Repository navigation
Conversation
… kicks while a contributor relay covers the role A contributor relay acting as an agent role used to work in parallel with the hive's resident agent: the governor never knew the relay was there, so the hive's own inference budget kept being spent on a lane a contributor was already covering. Standby (RFC hivecommons#7629) is the opposite order on purpose and only fires once a lane is paused for budget. agents.<name>.contributor_first: true makes the connected relay the lane's primary worker. While at least one relay is acting as the role (fresh heartbeat, not failure-paused) the governor withholds the resident agent's kicks at the same four gates that honour an exhausted budget: cadence, continuous re-kick, CEL event kick and crash-recovery resume kick. Once no relay has covered the role for contributor_first_grace (default 10m) the resident agent resumes its normal cadence. Budget pause and standby are unchanged: a covered lane is never reported as paused for budget, and the budget gate is checked first. The load fails when the flag names a role contributors may not act as, so the setting can never sit inert. The Governor section shows each covered lane and who covers it; /api/status carries governor.covered_lanes. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WcWm7sQjJ5QDk9LVqL115w Signed-off-by: Douglas Baggett <doug.baggett@gmail.com>
|
[APPROVALNOTIFIER] This PR is NOT APPROVED This pull-request has been approved by: The full list of commands accepted by this bot can be found here. DetailsNeeds approval from an approver in each of these files:Approvers can indicate their approval by writing |
…tus-struct addition FrontendLaneCoverage and the covered_lanes field moved the route registrations in pkg/dashboard/server.go down by fourteen lines, so the pkg/dashboard/server.go:NNNN citations in src/docs/api-reference.md and src/docs/api-reference/*.md drifted. Rewritten with `bash src/scripts/check-api-reference-citations.sh --fix` (32 citations); the checker is clean again. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WcWm7sQjJ5QDk9LVqL115w Signed-off-by: Douglas Baggett <doug.baggett@gmail.com>
There was a problem hiding this comment.
Read the governor, hub and eval-cycle code at 1b7d95c plus the existing admission path (contribute_ws.go:2936-2950, contribute_agent_roles.go:61-96). The core is sound: the gate sits at the same four kick paths as the budget gate, SuppressedLanes/LanePauseReasons really do stay budget-only, coverage is only ever derived from c.role, which is set exclusively after roleClaimAllowed (delegatable role, tier ≥ trusted, grants) or by the owner's SetAssignedAgentRole (contribute_ws.go:2341) — so no relay can self-assign its way into withholding a lane. go build of the four touched packages is clean.
One verified gap — the operator's "Suspend contributions" switch idles the whole lane.
src/pkg/dashboard/contribute_select.go:484-489: whenhub.contribute_suspendedis true, every connected relay is refused work withcontribution_suspended. The relays stay connected and keep heartbeating.src/pkg/dashboard/contribute_coverage.go:44-57:RoleCoveragecounts a relay as covering if its role is set, its heartbeat is fresh and it is not in a failure-streak pause. It does not consultHub.ContributeSuspended.- Consequence: an operator who pauses contributions on a
contributor_firstlane gets neither worker — relays get no tasks, and the governor keeps withholding the resident agent for as long as the relays stay connected. The header comment atsrc/pkg/governor/contributor_first.go:23("connected, non-stale, non-suspended relay") promises exactly the check that is missing. Suggested fix: returnnilfromRoleCoverage(or fromServer.ContributorRoleCoverage) whendeps.Config.Hub.ContributeSuspendedis set, plus a test next to the stale/failure-paused cases incontribute_coverage_test.go.
Severity: medium — opt-in flag, operator-visible via the new tile, but it silently converts a "pause contributors" action into "pause the lane".
Body matches the diff otherwise (the four gates, grace, validation, covered_lanes). No other findings.
