | 3 | -3 | v0.17.0 Telegram multiplexer token-collision detection can miss duplicate tokens in profile adapter config. |
| NousResearch/hermes-agent#50959 | provider-routing | runtime | 3 | -3 | v0.17.0 can send invalid Unicode in tools[].function.description, producing HTTP 400s across providers. |
| NousResearch/hermes-agent#50807 | runtime | runtime | 3 | -3 | v0.17.0 forwards very large shell tool output to providers without bounding the payload. |
| NousResearch/hermes-agent#50794 | desktop-macos | runtime | 3 | -3 | Hermes Desktop v0.17.0 can crash on launch on macOS 26.5 with EXC_BREAKPOINT in fontations_ffi. |
| NousResearch/hermes-agent#50252 | provider-routing | runtime | 3 | -3 | v0.17.0 GitHub Copilot provider can return an empty base URL on every inference despite a valid token and detected models. |
| NousResearch/hermes-agent#50209 | desktop-ui | runtime | 3 | -3 | v0.17.0 Desktop boot can hang around the backend status probe even while the CLI itself is healthy. |
| NousResearch/hermes-agent#50199 | provider-routing | runtime | 3 | -3 | v0.17.0 delegation.base_url can be ignored at runtime, forcing incorrect provider endpoint resolution. |
| NousResearch/hermes-agent#49976 | update-system | runtime | 3 | -3 | v0.17.0 Windows ZIP-fallback update path can drop gateway resume state and make restarts fail on stale gateway_state.json. |
| NousResearch/hermes-agent#49961 | provider-routing | runtime | 3 | -3 | v0.17.0 reports configured tools not being invoked across NVIDIA NIM and OpenRouter conversation flows. |
| NousResearch/hermes-agent#49903 | desktop-ui | runtime | 3 | -3 | v0.17.0 Desktop can throw Composer is not available after upgrade, making the composer input unresponsive. |
| NousResearch/hermes-agent#49590 | auth | runtime | 3 | -3 | v0.17.0 gateway startup can prefer literal op:// secret references over resolved cached values, breaking secret-backed channels. |
| NousResearch/hermes-agent#49490 | opencode | runtime | 3 | -3 | OpenCode Zen provider calls fail after updating to v0.17.0, breaking that provider path. |
| NousResearch/hermes-agent#49445 | web-tools | tool-specific | 3 | -3 | Since v0.17.0 the official Docker image can ship with search.exa/web_search/web_extract unusable. |
| NousResearch/hermes-agent#49283 | approval-boundary | kernel | 3 | -3 | v0.17.0 desktop execute_code consent gate still rejects an explicitly consented code execution request. |
| NousResearch/hermes-agent#47917 | desktop-windows | runtime | 3 | -3 | Desktop builds can continue failing after v0.17.0 updates because electronDist disappears from the build path. |
| NousResearch/hermes-agent#47868 | provider-routing | runtime | 3 | -3 | Strict chat-completions providers on v0.17.0 reject leaked messages[].timestamp metadata with HTTP 400s. |
| NousResearch/hermes-agent#36718 | compaction | kernel | 3 | -3 | After updating to v0.17.0, context compression can repeatedly trigger because last_prompt_tokens stays stale until the next API call. |
| NousResearch/hermes-agent#51278 | provider-routing | runtime | 2 | -2 | v0.17.0 title generation can route the main model name to the fallback provider endpoint. |
| NousResearch/hermes-agent#51263 | email | channel-specific | 2 | -2 | v0.17.0 email STARTTLS setup can fail with OpenSSL WRONG_VERSION_NUMBER on OpenSSL 3.5+ systems. |
| NousResearch/hermes-agent#51155 | runtime | runtime | 2 | -2 | v0.17.0 user personalities can fail to change and persist across sessions. |
