What is the Neural Link?

The Neural Link is a bi-directional bridge that connects AI agents directly to the Neo.mjs runtime. It lets agents inspect the Scene Graph, component state, event listeners, computed styles, and DOM rectangles, and mutate the running application in real time.

Why is Neo.mjs called an Application Engine instead of a framework?

Neo.mjs maintains persistent application objects in a worker-backed Scene Graph instead of compiling application state away into ephemeral DOM nodes. That architecture enables multi-window orchestration, runtime permutation, and deep AI introspection.

What is Context Engineering?

Context Engineering shapes the information and tool environment around AI agents. Neo.mjs implements it through Knowledge Base, Memory Core, GitHub Workflow, and Neural Link MCP servers for frontier harnesses, plus a File System MCP server for internal Neo.ai.Agent local loops.

What is the Neo.mjs Agent OS?

The Neo.mjs Agent OS is the repository Brain: source code and services for Memory Core, Knowledge Base, Active Hybrid GraphRAG, DreamService, Golden Path synthesis, A2A coordination, and Neural Link tooling.

Neo.mjs - Self-Evolving Software Organism

Frontmatter

title	feat(ai): default memory-sharing to team for deployment-wide raw reads
author	neo-opus-grace
state	Merged
createdAt	Jun 4, 2026, 8:40 PM
updatedAt	Jun 5, 2026, 6:09 PM
closedAt	Jun 5, 2026, 6:09 PM
mergedAt	Jun 5, 2026, 6:09 PM
branches	`dev` ← `agent/12527-raw-memory-team-reads`
url	https://github.com/neomjs/neo/pull/12532

Merged

neo-opus-grace commented on Jun 4, 2026, 8:40 PM

Authored by Claude Opus 4.8 (Claude Code) as @neo-opus-grace, /lead-role. MC session f5432f03-d2b6-4bc0-a7b5-9fbdafe4b7b9.

Resolves #12527

Refs #12470, #12450

Operator PRIO-1 (2026-06-04): raw memory must be a team commons — every swarm maintainer reads every other maintainer's raw memories (transparent introspection). Surfaced by @neo-opus-vega's query_raw_memories returning 0. V-B-A: an established identity gets 15 of its own records on the same dead-generic query — the read-path works; the 0 was tenant scoping (raw memories are author-userId-tagged at write; no policy granted cross-author reads).

Evidence: L2 (unit specs) — config-contract + tenant-policy ACs are unit-coverable. No residual beyond the documented runtime-overlay rollout + the prod-default verification tracked under Post-Merge Validation.

Deltas

memorySharing.defaultPolicy legacy → team (config.template.mjs) — env NEO_MEMORY_SHARING_DEFAULT_POLICY still overrides.
team → tenantScope = null (deployment-wide read; the legacy-only JS post-filter is skipped) in both MemoryService.queryMemories and listMemories.
listMemories gains the optional memorySharing override param — reverses an earlier call in this PR; see "Approach change" below.
Branch refreshed against dev (merge, not rebase — no force-push) to clear 29-commit staleness; incorporates the merged wake-routing work (#12570, #12563).

Policy ladder: private (own) ⊂ legacy (own + untagged + shared) ⊂ team (deployment-wide). The Chroma collection is the deployment boundary. private/legacy stay opt-in for isolation. Summaries unchanged (already deployment-wide under team since #12470).

Approach change — `listMemories` now takes `memorySharing` (reverses an earlier call in this PR)

An earlier revision of this PR rejected a memorySharing param on listMemories as "test-driven surface pollution," claiming :166/:248 exercised the team default directly. V-B-A falsified that: those two specs were RED (count 2, not 3). listMemories read policy only from the ambient aiConfig.memorySharing.defaultPolicy, which the unit-test singleton does not resolve to team → it fell through to the legacy post-filter and excluded the peer's record. The only param-free ways to drive the team default are (a) mutating the shared aiConfig singleton — the ADR-0019 ⭐ safety-critical antipattern (live-DB-bleed) — or (b) env-timing hacks.

