Add v13 cost-safety release-note chapter

Context

The current v13 release note artifact is now strong on the Brain / Dream / Memory / Grid arcs, but the late 2026-06-08 cost-safety incident is not yet represented as a first-class v13 story.

V-B-A evidence before filing:

• Live latest-open sweep: checked the latest 20 open issues on 2026-06-08 with number, title, author, labels, and URL. Adjacent open work exists under #12740, including #12743, #12746, #12748, #12768, plus human-gated #12776 / #12780; no equivalent release-note chapter ticket was present.
• Targeted live search for cost safety release notes returned closed #12749 only, which is the /blob/main link guard follow-up, not the cost-safety narrative.
• Local mirror sweep over resources/content/issues and resources/content/discussions for cost-safety, Gemini incident, billing, remote model, local-first, and miniSummary backfill found related incident/provider material but no release-note chapter ticket.
• Semantic fallback sweep used query_documents(type='ticket') because ask_knowledge_base itself attempted stale Gemini synthesis in this harness. The top result was historical #12573; no equivalent release-note chapter ticket surfaced.
• Current release note inspection found resources/content/release-notes/v13.0.0.md covers mini-summary backfill and sync_all warm-cache repair, but has no cost-safety / Gemini-spend chapter.
• Current source shows ai/services/knowledge-base/SearchService.mjs already provider-routes synthesis through buildChatModel, so the observed ask_knowledge_base Gemini call is stale-runtime evidence, not proof that #12741 failed.

The Problem

The cost-safety incident is a release-quality story, not just cleanup bookkeeping. It demonstrates the same MX loop the v13 release is trying to explain:

• A real remote Gemini spend spike exposed that cheap Agent OS maintenance paths can become paid remote workloads when defaults, inherited keys, and maintenance loops line up badly.
• The swarm split containment into small leaves: provider routing, local-first defaults, harness key pass-through, test opt-in gates, backfill pacing, runtime freshness, and billing reconciliation.
• That sequence is exactly friction -> gold: an expensive operator-visible failure became repo-owned safety substrate.

If the release note omits this, v13 undersells one of its most important late hardening arcs. If it lands too early, it risks overclaiming while #12740 leaves and billing reconciliation are still moving.

The Architectural Reality

• Release artifact: resources/content/release-notes/v13.0.0.md.
• Parent release epic: #12696.
• Cost-safety incident epic: #12740.
• Relevant current leaves include #12743 spend reconciliation, #12746 miniSummary backfill pacing, #12748 local chat priority, #12768 lean Memory Core healthcheck, #12776 shared runtime-freshness helper, and #12780 sync CLI dev-branch guard.
• Already-closed / merged adjacent leaves include #12741 KB ask provider routing, #12742 local-first summary provider default, #12744 harness key pass-through removal, #12747 live-AI test opt-in inversion, and #12752 mailbox degradation tolerance.
• ADR 0019 remains the provider/config source-of-authority boundary: release prose must not describe parallel provider aliases as if they were the intended architecture.

The Fix

After the relevant cost-safety leaves are terminal, update the v13 release note with a bounded cost-safety chapter or war-story section.

Expected shape:

• Add a Symptom -> Investigation -> Culprit -> Fix story for the Gemini cost incident and miniSummary backfill trigger.
• Tie the story to local-first provider defaults, explicit remote opt-in, harness key hygiene, runtime freshness, and backfill pacing.
• Add representative anchors for the merged PRs only; do not cite open or human-gated PRs as shipped evidence.
• Keep #12743 as the authoritative billing/spend reconciliation surface and avoid inventing final token/SKU math if Cloud Billing data is still unavailable.
• Preserve the current release-note stance that v13 is a self-evolving organism: this chapter should show MX turning a cost failure into substrate, not read like a generic incident postmortem.

Contract Ledger Matrix

Decision Record impact

aligned-with ADR 0019. This is release documentation, but it must preserve the AiConfig provider SSOT framing and not imply a successor provider model.

Acceptance Criteria

• The v13 release note contains a cost-safety chapter, war-story block, or clearly named subsection explaining the 2026-06-08 Gemini spend incident as MX friction converted into substrate.
• The prose cites only merged PRs as shipped evidence and names any still-open leaves as residual gates rather than completed work.
• The chapter covers local-first provider defaults, explicit remote opt-in, inherited key hygiene, KB ask/provider routing, miniSummary/backfill pacing, and runtime freshness where those surfaces have merged evidence.
• The chapter keeps billing/SKU/token math scoped to #12743; if the external Cloud Billing export is still missing, the release note says that instead of inventing final numbers.
• The curated changelog digest gains representative cost-safety bullets if the chapter lands outside the existing Memory/Core or Provider sections.
• The release-cut checklist includes any remaining cost-safety gate that must be verified before isDraft: false.

Out of Scope

• Implementing or reviewing the cost-safety code leaves themselves.
• Closing #12740 or #12743.
• Changing provider defaults, scheduling behavior, or MCP healthcheck code.
• Turning the release note into a full incident report with private billing-console details.
• Creating the future release-notes skill (#10321).

Avoided Traps

• Adding the story before the implementation leaves land, then having public prose drift from shipped evidence.
• Treating key deletion as the release story. Key deletion stopped immediate damage; the v13 story is the substrate repair.
• Letting ask_knowledge_base stale-runtime behavior become a false public claim against the current merged source.
• Burying the cost incident in a bullet list; it deserves narrative treatment because it proves the MX loop under real pressure.

Related

Parent: #12696 Blocked by: #12740 Related: #12743, #12746, #12748, #12768, #12741, #12742, #12744, #12747, #12752, #12776, #12780

Origin Session ID: e8f07ef9-ef7e-4815-8ff4-7abe13720621 Retrieval Hint: v13 release notes cost-safety Gemini incident miniSummary backfill local-first provider defaults runtime freshness #12740 #12743