— hive: agent=reviewer backend=copilot model=claude-fable-5.1 copilot=1.0.88
|
@Danathar this PR is back in your court ( Review asks:
Reply here or push a new head to this branch (same sign-off); CI and the review swarm re-run automatically and the PR returns to the normal path. Queued in your relay fix queue (1 waiting): fixes are handed out before new work as soon as you have capacity. |
Open PR overlap checkBase branch:
Conflicts are checked with every PR merged onto the current tip of No conflicting open PRs or same-file overlaps were found. |
|
Hive status — updated in place; edits do not notify. Full history is in the Activity Monitor. Review findings from Head a524bf8 resolves both findings from the previous hive review; one new, verified CI break remains — a one-line test-harness fix. Previously blocking (replicas never withheld) — fixed. New, blocking: Also verified: the hub only reports a lane as covered from Housekeeping: the PR is Fork PR — comment only; nothing pushed. Review findings from Read the final delta 1. --- a/src/pkg/dashboard/standby_tile_visible.test.js
+++ b/src/pkg/dashboard/standby_tile_visible.test.js
-const src = ['standbyPauseFactsHtml', 'renderGovernor'].map(grab).join('\n');
+const src = ['standbyPauseFactsHtml', 'contributorFirstFactsHtml', 'renderGovernor'].map(grab).join('\n');2. The The replica fix in Fork PR — comment only; nothing pushed. Since Review findings from Correctness re-review of a524bf8 (delta since f24a477). Previous findings: replicas never withheld — fixed in the governor ( One new in-scope problem, in the fix itself: The config loader still rejects any contributor-first lane that has replicas or review heads, so the new governor replica support is unreachable. Suggested fix — validate the flag once, on the base entry (the lane), and skip derived agents: and add a config test that sets |
…; cover the crux grace helper Review finding on hivecommons#16469: with hub.contribute_suspended set, the hub refuses work to every relay while their sockets stay up, so RoleCoverage kept reporting the connected relays as covering the lane and the governor kept withholding the resident agent. Pausing contributors became pausing the lane. RoleCoverage now returns no coverage while contributions are suspended, mirroring the task-selection gate, with a test next to the stale and failure-paused cases. The crux/config coverage floor (90%) tripped at 89.9% because the new ContributorFirstGraceOrDefault helper was only exercised from pkg/config; it now has its own table test in crux/config (90.3%). Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WcWm7sQjJ5QDk9LVqL115w Signed-off-by: Douglas Baggett <doug.baggett@gmail.com>
|
Thanks, the suspend-switch gap was real. Addressed in f24a477:
PR body updated to match. |
…ir base lane; case-insensitive agent lookup Second review finding on hivecommons#16469: replicas (quality-2) and review heads (quality-fast-1) copy the base agent's contributor_first flag, but the hub reports coverage under the BASE role a relay acts as while the gate looked the record up by each agent's own name. Only replica 1 was ever withheld, and the tile still claimed the lane was covered. The gate now resolves every agent to its coverage lane through ReplicaOf, so a covered lane withholds all of its derived agents, and covered_lanes publishes one entry per lane rather than per replica. Third finding: SetContributorCoverage looked the lane's grace up by the lower-cased role while agents are keyed by their raw YAML name, so a mixed-case agent name expired on the default grace instead of its own. Agent lookups are now case-insensitive, and the loader additionally rejects a mixed-case contributor_first name, because relays claim roles in lower case and such an agent could never be covered at all. Tests: a replica/review-head lane test and a mixed-case grace test in pkg/governor, and the new load-error case in pkg/config. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WcWm7sQjJ5QDk9LVqL115w Signed-off-by: Douglas Baggett <doug.baggett@gmail.com>
There was a problem hiding this comment.
Delta re-review (correctness) of 1b7d95c → f24a477: the suspend-switch gate is correct — RoleCoverage short-circuits on contributionsSuspended() (contribute_coverage.go:34, 92-95), which mirrors the existing unsynchronized read in contribute_select.go:346-347, and TestRoleCoverage_SuspendedContributionsCoverNothing covers both the suspended and resumed paths. No new defects in the delta.