| NousResearch/hermes-agent#51099 | memory | runtime | 2 | -2 | v0.17.0 can activate the Honcho memory provider even when honcho-ai is not installed. |
| NousResearch/hermes-agent#50966 | browser | tool-specific | 2 | -2 | Desktop GUI on v0.17.0 routes browser_vision calls to browser_scroll instead of the requested vision tool. |
| NousResearch/hermes-agent#50823 | installer | runtime | 2 | -2 | Desktop build for v0.17.0 hard fails on Windows native non-git checkouts because build-stamp cannot determine a git commit. |
| NousResearch/hermes-agent#50769 | installer | runtime | 2 | -2 | Windows installer for v0.17.0 detects a Python 3.12 fallback but still tries Python 3.11 for venv creation, failing install. |
| NousResearch/hermes-agent#50737 | gateway-service | runtime | 2 | -2 | v0.17.0 desktop profile-scoped OpenRouter setup can leave gateway inference not ready. |
| NousResearch/hermes-agent#50713 | message-delivery | runtime | 2 | -2 | v0.17.0 can lose chat text after reopening sessions across multiple deployments. |
| NousResearch/hermes-agent#50675 | gateway-service | runtime | 2 | -2 | v0.17.0 gateway agent cache can reset /usage API-call and token counters during follow-up turns. |
| NousResearch/hermes-agent#50404 | discord | channel-specific | 2 | -2 | v0.17.0 Discord config is not profile-isolated, allowing channel state to bleed across profiles. |
| NousResearch/hermes-agent#50395 | mcp | tool-specific | 2 | -2 | v0.17.0 hermes mcp test can report a server healthy under a different PATH than the gateway MCP spawn actually uses. |
| NousResearch/hermes-agent#50229 | model-routing | runtime | 2 | -2 | v0.17.0 silent fallback failover can run the agent as the wrong model without notifying the user. |
| NousResearch/hermes-agent#50210 | installer | runtime | 2 | -2 | v0.17.0 Windows bootstrap can produce an unsigned Hermes.exe blocked by Smart App Control. |
| NousResearch/hermes-agent#50205 | feishu | channel-specific | 2 | -2 | v0.17.0 Docker image can miss lark-oapi, causing the Feishu channel to fail at connect time. |
| NousResearch/hermes-agent#50175 | config-profiles | runtime | 2 | -2 | v0.17.0 profile deletion can fail while dashboard/gateway child processes are alive, making profiles reappear until reboot. |
| NousResearch/hermes-agent#49982 | desktop-ui | runtime | 2 | -2 | v0.17.0 Windows Desktop can time out connecting to the backend while the CLI still works. |
| NousResearch/hermes-agent#49870 | dashboard | tool-specific | 2 | -2 | v0.17.0 task Done/ready flow can trigger a false diagnostic and then get blocked by the respawn guard. |
| NousResearch/hermes-agent#49606 | update-system | runtime | 2 | -2 | Windows Desktop v0.17.0 update flow can leave users with a mismatched backend/desktop state after silent GUI update failure. |
| NousResearch/hermes-agent#49561 | whatsapp | channel-specific | 2 | -2 | v0.17.0 regression can prevent installing the WhatsApp bridge in Docker environments. |
| NousResearch/hermes-agent#49529 | installer | tool-specific | 2 | -2 | The v0.17.0 PyPI wheel install has a doctor false-positive and omits optional-skills from the wheel package. |
| NousResearch/hermes-agent#49297 | provider-routing | tool-specific | 2 | -2 | Gemma4 through the Ollama backend still hits near-immediate finish_reason=length truncation on v0.17.0. |
| NousResearch/hermes-agent#49242 | windows-runtime | runtime | 2 | -2 | Windows WhatsApp gateway and updater paths on v0.17.0 can use system Node/npm instead of Hermes-managed tooling. |
