Environment: prime-agent 0.9.1 (upgraded from 0.8.1 on September 2, 2026; the daemon was replaced on update, protocol schema 22 → 25) · Node v24.13.0 · Linux Mint 21.x (XFCE)
Symptom: In the agents view, selecting a running RLM subagent session shows "waiting to load" indefinitely and never attaches. The same action worked on 0.8.1 on September 1, 2026.
Reproduction Steps:
- In a root session, spawn an RLM subagent via the
rlm tool.
- Open the agents view while the subagent is running.
- Select the subagent's session row.
- Observe: "waiting to load" persists; the session never opens.
Evidence:
- The daemon log shows no attach request and no error at the time of the attempts — the hang appears to be on the client/view side.
- The subagent session itself is healthy: the roster reports it as running and streaming with an increasing message count, and API-level observation (recent-messages preview) returns live content.
attachedClients stays 0.
- No relevant entries appear in
client-errors.log for the attempts.
Suspected relation: the 0.9.0 roster-subscription rewrite (#1900) and the roster boot-seed scoping. Version 0.9.1 already fixed one sibling regression (the Inactive section loading empty until a search, same root cause). This looks like another: the session-open path for subagent rows may be waiting on data the scoped seed/subscription no longer delivers.
Disclosure: this install carries small local patches to agents-view section classification (list rendering only — nothing in the attach/open path). Happy to re-test against a clean, unpatched install if useful.
Environment: prime-agent 0.9.1 (upgraded from 0.8.1 on September 2, 2026; the daemon was replaced on update, protocol schema 22 → 25) · Node v24.13.0 · Linux Mint 21.x (XFCE)
Symptom: In the agents view, selecting a running RLM subagent session shows "waiting to load" indefinitely and never attaches. The same action worked on 0.8.1 on September 1, 2026.
Reproduction Steps:
rlmtool.Evidence:
attachedClientsstays 0.client-errors.logfor the attempts.Suspected relation: the 0.9.0 roster-subscription rewrite (#1900) and the roster boot-seed scoping. Version 0.9.1 already fixed one sibling regression (the Inactive section loading empty until a search, same root cause). This looks like another: the session-open path for subagent rows may be waiting on data the scoped seed/subscription no longer delivers.
Disclosure: this install carries small local patches to agents-view section classification (list rendering only — nothing in the attach/open path). Happy to re-test against a clean, unpatched install if useful.