Still present (blocking): replicas of a contributor-first lane are never withheld. The delta does not touch the gate. pkg/config/config.go:361 still does derived := agent (so quality-2, quality-3 inherit ContributorFirst, with derived.ReplicaOf = name at :366), while the hub keys coverage by role ("quality", contribute_coverage.go:48) and the gate keys by the agent's own name:
rec, ok := g.coverage[strings.ToLower(strings.TrimSpace(agentName))] // pkg/governor/contributor_first.go:160So only replica 1 is withheld; the others are kicked on cadence while CoveredLanes reports the lane as covered. Suggested fix — resolve replicas to their base lane in the three lookups:
--- a/src/pkg/governor/contributor_first.go
+++ b/src/pkg/governor/contributor_first.go
@@ func (g *Governor) contributorCoveredLocked(agentName string) (LaneCoverage, bool) {
ac, ok := g.agents[agentName]
if !ok || !ac.ContributorFirst {
return LaneCoverage{}, false
}
- rec, ok := g.coverage[strings.ToLower(strings.TrimSpace(agentName))]
+ rec, ok := g.coverage[coverageKey(agentName, ac)]
if !ok {
return LaneCoverage{}, false
}
@@
+// coverageKey is the role the hub reports coverage under: the base lane for a
+// replica (quality-2 → quality), the agent's own name otherwise.
+func coverageKey(agentName string, ac config.AgentConfig) string {
+ if ac.ReplicaOf != "" {
+ return strings.ToLower(strings.TrimSpace(ac.ReplicaOf))
+ }
+ return strings.ToLower(strings.TrimSpace(agentName))
+}contributorFirstLocked/contributorFirstGraceLocked (:196-206) take the lower-cased role, so they need to consider any agent whose coverageKey matches (base lane or its replicas) rather than g.agents[role] alone — otherwise the "covering a contributor-first lane" log/grace still only sees the base agent. A replicas: 2 case in contributor_first_test.go (currently no replica coverage, contributor_first_test.go:11-28) would pin it.
Still present (low): mixed-case agent name expires its record after the default grace — SetContributorCoverage looks up g.agents[role] with role lower-cased (contributor_first.go:103, 141, 201-205) while g.agents is keyed by the raw YAML name; the same coverageKey normalisation, or rejecting mixed-case names in validateContributorFirst, closes it.
Fork PR — comment only; nothing pushed.
— hive: agent=adjudicator backend=copilot model=claude-fable-5.1 copilot=1.0.88
|
Both of reviewer-2's findings addressed in a524bf8:
PR body updated. The overlap with #16378 is only the api-reference citation files, which the citation fixer regenerates; whichever lands second re-runs |
|
@Danathar this PR is back in your court ( Failing checks:
Reply here or push a new head to this branch (same sign-off); CI and the review swarm re-run automatically and the PR returns to the normal path. Queued in your relay fix queue (1 waiting): fixes are handed out before new work as soon as you have capacity. |
|
Cause: Smallest fix — add it to the stub map so the standby test stays focused on the standby tile: --- a/src/pkg/dashboard/standby_tile_visible.test.js
+++ b/src/pkg/dashboard/standby_tile_visible.test.js
@@ -34,6 +34,7 @@ const stubs = {
fleetRunStatHtml: () => '',
+ contributorFirstFactsHtml: () => '',
formatGovernorNextRun: () => '',(Alternative, if you'd rather exercise the real tile alongside standby: Fork branch, so I can't push this; flagging for you to apply with your next push. 🐝 Hive Agent: — hive: agent=scanner backend=copilot model=claude-fable-5.1 copilot=1.0.88 |
|
@clubanderson asked the hive to adopt this pull request now, without waiting for its author's hive. The hive opens its own PR carrying the author's commits and sign-off unchanged on the pass now running; this PR closes when that copy merges. |
|
Supersede failed: the hive could not open a fresh PR with these commits. Merging Rebase this branch onto |
1 similar comment
|
Supersede failed: the hive could not open a fresh PR with these commits. Merging Rebase this branch onto |
…; cover the crux grace helper Review finding on #16469: with hub.contribute_suspended set, the hub refuses work to every relay while their sockets stay up, so RoleCoverage kept reporting the connected relays as covering the lane and the governor kept withholding the resident agent. Pausing contributors became pausing the lane. RoleCoverage now returns no coverage while contributions are suspended, mirroring the task-selection gate, with a test next to the stale and failure-paused cases. The crux/config coverage floor (90%) tripped at 89.9% because the new ContributorFirstGraceOrDefault helper was only exercised from pkg/config; it now has its own table test in crux/config (90.3%). Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WcWm7sQjJ5QDk9LVqL115w Signed-off-by: Douglas Baggett <doug.baggett@gmail.com>
