A proving Run opens a Proposal for vaulted-agent#92

[JacobStephens2]

Parent

#51

What to build

A proving Run on vaulted-agent#92 opens a draft Proposal, Routes the issue, appears on Run History, and mails the operator.

Acceptance criteria

• The first Dispatch is vaulted-agent#92 (first Target, lowest Eligible).
• A draft Proposal exists for that Run.
• The issue is Routed from the Run's ending bound and the Proposal's checks.
• Run History shows the ended Run.
• Notify mail arrives at jacob@stephens.page.

Blocked by

• The Dashboard is on HTTPS at the public origin #58
• A dry-run Cycle succeeds #61

[JacobStephens2]

Handoff from #61 (dry-run green, issue closed). What the proving Run inherits: the Journal already holds dry-run cycles 2 (vaulted-agent, picked #92) and 5 (career, picked #81) with empty dispatches and no in-flight lock, so the first live cycle should pick #92 as the first Dispatch. The owner-wide search journaled nine unenrolled repos carrying ready-for-agent (including tracewake itself, expected) — the first live cycle may queue gap mail for these, and the notifier is still failed, so decide whether to revive it before or after the proving Run. Still open from the #60 handoff: /var/lib/conductor/work/* and /home/loop/career do not exist; box facts/progress commands use SSH despite SELECTOR_BOX_HOST=local; the Cycle runs as conductor while Box credentials belong to loop. Full detail: notes/dry-run-61-evidence.md.

[JacobStephens2]

Proving Run complete. vaulted-agent#92 was the first Dispatch in both live cycles (Journal rows 15/17, 22/24), draft Proposal JacobStephens2/vaulted-agent#109 exists (draft, checks green, Run comment posted), the issue is routed to ready-for-human from the agent-failed bound at attempt 2/2 (selector's give-up comment posted; I completed the swap after creating the missing repo label), Run History content verified against the live Journal, and the notifier relayed three mails (gap notice, dispatch-failed, agent-failed outcome) - please confirm they reached your inbox.

What it took is in notes/proving-run-62-evidence.md: revived the stale-failed notifier, wired conductor->loop dispatch (/etc/tracewake/box-as-loop.sh + sudoers, copy at deploy/instance/), taught the credential gate --agent (TDD, green on the Host), cloned the work checkout, and created the missing lifecycle labels in vaulted-agent. Review PR for the product half: #76. Left open for your inbox confirmation and that review. Known follow-ups (facts/progress SSH noise, guardrail unreadable until the Host returns to main, career checkouts/labels, missing OnFailure unit, timer still disabled) are listed in the note.

[JacobStephens2]

Mailbox confirms it: gap notice and agent-failed outcome both arrived. The relay also accepted 'Dispatch failed' at 05:20:27 - if it is not in the inbox, check spam (three same-sender mails inside two minutes). Follow-up transient, not this issue: unattended-upgrades churned services ~06:32-06:38, the notifier crash-looped on dropped Postgres connections and self-recovered via Restart; stable since 06:39 with zero tracebacks. Worth a robustness issue (notifier reconnect + upgrade restarts landing mid-Run). All five criteria verified; evidence in notes/proving-run-62-evidence.md, product half in PR #76.