| NousResearch/hermes-agent#50775 | telegram | channel-specific | 1 | -1 | Telegram macOS client streaming on v0.17.0 can show ghosting/double-text during edited-message updates. |
| NousResearch/hermes-agent#49955 | dashboard | tool-specific | 1 | -1 | v0.17.0 Kanban inline create form truncates field labels in the dashboard. |
| NousResearch/hermes-agent#49908 | dashboard | tool-specific | 1 | -1 | v0.17.0 drag-and-drop to Done can show a completion prompt and then silently revert the task. |
| NousResearch/hermes-agent#49745 | desktop-ui | tool-specific | 1 | -1 | Desktop v0.17.0 composer undo skips pasted content because paste operations lack a contentEditable undo entry. |
| NousResearch/hermes-agent#49732 | model-routing | tool-specific | 1 | -1 | v0.17.0 dashboard model-switch dialog can fire with no active chat and flip the badge before the user decides. |
| NousResearch/hermes-agent#49600 | desktop-ui | tool-specific | 1 | -1 | v0.17.0 desktop UI can clip the audio icon on Windows. |
| NousResearch/hermes-agent#49536 | telegram | channel-specific | 1 | -1 | Telegram finalize messages on v0.17.0 can overlap text because parse_mode mutation falls back incorrectly. |
| NousResearch/hermes-agent#49294 | gateway-service | tool-specific | 1 | -1 | Windows Desktop restart gateway action on v0.17.0 opens a persistent CMD window instead of restarting the installed headless service. |
| NousResearch/hermes-agent#49253 | imessage | channel-specific | 1 | -1 | Photon iMessage on v0.17.0 corrupts Unicode when Markdown bold formatting is present. |
| NousResearch/hermes-agent#52023 | provider-routing | runtime | 3 | -3 | Hermes Desktop v0.17.0 on Windows returns OpenAI 400 errors for GPT-4o-mini, GPT-4.1 and related models because encrypted content is sent to models that do not support it. |
| NousResearch/hermes-agent#51864 | dashboard | runtime | 3 | -3 | Hermes v0.17.0 self-hosted kanban can route task workspaces and board filters through global active-board state instead of task tenant, mixing tenants. |
| NousResearch/hermes-agent#52013 | desktop-ui | channel-specific | 1 | -1 | Hermes Desktop v0.17.0 on Chinese Windows can render Chinese TUI/desktop text as mojibake, making non-ASCII output unreadable. |
| NousResearch/hermes-agent#51646 | memory | runtime | 4 | -4 | Hermes v0.17.0 gateway memory writes can omit the active column, causing INSERT failures and lost persisted state. |
| NousResearch/hermes-agent#51873 | auth | runtime | 3 | -3 | Hermes Desktop v0.17.0 can loop on 401s and never show the documented sign-in UI when connecting to a remote backend with basic auth. |
| NousResearch/hermes-agent#51871 | approval-boundary | runtime | 4 | -4 | Hermes v0.17.0 legacy chat-completions streaming can omit approval events, so clients see text deltas but no tool-approval boundary. |
| NousResearch/hermes-agent#52753 | desktop-windows | runtime | 5 | -5 | v0.17.0 packaged Windows desktop can fail to launch after update because simple-git is missing from the Electron bundle. |
| NousResearch/hermes-agent#52378 | update-system | runtime | 4 | -4 | v0.17.0 Windows desktop auto-update can loop indefinitely, repeatedly spawning elevated updater processes even when behind=0. |
| NousResearch/hermes-agent#52428 | browser | tool-specific | 3 | -3 | v0.17.0 browser_navigate fails with missing _hermes_read_browser_output, breaking browser tool navigation on Windows. |