…ir base lane; case-insensitive agent lookup Second review finding on #16469: replicas (quality-2) and review heads (quality-fast-1) copy the base agent's contributor_first flag, but the hub reports coverage under the BASE role a relay acts as while the gate looked the record up by each agent's own name. Only replica 1 was ever withheld, and the tile still claimed the lane was covered. The gate now resolves every agent to its coverage lane through ReplicaOf, so a covered lane withholds all of its derived agents, and covered_lanes publishes one entry per lane rather than per replica. Third finding: SetContributorCoverage looked the lane's grace up by the lower-cased role while agents are keyed by their raw YAML name, so a mixed-case agent name expired on the default grace instead of its own. Agent lookups are now case-insensitive, and the loader additionally rejects a mixed-case contributor_first name, because relays claim roles in lower case and such an agent could never be covered at all. Tests: a replica/review-head lane test and a mixed-case grace test in pkg/governor, and the new load-error case in pkg/config. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WcWm7sQjJ5QDk9LVqL115w Signed-off-by: Douglas Baggett <doug.baggett@gmail.com>
…g a role withholds the resident agent's kicks (#16518) * ✨ [v6] feature: contributor-first lanes withhold the resident agent's kicks while a contributor relay covers the role A contributor relay acting as an agent role used to work in parallel with the hive's resident agent: the governor never knew the relay was there, so the hive's own inference budget kept being spent on a lane a contributor was already covering. Standby (RFC #7629) is the opposite order on purpose and only fires once a lane is paused for budget. agents.<name>.contributor_first: true makes the connected relay the lane's primary worker. While at least one relay is acting as the role (fresh heartbeat, not failure-paused) the governor withholds the resident agent's kicks at the same four gates that honour an exhausted budget: cadence, continuous re-kick, CEL event kick and crash-recovery resume kick. Once no relay has covered the role for contributor_first_grace (default 10m) the resident agent resumes its normal cadence. Budget pause and standby are unchanged: a covered lane is never reported as paused for budget, and the budget gate is checked first. The load fails when the flag names a role contributors may not act as, so the setting can never sit inert. The Governor section shows each covered lane and who covers it; /api/status carries governor.covered_lanes. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WcWm7sQjJ5QDk9LVqL115w Signed-off-by: Douglas Baggett <doug.baggett@gmail.com> * 🐛 fix(contributor-first): a suspended contribute queue covers nothing; cover the crux grace helper Review finding on #16469: with hub.contribute_suspended set, the hub refuses work to every relay while their sockets stay up, so RoleCoverage kept reporting the connected relays as covering the lane and the governor kept withholding the resident agent. Pausing contributors became pausing the lane. RoleCoverage now returns no coverage while contributions are suspended, mirroring the task-selection gate, with a test next to the stale and failure-paused cases. The crux/config coverage floor (90%) tripped at 89.9% because the new ContributorFirstGraceOrDefault helper was only exercised from pkg/config; it now has its own table test in crux/config (90.3%). Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WcWm7sQjJ5QDk9LVqL115w Signed-off-by: Douglas Baggett <doug.baggett@gmail.com> * 🐛 fix(contributor-first): withhold replicas and review heads with their base lane; case-insensitive agent lookup Second review finding on #16469: replicas (quality-2) and review heads (quality-fast-1) copy the base agent's contributor_first flag, but the hub reports coverage under the BASE role a relay acts as while the gate looked the record up by each agent's own name. Only replica 1 was ever withheld, and the tile still claimed the lane was covered. The gate now resolves every agent to its coverage lane through ReplicaOf, so a covered lane withholds all of its derived agents, and covered_lanes publishes one entry per lane rather than per replica. Third finding: SetContributorCoverage looked the lane's grace up by the lower-cased role while agents are keyed by their raw YAML name, so a mixed-case agent name expired on the default grace instead of its own. Agent lookups are now case-insensitive, and the loader additionally rejects a mixed-case contributor_first name, because relays claim roles in lower case and such an agent could never be covered at all. Tests: a replica/review-head lane test and a mixed-case grace test in pkg/governor, and the new load-error case in pkg/config. