mosaicstack/stack
1
0
Fork 0
Code Issues 36 Pull Requests 2 Actions Packages Projects Releases 15 Wiki Activity

Fleet never-sleeps: auto-trust workdir for fleet-launched agent sessions (factory orchestrator stalled at 'trust this folder?' gate) #644

New Issue
Open
opened 2026-06-24 03:11:25 +00:00 by jason.woltje · 0 comments
jason.woltje commented 2026-06-24 03:11:25 +00:00
Owner
Copy Link

Goal

Toward an always-on fleet: a single interactive prompt must never be able to silently halt a fleet-launched agent.

Observed (dragon-lin, 2026-06-23)

The mosaic-factory orchestrator (claude) session was stuck at Claude Code's Do you trust the files in this folder? gate. It never started work — 0.0% ctx, heartbeat watchdog still "ok", but zero progress. It only cleared when the 0.0.41 reseed relaunched the session. coder0/coder1/reviewer (pi) were fine; the trust gate is claude-runtime-specific.

Impact

This is a textbook "fleet asleep" failure: the watchdog reports healthy (process up, heartbeating) while the agent is indefinitely blocked on a TTY prompt. No work, no error, no alert.

Proposed fix

Fleet-launched (non-interactive) agent sessions should never hit the trust gate:

  • Pre-trust the agent workdir when the fleet launcher creates the session (Claude Code records trusted folders in its config), and/or
  • Pass the runtime's trust-accept/bypass flag for fleet/systemd-launched sessions (these already run --dangerously-skip-permissions-style bypass in mosaic-agent@), and/or
  • Add a startup health probe that detects a session parked at a known prompt and auto-clears or flags it.

Related

Part of the "fleet that never sleeps" reliability set alongside #642 (stale-framework silent drift) and F1 premature-stop auto-continuation (mosaic coord continue/run, see ORCHESTRATOR-PROTOCOL.md).

## Goal Toward an always-on fleet: a single interactive prompt must never be able to silently halt a fleet-launched agent. ## Observed (dragon-lin, 2026-06-23) The mosaic-factory **orchestrator** (claude) session was stuck at Claude Code's `Do you trust the files in this folder?` gate. It never started work — 0.0% ctx, heartbeat watchdog still "ok", but zero progress. It only cleared when the 0.0.41 reseed relaunched the session. coder0/coder1/reviewer (pi) were fine; the trust gate is claude-runtime-specific. ## Impact This is a textbook "fleet asleep" failure: the watchdog reports healthy (process up, heartbeating) while the agent is indefinitely blocked on a TTY prompt. No work, no error, no alert. ## Proposed fix Fleet-launched (non-interactive) agent sessions should never hit the trust gate: - Pre-trust the agent workdir when the fleet launcher creates the session (Claude Code records trusted folders in its config), and/or - Pass the runtime's trust-accept/bypass flag for fleet/systemd-launched sessions (these already run `--dangerously-skip-permissions`-style bypass in `mosaic-agent@`), and/or - Add a startup health probe that detects a session parked at a known prompt and auto-clears or flags it. ## Related Part of the "fleet that never sleeps" reliability set alongside #642 (stale-framework silent drift) and F1 premature-stop auto-continuation (`mosaic coord continue`/`run`, see ORCHESTRATOR-PROTOCOL.md).
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
fleet-enhancement
Fleet-wide reliability/speed/cost improvements surfaced by the nightly enhance pass
No Label
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
jason.woltje
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: mosaicstack/stack#644
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?