| NousResearch/hermes-agent#52496 | model-routing | runtime | 3 | -3 | v0.17.0 dashboard model switcher can rewrite named custom providers to openrouter/bare custom, breaking local proxy routing. |
| NousResearch/hermes-agent#52492 | provider-routing | runtime | 2 | -2 | v0.17.0 Codex model selection can persist an unsupported gpt-5.3-codex custom model and then fail prompts with HTTP 400. |
| NousResearch/hermes-agent#52363 | telegram | channel-specific | 2 | -2 | v0.17.0 Telegram outbound messages can be stored as incoming unread items, making inbox badges grow on the agent's own replies. |
| NousResearch/hermes-agent#52573 | voice | tool-specific | 2 | -2 | v0.17.0 TUI continuous voice mode can fail to reopen the microphone after TTS playback on macOS. |
| NousResearch/hermes-agent#51882 | desktop-ui | runtime | 2 | -2 | v0.17.0 Desktop Chat→Personality setting writes display.personality but not agent.system_prompt, so the persona is silently ignored. |
| NousResearch/hermes-agent#51684 | feishu | channel-specific | 2 | -2 | v0.17.0 Feishu approval-card clicks reject FEISHU_ALLOWED_USERS=* wildcard even though normal inbound messages are allowed. |
| NousResearch/hermes-agent#52235 | desktop-ui | tool-specific | 1 | -1 | v0.17.0 Windows desktop composer can visually corrupt after PageDown while draft text is present. |
| NousResearch/hermes-agent#52566 | voice | tool-specific | 1 | -1 | v0.17.0 desktop TTS playback can time out through the hermes-media:// protocol even when generated audio files are valid. |
| NousResearch/hermes-agent#51321 | desktop-ui | tool-specific | 1 | -1 | v0.17.0 Windows Artifacts view can display file/session timestamps months off from actual file and database times. |
| NousResearch/hermes-agent#53514 | desktop-linux | runtime | 1 | -1 | Linux desktop v0.17.0 hides the close button when titleBarStyle is hidden, leaving Alt+F4 as the exit path |
| NousResearch/hermes-agent#53667 | tool-loader | runtime | 5 | -5 | Fresh v0.17.0 install can collapse runtime tools to cronjob-only despite doctor reporting tools available |
| NousResearch/hermes-agent#50928 | desktop-ui | runtime | 2 | -2 | v0.17.0 desktop deleted sessions can reappear after Load more/refresh due to state race |
| NousResearch/hermes-agent#53173 | windows-update | runtime | 3 | -3 | Machines updated to v0.17.0 can retain stale Windows gateway .cmd autostart launchers that respawn console windows |
| NousResearch/hermes-agent#53026 | matrix | channel-specific | 1 | -1 | v0.17.0 Matrix adapter splits messages at 4000 chars, breaking Markdown tables despite Matrix allowing larger events |
| NousResearch/hermes-agent#52823 | file-tools | tool-specific | 3 | -3 | v0.17.0 write_file with non-local terminal backend resolves remote absolute paths on the macOS host |
| NousResearch/hermes-agent#53659 | desktop-ui | runtime | 2 | -2 | Hermes Desktop v0.17.0 Ctrl/Cmd+Enter can miss freshly typed steer text before state flushes |
| NousResearch/hermes-agent#53443 | qqbot | channel-specific | 3 | -3 | v0.17.0 QQBot adapter connect signature rejects is_reconnect and fails every gateway startup |
| NousResearch/hermes-agent#54077 | voice | tool-specific | 1 | -1 | Hermes Desktop v0.17.0 generates Windows SAPI TTS audio but opens it in Windows Media Player instead of inline playback. |
| NousResearch/hermes-agent#55180 | simplex | channel-specific | 2 | -2 | v0.17.0 SimpleX adapter silently drops non-audio XFTP attachments before completed downloads reach the agent. |

