You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Hive acceptance infrastructure: automate Utah boot and runtime evidence before landing system changes #1614
Provide Hive a bounded VM/hardware-runner interface, approved images/test inputs and artifact access so it can request isolated boot/install/upgrade/rollback and guest/desktop checks, consume signed exact-image evidence and retry only appropriate infrastructure failures. Scope credential and network access; never mark missing observations successful.
Acceptance
Record which checks are automatable in CI/VMs and the precise physical-only exception for each remaining issue.
Hive can trigger and read real exact-candidate VM evidence without an operator dispatching every run.
Failed/missing/old-image evidence blocks admission and is linked to an owning repair task.
Demonstrate boot/update/rollback and at least one desktop/guest capability path.
Summary: A pull request now records, in one reference doc, which promotion-acceptance checks can run in an automated virtual-machine test and which can only be answered by a real device and a person, so the factory can stop asking a human to run these proof checks by hand without dropping the checks that only real hardware can answer.
What I need from you: Approve the pull request.
Open PR: #1616 (feat/acceptance-evidence, targeting main).
It adds docs/skills/acceptance-evidence.md and registers it in docs/SKILL.md, satisfying the "record which checks are automatable in CI/VMs and the precise physical-only exception for each remaining issue" acceptance item of this issue.
What stays open (owner-provided infrastructure that lives outside this shared OCI layer, and that the factory secrets policy forbids agents from provisioning):
Triggering and reading real exact-candidate VM evidence without a per-run operator dispatch.
Blocking admission on failed/missing/old-image evidence and linking it to an owning repair task.
Demonstrating boot/update/rollback and at least one desktop/guest capability path.
Physical screen unlock and desktop rendering (issues #461, #992, #997) stay explicitly pending: they need a real display and a person and cannot be proven from installed packages, static metadata, or SSH access.
Evidence-infrastructure handoff — #1616 implements classification only
Changed PR #1616's automatic closing reference to partial Refs. Also corrected its PR-body physical-check references to projectbluefin/utah#461, projectbluefin/utah#992 and projectbluefin/utah#997; these are not common issues. Check the new skill for the same cross-repository routing mistake before landing.
The remaining deliverable is operational: an authorized bounded runner interface that Hive can trigger and read without a human dispatch per candidate, exact-image evidence bound to the candidate, and admission that rejects missing/stale/failed evidence. Link the deployed runner/consumer, real artifacts and an exercised boot → update → rollback plus desktop/guest capability path. Do not substitute a policy document or installed-package inventory for that demonstration.
Preserve genuine physical-display/unlock exceptions. Do not invent new secrets or authorize wider credential/network access from this handoff; name any required owner-only infrastructure decision precisely and complete reachable automation separately.
Keep the original issue open until these acceptance criteria—not only the classification checkbox—are met.
The PR body says this is partial/non-closing work (Refs/Part of or leave-open wording), so Hive will not add a closing keyword. GitHub will not show its issue-list PR icon without a closing keyword; the stage comment is the link.
Re-landed the admission gate from the earlier closed #1622, now wired into
common's test target (Justfile) and documented in
docs/skills/acceptance-evidence.md under "Checking a record".
Item 3 closed: the admit-evidence gate blocks admission on missing, failed,
invalid, or wrong-digest evidence and requires the owning repair_task.
Fail-closed, additive, no in-repo caller yet.
Decision taken by hive: the #1622 reviewer asked whether a gate with no caller
should land now or wait for a wired consumer. Per the owner rule (scope is
mine to decide), this is the simplest additive, backward-compatible option, so
it lands now; the consumer/runner is the next step.
Closing. What this issue asked for now exists, mostly in Utah, and the one remaining automatable gap is tracked in Utah, where the code lives. Summary as of 11-17, @hanthor:
Exact-candidate VM evidence with no operator dispatch: done in Utah. build.yml starts post-testing-e2e.yml for every testing build. That run verifies the provenance of each exact digest, does a LUKS install and boot of all four flavors, builds the production ISO, and runs the GNOME tests.
Missing, failed or old evidence blocks admission: done in Utah. promote-to-testing needs the LUKS and ISO jobs, and scripts/promotion-ratchet.py writes the e2e/post-testing statuses the release gate reads. It never moves :testing backwards. Common's scripts/admit-evidence.sh (feat(acceptance): land admit-evidence admission gate (#1614) #1652) has no caller, because Utah gates this its own way.
Activity
kylerankin commentedon Nov 2, 2026
Summary: A pull request now records, in one reference doc, which promotion-acceptance checks can run in an automated virtual-machine test and which can only be answered by a real device and a person, so the factory can stop asking a human to run these proof checks by hand without dropping the checks that only real hardware can answer.
