[alexwill87]

Summary

• Forward attach_to_session through the registered cronjob tool handler.
• Preserve explicit false as the per-job override of cron.mirror_delivery.
• Add an E2E temporary-HERMES_HOME create/update/readback regression test for true then false.

Validation

• scripts/run_tests.sh tests/tools/test_cronjob_tools.py -v --tb=short (78 passed)
• scripts/run_tests.sh tests/cron/test_scheduler.py -v --tb=short (222 passed)
• git diff --check

[alt-glitch]

This was generated by AI during triage.

Duplicate of #57556 for the cronjob handler's attach_to_session passthrough. #57556 already covers the same registered-tool boundary; this PR also bundles unrelated changes that should be split before review.

[teknium1]

Thanks for tracing the registered-tool boundary. The focused cron fix is still needed on current main: tools/cronjob_tools.py:1039-1042 exposes attach_to_session, and cronjob() accepts/persists it (:643, :716, :894-895), but the registered handler at :1078-1100 does not forward it.

Problems

• The PR also includes unrelated commit 87c8244f150f (25 files, +2573/-328), including agent-loop, Kanban, dashboard-auth, and Langfuse work. Please keep this fix focused; AGENTS.md:102-107 also requires behavioral settings such as the added HERMES_KANBAN_BUDGET_PRESTOP_REMAINING to use config.yaml.
• Explicit False is a meaningful override because cron/scheduler.py:633-639 gives it precedence over cron.mirror_delivery, but _format_job at tools/cronjob_tools.py:531-565 does not return either boolean. The new test verifies storage but not that update readback preserves false.

Suggested changes

• Isolate the registry forwarding fix and its regression test from commit 87c8244f150f.
• Return either stored boolean from _format_job and assert the registered-handler update response contains attach_to_session: false.

This is an automated hermes-sweeper review.

[kshitijk4poor]

The attach_to_session persistence fix this PR set out to make has been salvaged onto current main via #95366 (from #84829). The remainder of this diff mixes a large amount of unrelated kanban/dashboard/langfuse work into the same branch, which made it unsalvageable as a unit. Closing — if any of the other changes here are still relevant against current main, please resubmit them as focused single-concern PRs.