Error when restarting DSH (preset failed to mount)

[elibroftw]

Error: agent-presents: present "standard" failed to mount: failed to apply loader entry persona (@deepseek-ai/dsh-persona): prompt section "deployment-persona" is already registered (for a per-agent override, register through the agent's agent.ctx instead).
(...agent.cordis.yml)

[elibroftw]

#4263

[nokkies]

This is a known failure class, not a persona-specific bug — and knowing that changes the workaround as well as the fix.

The message comes from the system-prompt registry (packages/core/system-prompt), which refuses a duplicate section name. But the duplicate is not really a second plugin registering the same name: it is the same preset mounted twice, with the first mount never torn down.

ensureStanding in agent-presets drops the standing pointer when a preset's composition file changes and mounts the next generation, and its own TODO says the superseded one is never reclaimed:

TODO: reclaim the superseded generation once the last agent joined to it is gone. The subtree is not inert — dsh-skill-filesystem watches its roots

The old generation's registrations therefore stay live, and the new generation collides with them. Whichever process-global registry the preset touches first is the one that reports it — which is why this looks like a different bug each time it happens:

• registering into the Cordis inspect registry produces Host Cordis inspect provider "Service" is already registered (tool-cordis inspect providers are process-global: two cordis-based presets cannot coexist in one process #4675, and [Bug Report] 运行中热重载 cordis.patch.yml 插入 MCP 插件后，新建会话/选择工作区失败（界面闪回未选择状态） #902 via hot reload);
• registering a prompt section produces exactly what you got.

Same cause, different messenger. So the linked #4263 and the cordis reports are all this one thing.

Practical consequence for you: a full process restart clears it, but only if the preset composition is not re-edited afterwards — the collision returns on the next composition change within the same process, not on the next launch. If you are editing preset files (or using the settings-page preset authoring flow, which writes on every save), expect it to recur, and that is not something a restart fixes permanently.

Worth knowing for anyone tempted to fix the registry: making the prompt-section registry tolerate duplicates would remove this message and leave the leak. The superseded generation still holds its watchers, and the next registry it touches produces the next variant of this report. I argued the same point on #902, where the cordis-inspect equivalent is being made duplicate-tolerant: the error disappearing will not mean the leak is gone, and the leak deserves tracking on its own rather than being noticed by whichever registry happens to complain.

(The registry guard itself is right, incidentally — its advice about registering through agent.ctx for a per-agent override is correct and worth keeping. It is simply answering a question you did not ask, because the second registration came from a generation that should not still exist.)