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01WcWm7sQjJ5QDk9LVqL115w Signed-off-by: Douglas Baggett <doug.baggett@gmail.com> * test: stub contributorFirstFactsHtml in standby tile isolation test Hive-Fix-Attempt: 1/6 Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Signed-off-by: architect <280983584+hivecommons-hive[bot]@users.noreply.github.com> * 📖 docs: resync api-reference source citations after rebase onto v6 Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Signed-off-by: scanner <280983584+hivecommons-hive[bot]@users.noreply.github.com> * chore: retrigger CI after proxy.golang.org stream flake in build-hub (linux/amd64) Signed-off-by: scanner <280983584+hivecommons-hive[bot]@users.noreply.github.com> --------- Signed-off-by: Douglas Baggett <doug.baggett@gmail.com> Signed-off-by: architect <280983584+hivecommons-hive[bot]@users.noreply.github.com> Signed-off-by: scanner <280983584+hivecommons-hive[bot]@users.noreply.github.com> Co-authored-by: Douglas Baggett <doug.baggett@gmail.com> Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com> Co-authored-by: architect <280983584+hivecommons-hive[bot]@users.noreply.github.com> Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Executive summary
A contributor who runs a relay "acting as" one of the hive's agent roles can now be that lane's primary worker. Set
contributor_first: trueon the agent and, while a healthy contributor relay is connected for that role, the governor stops kicking the hive's own agent for that lane. The hive's inference budget is no longer spent on work a contributor is already covering. When the contributor goes away, the resident agent resumes on its normal cadence after a short grace (default 10 minutes). Hives that do not set the flag see no change, and budget pause and standby keep their current meaning.Target: v6.
What this changes
Today a contributor relay with an agent role works in parallel with the resident agent: the governor does not know the relay is there and keeps kicking the resident agent, so operator budget is still spent on that lane. Standby (RFC #7629) is the opposite order on purpose: it only hands out work after a lane is paused for budget.
With this PR, per agent in
hive.yaml:contributor_first: true— while at least one connected contributor relay is acting as the role (fresh heartbeat, not in a failure-streak pause), the governor withholds the resident agent's kicks. Cadence kicks, continuous re-kicks, CEL event kicks and crash-recovery resume kicks all respect it: the same four gates that respect an exhausted budget.contributor_first_grace(default10m) — how long kicks stay withheld after the last relay disappears, so a routine relay reconnect does not flap the resident agent on and off./api/statuscarries the same facts asgovernor.covered_lanes.quality-2) and review heads (quality-fast-1) of a contributor-first lane are withheld together with their base agent: coverage is resolved through the base lane a relay acts as, and the tile shows one entry per lane. (Added after the second hive reviewer's finding, fourth commit.)hub.contribute_delegatable_rolesand not one of the default scanner / quality / outreach), onsupervisor, on an agent whose name is not lower-case (relays claim roles in lower case, so it could never be covered), or with a negative grace. A contributor-first lane nobody may claim would sit inert while the operator believes contributors are taking the load.What does not change
suppressed_lanesandlane_pause_reasonsstay budget-only, so standby dispatch keeps its meaning. The budget gate is checked first, so a lane that is both out of budget and covered reads as paused for budget exactly as before.hub.contribute_suspended) still pauses only contributors. While it is on, no relay counts as covering, so the resident agent keeps working; pausing contributors never pauses the lane. (Added after the hive reviewer's finding, third commit.)hive.yamlround-trips unchanged.How it works
Once per eval tick, before the governor evaluates, the eval cycle asks the contribute hub which roles connected relays are acting as (the effective role after the owner's "Acting as" assignment, a heartbeat inside the WebSocket timeout, no failure-streak pause) and hands the governor a role → logins map. The governor remembers the last sighting per role and withholds while a relay is present or the grace is still running. Coverage transitions (started, cover ended, grace elapsed) are logged once each, not every tick.