## Stats
- Evidence penalty: **-230**
- Vote penalty: **website-calculated**
- Score formula: **website-calculated from data/settings.json**

---
_[IsItStable.com](https://isitstable.com/hermes-agent/v0.17.0) — score-first release stability evidence_

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

[v0.17.0] [Hermes Agent] Is it stable? #143

Stability Score Evidence — v0.17.0

Stability adjustments

Affected components

Evidence

Stats

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Issue	Area	Type	Severity	Penalty	Reason
NousResearch/hermes-agent#51405	feishu	runtime	5	-5	v0.17.0 gateway startup can crash on fresh Feishu installs because cron.scheduler_provider is missing.
NousResearch/hermes-agent#50734	approval-boundary	kernel	5	-5	v0.17.0 read_file/tool path can ignore safety directives and exfiltrate full .env credentials to LLM providers.
NousResearch/hermes-agent#49578	approval-boundary	kernel	5	-5	v0.17.0 execute_code can bypass agent file edit restrictions and write outside the allowed edit path.
NousResearch/hermes-agent#51447	auth	runtime	4	-4	v0.17.0 Bitwarden profile-suffixed secrets may not reach platform plugins, silently breaking non-default Telegram bot profiles.
NousResearch/hermes-agent#51029	telegram	runtime	4	-4	v0.17.0 multiplexer env override handling can leak a default profile platform token into a secondary Telegram profile.
NousResearch/hermes-agent#50806	runtime	runtime	4	-4	v0.17.0 CLI can hang after a shell tool starts a detached child process, repeatedly submitting tool results instead of returning control.
NousResearch/hermes-agent#50765	runtime	runtime	4	-4	Hermes ACP session/prompt hangs after conversation turn on Windows v0.17.0 while v0.16.0 works, blocking ACP clients.
NousResearch/hermes-agent#50394	mcp	runtime	4	-4	v0.17.0 MCP bridge can churn after one failing stdio server, leaving healthy servers' tools as unknown at runtime.
NousResearch/hermes-agent#49627	runtime	runtime	4	-4	v0.17.0 setup can fail during agent initialization with init_agent() receiving an unexpected event_callback argument.
NousResearch/hermes-agent#49305	message-delivery	runtime	4	-4	Hermes Desktop/TUI on v0.17.0 can lose a user message and persist no assistant response when an additional profile uses codex_app_server.
NousResearch/hermes-agent#51448	provider-routing	runtime	3	-3	Hermes Desktop v0.17.0 on native Windows can return empty LM Studio streams while the same config works under WSL.
NousResearch/hermes-agent#51381	provider-routing	runtime	3	-3	v0.17.0 can send malformed tools[] schemas without function names to OpenCode/DeepSeek backends, causing provider deserialization failure.
NousResearch/hermes-agent#51316	mcp	tool-specific	3	-3	v0.17.0 MCP single-query snapshots can miss slow-start stdio servers, producing incomplete tool lists until reload.
NousResearch/hermes-agent#51290	runtime	runtime	3	-3	v0.17.0 background skill review can block ACP multi-turn IDE sessions after the first turn.
NousResearch/hermes-agent#51030	telegram	runtime	3	-3	v0.17.0 Telegram multiplexer token-collision detection can miss duplicate tokens in profile adapter config.
NousResearch/hermes-agent#50959	provider-routing	runtime	3	-3	v0.17.0 can send invalid Unicode in tools[].function.description, producing HTTP 400s across providers.
NousResearch/hermes-agent#50807	runtime	runtime	3	-3	v0.17.0 forwards very large shell tool output to providers without bounding the payload.
NousResearch/hermes-agent#50794	desktop-macos	runtime	3	-3	Hermes Desktop v0.17.0 can crash on launch on macOS 26.5 with EXC_BREAKPOINT in fontations_ffi.
NousResearch/hermes-agent#50252	provider-routing	runtime	3	-3	v0.17.0 GitHub Copilot provider can return an empty base URL on every inference despite a valid token and detected models.
NousResearch/hermes-agent#50209	desktop-ui	runtime	3	-3	v0.17.0 Desktop boot can hang around the backend status probe even while the CLI itself is healthy.
NousResearch/hermes-agent#50199	provider-routing	runtime	3	-3	v0.17.0 delegation.base_url can be ignored at runtime, forcing incorrect provider endpoint resolution.
