[v0.15.0] [Hermes Agent] Is it stable?

[exisz]

Stability Score Evidence — v0.15.0

Major release, major blast radius: PyPI manifests, Docker startup, dashboard auth/events, Codex request shaping, gateway/session bugs, and platform delivery regressions all had explicit 0.15.0/v2026.5.28 evidence.

Stability adjustments

Affected components

• plugin-system
• plugin-install
• runtime
• linux-runtime
• ui
• gateway
• mcp
• codex
• provider-integration
• windows
• telegram
• model-routing
• feishu
• approval-boundary
• browser
• cron
• tool-registration
• message-delivery
• wechat
• memory
• docker
• kanban

Evidence

Issue	Area	Type	Severity	Penalty	Reason
NousResearch/hermes-agent#34034	plugin-system	packaging	5	-5	0.15.0 PyPI wheel omitted plugin.yaml manifests, so bundled platform adapters did not load
NousResearch/hermes-agent#34071	runtime	packaging	5	-5	0.15.0 Docker image missed init scripts/container_boot replacement and exited on startup
NousResearch/hermes-agent#36208	runtime	container-runtime	4	-4	v2026.5.28 Docker/s6 image crash-looped in non-PID-1 runtimes
NousResearch/hermes-agent#34091	ui	runtime	3	-3	0.15.0 Docker dashboard TUI/event feed failed due WebSocket peer IP guard
NousResearch/hermes-agent#34289	ui	auth-runtime	3	-3	0.15.0 dashboard loopback mode reload-looped on /api/auth/me 401
NousResearch/hermes-agent#34871	mcp	packaging	3	-3	0.15.0 pip wheel omitted top-level mcp_serve, crashing hermes mcp serve
NousResearch/hermes-agent#34113	codex	provider-integration	4	-4	0.15.0 Codex requests failed HTTP 400 when participant display_name contained emoji/non-ASCII
NousResearch/hermes-agent#34140	windows	runtime	3	-3	0.15.0 WSL interactive CLI froze after /new or /reset
NousResearch/hermes-agent#33961	windows	runtime	3	-3	0.15.0 still froze terminal on /new or /reset without the now suffix
NousResearch/hermes-agent#34205	telegram	provider-integration	4	-4	0.15.0 Telegram sessions could be poisoned by stale encrypted_content after provider switch
NousResearch/hermes-agent#34179	codex	provider-integration	3	-3	0.15.0 direct Copilot large resumes hit 12s Codex first-byte watchdog
NousResearch/hermes-agent#35032	feishu	approval-boundary	3	-3	Upgrade from 0.14.0 to 0.15.0/0.15.1 broke Feishu approval buttons when admins were not configured
NousResearch/hermes-agent#35207	feishu	approval-boundary	3	-3	v0.15 silently deprecated FEISHU_ALLOW_ALL_USERS, blocking approval-card callbacks under default allowlist
NousResearch/hermes-agent#35595	ui	tool-specific	2	-2	0.15.0 slash commands returned raw structured i18n keys instead of human-readable text
NousResearch/hermes-agent#35306	browser	tool-specific	2	-2	0.15.0 browser_console could crash CDP when JavaScript returned DOM nodes
NousResearch/hermes-agent#34252	gateway	persistence	3	-3	v0.14.0+ including v2026.5.28 could fail atomic_replace with EXDEV on cross-filesystem HERMES_HOME
NousResearch/hermes-agent#34253	feishu	sessions-management	3	-3	v2026.5.28 Feishu cancellation path could orphan session guards and permanently block messages
NousResearch/hermes-agent#35561	cron	tool-registration	2	-2	v2026.5.28 cronjob tool could remain hidden when tool definitions cached before gateway env setup
NousResearch/hermes-agent#36264	mcp	tool-specific	3	-3	v0.15.x HTTP MCP test could pass while actual tool calls failed as not connected
NousResearch/hermes-agent#37005	message-delivery	platform-adapter	3	-3	v0.15.x Mattermost plugin send_message failed inside gateway adapter path
NousResearch/hermes-agent#35062	wechat	message-delivery	3	-3	v0.15 Weixin cron push regressed from v0.14 and silently failed tokenless retry delivery
NousResearch/hermes-agent#37011	gateway	runtime	4	-4	0.15.0/0.15.1 gateway reconnect loop leaked sqlite file descriptors until all platforms failed

| NousResearch/hermes-agent#34070 | memory | runtime | 3 | -3 | v0.15.0 fresh hermes chat -q subprocesses can hang indefinitely in Honcho memory prefetch. |
| NousResearch/hermes-agent#34270 | message-delivery | security | 4 | -4 | Hermes v0.15.0 send_message can bypass validate_media_delivery_path for Telegram media delivery. |
| NousResearch/hermes-agent#34457 | docker | runtime | 3 | -3 | v0.15.0 multi-container gateway/dashboard shared volumes can trigger endless s6-log lock crash loops. |
| NousResearch/hermes-agent#35202 | kanban | runtime | 3 | -3 | Hermes Agent v0.15.0 Docker kanban workers can bypass failure-limit circuit breakers when crashing before first heartbeat. |
| NousResearch/hermes-agent#32747 | kanban | runtime | 2 | -2 | v0.15.0 kanban protocol violations with rc=0 but rejected completion can fail to increment failure_limit. |

Evidence gate notes

• Included: #34070, #34270, #34457, #35202, and #32747 after Gate A/Gate B review because each body/comment explicitly identifies v0.15.0 / Hermes Agent v0.15.0 as affected or reproduced.
• Excluded: #39538 for v0.15.0 because its negative report is v0.16/latest Desktop behavior; v0.15.0 appears only as historical IME/TUI context.
• Included: Gate A included #34034, #34071, #34091, #34289, #34871, #34113, #34140, #33961, #34205, #34179, #35032, #35595, #35306, and #37011 because each body/comment/debug report explicitly names 0.15.0 or Hermes Agent v0.15.0 (2026.5.28) as affected/reproduced. Gate B included #36208, #34252, #34253, and #35561 because v2026.5.28 is the release tag for this version, and #35207, #36264, #37005, #35062 because their explicit v0.15/v0.15.x ranges cover 0.15.0.
• Excluded notable candidates: #34206 and #34398 are duplicate dashboard reload-loop reports covered by #34289; #30149 and #6931 only matched dependency strings containing 0.15.0, not Hermes version evidence; #33365 reports a v0.14 WhatsApp crash resolved after fresh v0.15.0 install; #33214 and #33932 describe v0.14 Codex failures fixed by/updating to v0.15; #36248 is reported on v0.15.2 and was not safely attributable to 0.15.0.

Stats

• Evidence count: 27
• Evidence penalty: -86
• Vote penalty: website-calculated
• Score formula: website-calculated from data/settings.json
• Freshness: historical version; evidence backfilled, score bucket intentionally deferred

---

IsItStable.com — score-first release stability evidence