Files
crux/config/agent_config.go,crux/config/contributor_first.go: the two keys and the default grace.pkg/config/contributor_first.go,pkg/config/validate.go: load-time validation.pkg/governor/contributor_first.go,pkg/governor/governor.go: coverage memory, the gate at the four kick paths,covered_lanesin state and eval history.pkg/dashboard/contribute_coverage.go: the hub's view of who is acting as which role;server.goandstatus_builder.go:covered_laneson the status payload.cmd/hive/contributor_first.go,cmd/hive/main_eval_cycle.go,cmd/hive/eval_cycle_seams.go: the per-tick feed and the resume-kick log line.pkg/dashboard/static-src/js/sidebar.js,governor-view.js,agent-detail.js, and the regeneratedstatic/index.html: the tile.docs/agent-configuration.md,docs/contributor-trust-and-roles.md,hive.yaml.example,pkg/dashboard/settings_parity_allowlist.txt, and a changelog fragment.docs/api-reference.mdanddocs/api-reference/*.md: line-number citations intoserver.goresynced with the repo'scheck-api-reference-citations.sh --fix(second commit; the new status struct moved the route registrations down).Related issues
None. This is an operator-requested feature; there is no tracking issue.
Tests
New tests:
pkg/governor/contributor_first_test.go: a covered lane is not due while an uncovered one still is; the flag off ignores coverage; grace then resume on a fixed clock, including a relay returning inside the grace; default grace; the budget gate still wins (the lane stays budget-suppressed); CEL and resume-kick gates; the continuous-mode blocker; report normalisation;GetStatereturns a copy; replicas and review heads follow their base lane and resume with it; a mixed-case agent name keeps its own grace.pkg/config/contributor_first_test.go: delegatable, undelegatable, supervisor, mixed-case-name and negative-grace cases; wiring intovalidate(); default grace.pkg/dashboard/contribute_coverage_test.go: stale, roleless, failure-paused and profile-less connections are excluded; role names normalised; nil hub and server; a suspended contribute queue reports no coverage; status rendering.crux/config/contributor_first_test.go: the grace default helper, so the crux/config coverage floor (90%) holds; it had dipped to 89.9% because the helper was only exercised frompkg/config.cmd/hive/contributor_first_test.go: no dashboard means no coverage.dashboard/test/contributor-first-tile.test.js: the tile renders nothing when empty, names the lane and contributors in both variants, grace wording, sorted and escaped output.Results on this machine:
go test ./pkg/governor ./crux/config ./pkg/dashboard ./cmd/hive(full packages,-count=1): all pass. The first full dashboard run failedTestSettingsDialogReferencesEveryHiveYAMLLeaf, which requires every newhive.yamlkey to have a Settings control or an allowlist entry. The two keys arehive.yaml-only in this PR, so they are allowlisted next to the standby keys under the same TODO.golangci-lint runon the five packages: 0 issues.go test ./pkg/config: all new tests pass. The one failure is the pre-existingTestGateFailsClosedWithoutIP6Tables, which expects ip6tables to be missing and this host has it installed (the same note as on 🐛 unpark: keep the "What to reply" commands as the last comment, in a green box #16425; the package is otherwise untouched).go vetis clean on the five packages;gofmtis clean on the touched files.node --test: 42 of 45 dashboard JS tests pass, including the new file. The 3 failures areCannot find module 'ws'/'express'incontribute.test.js,contributors-authz.test.jsandnous-authz.test.js: dependencies are not installed on this machine, and those files are untouched.make dashboard-genregeneratedindex.html;TestIndexHTMLIsGeneratedpasses.Not run: a live hive with a connected contributor relay. To verify on one: set
contributor_first: trueonquality, connect a relay acting as quality, confirm the contributor-first tile appears and the resident quality agent gets no kicks; disconnect the relay and confirm the resident agent resumes after the grace.🤖 Generated with Claude Code
https://claude.ai/code/session_01WcWm7sQjJ5QDk9LVqL115w