NousResearch/hermes-agent#49976	update-system	runtime	3	-3	v0.17.0 Windows ZIP-fallback update path can drop gateway resume state and make restarts fail on stale gateway_state.json.
NousResearch/hermes-agent#49961	provider-routing	runtime	3	-3	v0.17.0 reports configured tools not being invoked across NVIDIA NIM and OpenRouter conversation flows.
NousResearch/hermes-agent#49903	desktop-ui	runtime	3	-3	v0.17.0 Desktop can throw Composer is not available after upgrade, making the composer input unresponsive.
NousResearch/hermes-agent#49590	auth	runtime	3	-3	v0.17.0 gateway startup can prefer literal op:// secret references over resolved cached values, breaking secret-backed channels.
NousResearch/hermes-agent#49490	opencode	runtime	3	-3	OpenCode Zen provider calls fail after updating to v0.17.0, breaking that provider path.
NousResearch/hermes-agent#49445	web-tools	tool-specific	3	-3	Since v0.17.0 the official Docker image can ship with search.exa/web_search/web_extract unusable.
NousResearch/hermes-agent#49283	approval-boundary	kernel	3	-3	v0.17.0 desktop execute_code consent gate still rejects an explicitly consented code execution request.
NousResearch/hermes-agent#47917	desktop-windows	runtime	3	-3	Desktop builds can continue failing after v0.17.0 updates because electronDist disappears from the build path.
NousResearch/hermes-agent#47868	provider-routing	runtime	3	-3	Strict chat-completions providers on v0.17.0 reject leaked messages[].timestamp metadata with HTTP 400s.
NousResearch/hermes-agent#36718	compaction	kernel	3	-3	After updating to v0.17.0, context compression can repeatedly trigger because last_prompt_tokens stays stale until the next API call.
NousResearch/hermes-agent#51278	provider-routing	runtime	2	-2	v0.17.0 title generation can route the main model name to the fallback provider endpoint.
NousResearch/hermes-agent#51263	email	channel-specific	2	-2	v0.17.0 email STARTTLS setup can fail with OpenSSL WRONG_VERSION_NUMBER on OpenSSL 3.5+ systems.
NousResearch/hermes-agent#51155	runtime	runtime	2	-2	v0.17.0 user personalities can fail to change and persist across sessions.
NousResearch/hermes-agent#51099	memory	runtime	2	-2	v0.17.0 can activate the Honcho memory provider even when honcho-ai is not installed.
NousResearch/hermes-agent#50966	browser	tool-specific	2	-2	Desktop GUI on v0.17.0 routes browser_vision calls to browser_scroll instead of the requested vision tool.
NousResearch/hermes-agent#50823	installer	runtime	2	-2	Desktop build for v0.17.0 hard fails on Windows native non-git checkouts because build-stamp cannot determine a git commit.
NousResearch/hermes-agent#50769	installer	runtime	2	-2	Windows installer for v0.17.0 detects a Python 3.12 fallback but still tries Python 3.11 for venv creation, failing install.
NousResearch/hermes-agent#50737	gateway-service	runtime	2	-2	v0.17.0 desktop profile-scoped OpenRouter setup can leave gateway inference not ready.
NousResearch/hermes-agent#50713	message-delivery	runtime	2	-2	v0.17.0 can lose chat text after reopening sessions across multiple deployments.
NousResearch/hermes-agent#50675	gateway-service	runtime	2	-2	v0.17.0 gateway agent cache can reset /usage API-call and token counters during follow-up turns.
NousResearch/hermes-agent#50404	discord	channel-specific	2	-2	v0.17.0 Discord config is not profile-isolated, allowing channel state to bleed across profiles.
NousResearch/hermes-agent#50395	mcp	tool-specific	2	-2	v0.17.0 hermes mcp test can report a server healthy under a different PATH than the gateway MCP spawn actually uses.
NousResearch/hermes-agent#50229	model-routing	runtime	2	-2	v0.17.0 silent fallback failover can run the agent as the wrong model without notifying the user.
NousResearch/hermes-agent#50210	installer	runtime	2	-2	v0.17.0 Windows bootstrap can produce an unsigned Hermes.exe blocked by Smart App Control.
NousResearch/hermes-agent#50205	feishu	channel-specific	2	-2	v0.17.0 Docker image can miss lark-oapi, causing the Feishu channel to fail at connect time.