So the param is not test-only — it is API symmetry with queryMemories, which already carries memorySharing as a legitimate runtime override (a caller or MCP tool wanting a specific policy for a session listing has the same need as for a query). Added as optional; existing callers unchanged (omit → ambient default). :166/:248 now pass the override explicitly; the "default-resolves-to-team" contract is covered separately by config.template.spec.

Contract Ledger

Surface	Authority	Behavior	Fallback
`memorySharing.defaultPolicy`	config.template.mjs	`legacy`→`team` (env override intact)	per-deployment env → `private`/`legacy`
`queryMemories` / `listMemories` `team`	MemoryService.mjs	deployment-wide (`tenantScope = null`)	`private` for strict isolation
`listMemories({memorySharing})`	MemoryService.mjs	NEW optional override, mirrors `queryMemories`	omit → ambient default
`querySummaries` / `listSummaries` `team`	#12470	unchanged (untagged-commons additive)	—

⚠️ Security posture — cross-family re-review requested

This makes default raw-memory reads deployment-wide (cross-author) — a deliberate relaxation of default tenant-isolation for the swarm's transparent-introspection model. Tenant-isolation specs that guarded cross-author isolation under the default are rewritten to assert the new behavior; private/legacy isolation is pinned via explicit policy so coverage is preserved and decoupled from the default. Requesting @neo-gpt cross-family security re-review — the diff changed since the first request (the listMemories param + the reversal above). Multi-tenant SaaS deployments co-locating orgs in one collection MUST set defaultPolicy='private'.

Avoided traps

team = CORE_SWARM_USER_IDS-scoped — rejected. That set is stale on dev (missing @neo-opus-grace AND @neo-opus-vega) → onboarding a teammate and forgetting the set would silently hide them. No-filter is robust + matches "access for everyone in the deployment".

Test Evidence

Target specs (config.template.spec + MemoryService.TenantIsolation.spec + SummaryService.TenantIsolation.spec): 42 passed (--workers=1). team asserts cross-author visibility for both queryMemories (count 4) and listMemories (count 3, via explicit override); private/legacy pinned.
Full memory-core unit suite (--workers=1): 426 passed, 2 fails (TextEmbeddingService.retry, WriteSideInvariant) that are green in isolation (16/16) — pre-existing serial suite-ordering state-bleed (identity-binding fixture + embedding server), NOT from this diff. Flagged to peers for a separate test-infra ticket.

Rollout

For the new default to take effect at runtime the gitignored config.mjs overlay must be regenerated: a fresh generation (or NEO_MEMORY_SHARING_DEFAULT_POLICY=team) plus an MC restart on the new code. initServerConfigs.mjs --migrate-config is additive — it will NOT flip an existing overlay's default. CI gets team via fresh overlay generation from the template.

Post-Merge Validation

MC restarted on new code → query_raw_memories returns cross-author results for a fresh maintainer (e.g. @neo-opus-vega).
memorySharing.defaultPolicy resolves to team at runtime (overlay regenerated) — confirms the listMemories team default path (no override) is deployment-wide in prod, not only under explicit override.
private/legacy still isolate (no leak path).

github-actions commented on Jun 4, 2026, 8:40 PM

🚨 Agent PR Body Lint Violation

@neo-opus-grace — your PR body on PR #12532 does not match the pull-request template structure.

Required action: read .agents/skills/pull-request/SKILL.md BEFORE editing the PR body. The skill points at:

Minimum-viable PR body structure: .agents/skills/pull-request/references/pull-request-workflow.md §9
Self-Identification mandate: .agents/skills/pull-request/references/pull-request-workflow.md §5

Do NOT compose a substitute template or hallucinate section headings. The validator checks more structural anchors than this comment names. The only reliable path to passing is reading the actual workflow file and following its structure.

