Land the reliability half of the agent stack audit: durable squad state, queue concurrency honesty and end-to-end tool-event fidelity.
Why: the approved remediation plan's batch 2 fixes F11 (mailbox/handoffs/board persistence, lossy content-keyed decode, swallowed checkpoint/terminal-save errors), F12 (queue claim enforced only the per-agent cap) and F14 (tool events lost call identity and real status from native through squad to the backend host), via units U8, U9 and U11 (split U11a/U11b/U11c across repos). Also restores the batch-1 pins that the parallel #682 merge accidentally reverted, and carries one reviewed repair per defect found by verification (probe frame ordering; a -race exposure in the direct tool path).
Evidence: per-unit agent-per-unit receipts in the task card with SELF verification (all exit=0); full combination make test green on the final pinned combination (verify-combo.json). Independent review lands in the task card before merge.
Impact: three component pins advanced (native cff6935, squad 9edb6ba, backend 9b40595); no deploy, no migration, no release. Functional follow-ups A1-A4 and A2A remain out of scope.
Land the reliability half of the agent stack audit: durable squad state, queue concurrency honesty and end-to-end tool-event fidelity.
Why: the approved remediation plan's batch 2 fixes F11 (mailbox/handoffs/board persistence, lossy content-keyed decode, swallowed checkpoint/terminal-save errors), F12 (queue claim enforced only the per-agent cap) and F14 (tool events lost call identity and real status from native through squad to the backend host), via units U8, U9 and U11 (split U11a/U11b/U11c across repos). Also restores the batch-1 pins that the parallel #682 merge accidentally reverted, and carries one reviewed repair per defect found by verification (probe frame ordering; a -race exposure in the direct tool path).
Evidence: per-unit agent-per-unit receipts in the task card with SELF verification (all exit=0); full combination `make test` green on the final pinned combination (verify-combo.json). Independent review lands in the task card before merge.
Impact: three component pins advanced (native cff6935, squad 9edb6ba, backend 9b40595); no deploy, no migration, no release. Functional follow-ups A1-A4 and A2A remain out of scope.
Refs: docs/superpowers/plans/2026-10-21-agent-stack-audit-remediation-plan.md
Why: establish the authorized task workspace and its initial baseline.
Evidence: preserved precheck receipt in .task/feature-agent-stack-audit-remediation-b2.md; R0=f23f5bb01a8b96ab90686d3a1b07791d7f473de2.
Impact: task metadata only; the feature workspace starts at f23f5bb01a.
Refs: .task/feature-agent-stack-audit-remediation-b2.md
Why: PR #682 (f23f5bb0) merged a stale branch whose tree carried
pre-batch component pins, silently reverting the batch-1 composition
on main; batch 2 (U8/U9/U11) is defined on top of batch 1.
Evidence: restoration transport success (pinrestore-run.json); batch-1
tips pushed on all five component origins.
Impact: five component gitlinks restored to the audited batch-1 tips.
Refs: feature/agent-stack-audit-remediation-b2
Why: carry authorization, the discovered composition regression
(#682 reset batch-1 pins) and its restoration, plus the wave plan.
Evidence: restoration transport success (pinrestore-run.json).
Impact: task metadata only.
Refs: feature/agent-stack-audit-remediation-b2
Why: per-unit audit trail for U8/U9/U11a/U11b/U11c including the
fail-stop redispatches and the frame-order repair.
Evidence: per-unit SELF receipts under /tmp/agent-stack-b2.
Impact: task metadata only.
Refs: feature/agent-stack-audit-remediation-b2
Why: land U8, U9 and U11 across agent-native (cff6935), agent-squad
(9edb6ba) and modules/backend (race-repair tip) per the approved plan.
Evidence: per-unit SELF receipts under /tmp/agent-stack-b2 (all
exit=0); full make test re-run follows on this branch.
Impact: three component gitlinks.
Refs: docs/superpowers/plans/2026-10-21-agent-stack-audit-remediation-plan.md
Why: the backend branch tip 9b40595 additionally pins the nested
agent-squad (and its nested native) so the composed chain carries the
batch-2 event fields end to end.
Evidence: integration transport success (gitlinkb-run.json); full
make test re-run follows on this branch.
Impact: one component gitlink.
Refs: feature/agent-stack-audit-remediation-b2
Why: all batch-2 units implemented, repaired and verified; full make
test green on the final pinned combination.
Evidence: verify-combo.json exit=0; component branches pushed
(native cff6935, squad 9edb6ba, backend 9b40595).
Impact: task metadata only.
Refs: feature/agent-stack-audit-remediation-b2
Why: close the task card inside its own PR so landing it needs no post-merge backfill or separate close PR.
Evidence: .task/feature-agent-stack-audit-remediation-b2.md -> .task/archive/feature-agent-stack-audit-remediation-b2.md; status 已完成（合入前闭合）; PR #689https://git-home.u1680832.nyat.app:30319/weichen/Agent-System-Dev/pulls/689.
Impact: task metadata only; the card is archived within PR #689 and lands with its merge commit.
Refs: task.sh close --self
Why: include the current integration baseline before merge.
Evidence: fetched origin/main=1de7c28c.
Impact: integrate upstream changes without rewriting task history.
Refs: task-sync
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Land the reliability half of the agent stack audit: durable squad state, queue concurrency honesty and end-to-end tool-event fidelity.
Why: the approved remediation plan's batch 2 fixes F11 (mailbox/handoffs/board persistence, lossy content-keyed decode, swallowed checkpoint/terminal-save errors), F12 (queue claim enforced only the per-agent cap) and F14 (tool events lost call identity and real status from native through squad to the backend host), via units U8, U9 and U11 (split U11a/U11b/U11c across repos). Also restores the batch-1 pins that the parallel #682 merge accidentally reverted, and carries one reviewed repair per defect found by verification (probe frame ordering; a -race exposure in the direct tool path).
Evidence: per-unit agent-per-unit receipts in the task card with SELF verification (all exit=0); full combination
make testgreen on the final pinned combination (verify-combo.json). Independent review lands in the task card before merge.Impact: three component pins advanced (native cff6935, squad 9edb6ba, backend 9b40595); no deploy, no migration, no release. Functional follow-ups A1-A4 and A2A remain out of scope.
Refs: docs/superpowers/plans/2026-10-21-agent-stack-audit-remediation-plan.md