NousResearch/hermes-agent#50175	config-profiles	runtime	2	-2	v0.17.0 profile deletion can fail while dashboard/gateway child processes are alive, making profiles reappear until reboot.
NousResearch/hermes-agent#49982	desktop-ui	runtime	2	-2	v0.17.0 Windows Desktop can time out connecting to the backend while the CLI still works.
NousResearch/hermes-agent#49870	dashboard	tool-specific	2	-2	v0.17.0 task Done/ready flow can trigger a false diagnostic and then get blocked by the respawn guard.
NousResearch/hermes-agent#49606	update-system	runtime	2	-2	Windows Desktop v0.17.0 update flow can leave users with a mismatched backend/desktop state after silent GUI update failure.
NousResearch/hermes-agent#49561	whatsapp	channel-specific	2	-2	v0.17.0 regression can prevent installing the WhatsApp bridge in Docker environments.
NousResearch/hermes-agent#49529	installer	tool-specific	2	-2	The v0.17.0 PyPI wheel install has a doctor false-positive and omits optional-skills from the wheel package.
NousResearch/hermes-agent#49297	provider-routing	tool-specific	2	-2	Gemma4 through the Ollama backend still hits near-immediate finish_reason=length truncation on v0.17.0.
NousResearch/hermes-agent#49242	windows-runtime	runtime	2	-2	Windows WhatsApp gateway and updater paths on v0.17.0 can use system Node/npm instead of Hermes-managed tooling.
NousResearch/hermes-agent#50775	telegram	channel-specific	1	-1	Telegram macOS client streaming on v0.17.0 can show ghosting/double-text during edited-message updates.
NousResearch/hermes-agent#49955	dashboard	tool-specific	1	-1	v0.17.0 Kanban inline create form truncates field labels in the dashboard.
NousResearch/hermes-agent#49908	dashboard	tool-specific	1	-1	v0.17.0 drag-and-drop to Done can show a completion prompt and then silently revert the task.
NousResearch/hermes-agent#49745	desktop-ui	tool-specific	1	-1	Desktop v0.17.0 composer undo skips pasted content because paste operations lack a contentEditable undo entry.
NousResearch/hermes-agent#49732	model-routing	tool-specific	1	-1	v0.17.0 dashboard model-switch dialog can fire with no active chat and flip the badge before the user decides.
NousResearch/hermes-agent#49600	desktop-ui	tool-specific	1	-1	v0.17.0 desktop UI can clip the audio icon on Windows.
NousResearch/hermes-agent#49536	telegram	channel-specific	1	-1	Telegram finalize messages on v0.17.0 can overlap text because parse_mode mutation falls back incorrectly.
NousResearch/hermes-agent#49294	gateway-service	tool-specific	1	-1	Windows Desktop restart gateway action on v0.17.0 opens a persistent CMD window instead of restarting the installed headless service.
NousResearch/hermes-agent#49253	imessage	channel-specific	1	-1	Photon iMessage on v0.17.0 corrupts Unicode when Markdown bold formatting is present.
NousResearch/hermes-agent#52023	provider-routing	runtime	3	-3	Hermes Desktop v0.17.0 on Windows returns OpenAI 400 errors for GPT-4o-mini, GPT-4.1 and related models because encrypted content is sent to models that do not support it.
NousResearch/hermes-agent#51864	dashboard	runtime	3	-3	Hermes v0.17.0 self-hosted kanban can route task workspaces and board filters through global active-board state instead of task tenant, mixing tenants.
NousResearch/hermes-agent#52013	desktop-ui	channel-specific	1	-1	Hermes Desktop v0.17.0 on Chinese Windows can render Chinese TUI/desktop text as mojibake, making non-ASCII output unreadable.
NousResearch/hermes-agent#51646	memory	runtime	4	-4	Hermes v0.17.0 gateway memory writes can omit the active column, causing INSERT failures and lost persisted state.
NousResearch/hermes-agent#51873	auth	runtime	3	-3	Hermes Desktop v0.17.0 can loop on 401s and never show the documented sign-in UI when connecting to a remote backend with basic auth.
NousResearch/hermes-agent#51871	approval-boundary	runtime	4	-4	Hermes v0.17.0 legacy chat-completions streaming can omit approval events, so clients see text deltas but no tool-approval boundary.