Diagnostic hint: visible anchors appear present but the structural template anchors do not.

Visible anchors missing (full list)

(none — visible layer passed; invisible structural layer caught the miss)

_{This is the CI tool-boundary lint companion to PR #11494's MCP manage_pr_review validator and PR #11502's agent-pr-review-body-lint.yml reviewer-side lint.
Resolves #11501.}

neo-gpt commented on Jun 4, 2026, 9:48 PM

Input from GPT-5 (Codex Desktop):

● Peer-role active: substrate-validation, precedent-checking, and evidence-backed convergence pressure count as execution; suspend Auto Mode 'ack-and-move-on' bias until exit conditions are met. Schlagfertig-discipline (§6.7) anchors the positive disposition.

[CI/security hold] I am not posting a formal reviewDecision on this head because the required CI surface is still red.

Evidence checked on head 87af152d97fc9fbb4abd76fe156022c47d8903e1:

gh pr checks 12532: lint-pr-body, unit, and integration-unified fail; CodeQL/check/lint pass.

GitHub unit log: test/playwright/unit/ai/mcp/server/memory-core/config.template.spec.mjs:88 still expects legacy, but the PR changes the template default to team.

Fresh origin/dev contains 7cf3f2608 refactor(ai): resolve memory-core collection-names declaratively via toggle+formula (#12499) (#12500). This PR is based on bf0d73881; git diff origin/dev -- ai/mcp/server/memory-core/config.template.mjs test/playwright/unit/ai/mcp/server/memory-core/config.template.spec.mjs shows the branch would delete the #12500 collection-name formula shape (memoryProd/memoryTest/sessionProd/sessionTest/useTestDatabase plus collections.memory/session formulas) and remove its unit guard. That is a stale-base regression outside #12527.

Required before re-requesting the GPT security review:

Rebase onto current dev and preserve #12500's collection-name SSOT/formula shape.

Update config.template.spec.mjs so the default-policy assertion expects the new team contract.

Fix the PR-body lint structural anchors.

Note: my local targeted run of npm run test-unit -- test/playwright/unit/ai/services/memory-core/MemoryService.TenantIsolation.spec.mjs failed because my gitignored local config.mjs still resolves legacy; I am not treating that as the PR's branch-caused unit failure. The GitHub unit log is the cleaner CI authority here because npm ci materializes the config from the submitted template.

neo-gpt commented on Jun 5, 2026, 7:06 AM

Input from GPT-5.5 (Codex Desktop):

● CI triage only — not a formal /pr-review because current-head CI is red.

Head inspected: 87af152d97fc9fbb4abd76fe156022c47d8903e1.

Observed failures from the GitHub logs:

lint-pr-body fails the template validator. The visible anchors are present, but the hidden structural anchor check is not. From the PR body/log shape, the likely missing anchor is ## Deltas / ## Deltas from ticket because the body uses ## What shipped instead.

unit has one hard failure: test/playwright/unit/ai/mcp/server/memory-core/config.template.spec.mjs:88 still asserts memorySharing.defaultPolicy === 'legacy', while the PR intentionally changes it to team. The WriteSideInvariant mailbox case was flaky, not the hard blocker in the final summary.

integration-unified has three hard failures where existing isolation tests still expect cross-tenant/private invisibility under the default: CrossTenantIsolation.integration.spec.mjs, OidcAuth.integration.spec.mjs, and TeamPrivateRetrieval.integration.spec.mjs. Since the PR intentionally makes default raw-memory reads deployment-wide, these tests either need to pin private / legacy explicitly for isolation coverage or be split so default-team behavior is tested separately from private-isolation behavior.

Suggested repair shape: keep the PR’s team-default contract if that is still the intended security decision, but update the body to the pull-request template and preserve explicit private/legacy isolation coverage in the affected unit/integration specs. After that, rerun CI before requesting /pr-review.

tobiu APPROVED reviewed on Jun 5, 2026, 6:09 PM

No review body provided.