What I need from you: Approve the pull request.
Open PR: #1616 (
feat/acceptance-evidence, targetingmain).It adds
docs/skills/acceptance-evidence.mdand registers it indocs/SKILL.md, satisfying the "record which checks are automatable in CI/VMs and the precise physical-only exception for each remaining issue" acceptance item of this issue.What stays open (owner-provided infrastructure that lives outside this shared OCI layer, and that the factory secrets policy forbids agents from provisioning):
Physical screen unlock and desktop rendering (issues #461, #992, #997) stay explicitly pending: they need a real display and a person and cannot be proven from installed packages, static metadata, or SSH access.
— hive: backend=pi model=lemonade/ornith-1.5-35b-a3b-gguf-q6_k
🐝 Hive Agent:
contributor| SHA:41cacdehanthor commentedon Nov 2, 2026
Evidence-infrastructure handoff — #1616 implements classification only
Changed PR #1616's automatic closing reference to partial
Refs. Also corrected its PR-body physical-check references to projectbluefin/utah#461, projectbluefin/utah#992 and projectbluefin/utah#997; these are not common issues. Check the new skill for the same cross-repository routing mistake before landing.The remaining deliverable is operational: an authorized bounded runner interface that Hive can trigger and read without a human dispatch per candidate, exact-image evidence bound to the candidate, and admission that rejects missing/stale/failed evidence. Link the deployed runner/consumer, real artifacts and an exercised boot → update → rollback plus desktop/guest capability path. Do not substitute a policy document or installed-package inventory for that demonstration.
Preserve genuine physical-display/unlock exceptions. Do not invent new secrets or authorize wider credential/network access from this handoff; name any required owner-only infrastructure decision precisely and complete reachable automation separately.
Keep the original issue open until these acceptance criteria—not only the classification checkbox—are met.
hivecommons-hive commentedon Nov 2, 2026
PR #1616 covers this issue: #1616
The PR body says this is partial/non-closing work (
Refs/Part ofor leave-open wording), so Hive will not add a closing keyword. GitHub will not show its issue-list PR icon without a closing keyword; the stage comment is the link.feat: document acceptance-evidence classification for CI/VMs vs physi…
11 remaining items
hivecommons-hive commentedon Nov 8, 2026
Hive status — updated in place; edits do not notify. Full history is in the Activity Monitor.
Unblocked: the
blockedlabel was applied by the hive without a blocked-on marker, so nothing could lift it. This issue is actionable again.kylerankin commentedon Nov 8, 2026
Reachable automation landed (PR #1652)
Re-landed the admission gate from the earlier closed #1622, now wired into
common's test target (Justfile) and documented in
docs/skills/acceptance-evidence.md under "Checking a record".
invalid, or wrong-digest evidence and requires the owning repair_task.
Fail-closed, additive, no in-repo caller yet.
runner, signed evidence bound to the candidate) and the physical-display
checks. Left pending per Hive acceptance infrastructure: automate Utah boot and runtime evidence before landing system changes #1614.
Decision taken by hive: the #1622 reviewer asked whether a gate with no caller
should land now or wait for a wired consumer. Per the owner rule (scope is
mine to decide), this is the simplest additive, backward-compatible option, so
it lands now; the consumer/runner is the next step.
🐝 Hive Agent:
contributor| SHA:e3d4cbbfeat(acceptance): land admit-evidence admission gate (projectbluefin#…
Danathar commentedon Nov 17, 2026
Closing. What this issue asked for now exists, mostly in Utah, and the one remaining automatable gap is tracked in Utah, where the code lives. Summary as of 11-17, @hanthor:
Acceptance items
docs/skills/acceptance-evidence.md).build.ymlstartspost-testing-e2e.ymlfor every testing build. That run verifies the provenance of each exact digest, does a LUKS install and boot of all four flavors, builds the production ISO, and runs the GNOME tests.promote-to-testingneeds the LUKS and ISO jobs, andscripts/promotion-ratchet.pywrites thee2e/post-testingstatuses the release gate reads. It never moves:testingbackwards. Common'sscripts/admit-evidence.sh(feat(acceptance): land admit-evidence admission gate (#1614) #1652) has no caller, because Utah gates this its own way.just lifecycle-test). Wiring it in, and gating promotion on it, is now ci: run the bootc upgrade/rollback lifecycle harness in post-testing E2E and gate promotion on it utah#1323.If common's unused
admit-evidence.shshould go, that's a separate cleanup.