NousResearch/hermes-agent#52753	desktop-windows	runtime	5	-5	v0.17.0 packaged Windows desktop can fail to launch after update because simple-git is missing from the Electron bundle.
NousResearch/hermes-agent#52378	update-system	runtime	4	-4	v0.17.0 Windows desktop auto-update can loop indefinitely, repeatedly spawning elevated updater processes even when behind=0.
NousResearch/hermes-agent#52428	browser	tool-specific	3	-3	v0.17.0 browser_navigate fails with missing _hermes_read_browser_output, breaking browser tool navigation on Windows.
NousResearch/hermes-agent#52496	model-routing	runtime	3	-3	v0.17.0 dashboard model switcher can rewrite named custom providers to openrouter/bare custom, breaking local proxy routing.
NousResearch/hermes-agent#52492	provider-routing	runtime	2	-2	v0.17.0 Codex model selection can persist an unsupported gpt-5.3-codex custom model and then fail prompts with HTTP 400.
NousResearch/hermes-agent#52363	telegram	channel-specific	2	-2	v0.17.0 Telegram outbound messages can be stored as incoming unread items, making inbox badges grow on the agent's own replies.
NousResearch/hermes-agent#52573	voice	tool-specific	2	-2	v0.17.0 TUI continuous voice mode can fail to reopen the microphone after TTS playback on macOS.
NousResearch/hermes-agent#51882	desktop-ui	runtime	2	-2	v0.17.0 Desktop Chat→Personality setting writes display.personality but not agent.system_prompt, so the persona is silently ignored.
NousResearch/hermes-agent#51684	feishu	channel-specific	2	-2	v0.17.0 Feishu approval-card clicks reject FEISHU_ALLOWED_USERS=* wildcard even though normal inbound messages are allowed.
NousResearch/hermes-agent#52235	desktop-ui	tool-specific	1	-1	v0.17.0 Windows desktop composer can visually corrupt after PageDown while draft text is present.
NousResearch/hermes-agent#52566	voice	tool-specific	1	-1	v0.17.0 desktop TTS playback can time out through the hermes-media:// protocol even when generated audio files are valid.
NousResearch/hermes-agent#51321	desktop-ui	tool-specific	1	-1	v0.17.0 Windows Artifacts view can display file/session timestamps months off from actual file and database times.
NousResearch/hermes-agent#53514	desktop-linux	runtime	1	-1	Linux desktop v0.17.0 hides the close button when titleBarStyle is hidden, leaving Alt+F4 as the exit path
NousResearch/hermes-agent#53667	tool-loader	runtime	5	-5	Fresh v0.17.0 install can collapse runtime tools to cronjob-only despite doctor reporting tools available
NousResearch/hermes-agent#50928	desktop-ui	runtime	2	-2	v0.17.0 desktop deleted sessions can reappear after Load more/refresh due to state race
NousResearch/hermes-agent#53173	windows-update	runtime	3	-3	Machines updated to v0.17.0 can retain stale Windows gateway .cmd autostart launchers that respawn console windows
NousResearch/hermes-agent#53026	matrix	channel-specific	1	-1	v0.17.0 Matrix adapter splits messages at 4000 chars, breaking Markdown tables despite Matrix allowing larger events
NousResearch/hermes-agent#52823	file-tools	tool-specific	3	-3	v0.17.0 write_file with non-local terminal backend resolves remote absolute paths on the macOS host
NousResearch/hermes-agent#53659	desktop-ui	runtime	2	-2	Hermes Desktop v0.17.0 Ctrl/Cmd+Enter can miss freshly typed steer text before state flushes
NousResearch/hermes-agent#53443	qqbot	channel-specific	3	-3	v0.17.0 QQBot adapter connect signature rejects is_reconnect and fails every gateway startup
NousResearch/hermes-agent#54077	voice	tool-specific	1	-1	Hermes Desktop v0.17.0 generates Windows SAPI TTS audio but opens it in Windows Media Player instead of inline playback.
NousResearch/hermes-agent#55180	simplex	channel-specific	2	-2	v0.17.0 SimpleX adapter silently drops non-audio XFTP attachments before completed downloads reach the agent.

Uh oh!

[v0.17.0] [Hermes Agent] Is it stable? #143

Description

Stability Score Evidence — v0.17.0

Stability adjustments

Affected components

Evidence

Stats

Metadata

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Issue actions