chore(harness): reopen the 0.0.6 runtime lane for three public-runtime blockers

[rickylabs]

Summary

Control PR for the narrow reopen of the 0.0.6 runtime / public-surface lane, covering three newly triaged public-runtime blockers. Carries orchestration evidence only — the run dir and nothing else. It is not an implementation umbrella and carries no closing keyword; each issue lands through its own leaf branch and PR against main.

The lane's first pass is closed and unchanged: #1405 → 8ff1bcb8f, #1398 → d7e2b67b2, control #1525 → 0f0b6b6a3.

Scope

• Archetype / area: harness run artifacts (.llm/runs/**)
• Owns fix(fresh/ai): chat stream proxy drops durable State-Protocol query parameters #1457, bug(fresh): DeferComponent is not hydrated, so partial-miss regions never refresh #1459, Fresh Streams browser resolver cannot see Aspire VITE service references #1548 — referenced without closing keywords; the implementation PRs carry those.
• No other issue is adopted.

Grouping decision — three separate PRs

Instructed to group only if technically inseparable. Decided from located surfaces, not titles:

Issue	Surface	Shape
#1457 chat proxy drops State-Protocol query params	packages/fresh/src/runtime/ai/stream-proxy.ts	small, fully specified
#1459 DeferComponent never hydrated	packages/fresh/src/application/defer/{DeferPage,DeferIsland}.tsx, policy.ts	substantive — hydration + client-bundle regression test
#1548 browser resolver cannot see Vite service refs	packages/plugin-streams-core/src/application/stream-url-resolver.ts	known cause, real design choice

No file overlap. #1457 and #1459 both live in packages/fresh but in unrelated subtrees (runtime/ai vs application/defer), so they neither conflict nor share acceptance. #1548 is in a different package, and stream-url-resolver.ts was not touched by #1405 — no interaction with what this lane already landed.

Nothing is technically inseparable, so they stay three independent PRs.

Slices

• S1 Reopen: scope, grouping decision, eval routing recorded — fe9bb409e
• S2 fix(fresh/ai): chat stream proxy drops durable State-Protocol query parameters #1457 implementation + automatic IMPL-EVAL + merge
• S3 bug(fresh): DeferComponent is not hydrated, so partial-miss regions never refresh #1459 plan → PLAN-EVAL → implementation + automatic IMPL-EVAL + merge
• S4 Fresh Streams browser resolver cannot see Aspire VITE service references #1548 plan → PLAN-EVAL → implementation + automatic IMPL-EVAL + merge
• S5 Reopen closeout: cut-trace extended, retrospective addendum

Validation

Diff is .llm/runs/** only — no publishable source, so no package gate applies to this PR. Gate evidence lands on the three implementation PRs.

Harness

• Run dir: .llm/runs/release-0.0.6-features--orchestration/ (extended below a # Reopen divider; the closed first pass is not rewritten)
• Profile: .llm/harness/workflow/milestone-run.md; role skill agent-milestone-orchestrator
• Baseline: origin/main@0f0b6b6a3
• Eval policy: automatic (drift D-4/D-5). PLAN-EVAL by the openhands + status:plan-eval label pair; initial IMPL-EVAL automatically on draft → ready; reruns only by moving away from and back to the relevant status: label. No manual OpenHands dispatch, no local evaluator.
• PLAN-EVAL selected for bug(fresh): DeferComponent is not hydrated, so partial-miss regions never refresh #1459 and Fresh Streams browser resolver cannot see Aspire VITE service references #1548 (decision-heavy); fix(fresh/ai): chat stream proxy drops durable State-Protocol query parameters #1457 goes straight to implementation as fully specified, and its IMPL-EVAL still fires automatically — impl-eval:skip is deliberately not applied.

Drift / Debt

Carried from the first pass, still in force: D-1 sub-agent lane override · D-2 evaluator transport · D-3 IMPL-EVAL waiver for the small deterministic class · D-4 automatic dispatcher · D-5 label-driven trigger contract.

Open follow-ups filed by this lane and not adopted here: #1542 (quality:gate root coverage), #1543 (undeclared plugin-streams-core imports).

Definition of Done

• fix(fresh/ai): chat stream proxy drops durable State-Protocol query parameters #1457 merged to main through the pre-merge gate, acceptance truthfully ticked with linked evidence
• bug(fresh): DeferComponent is not hydrated, so partial-miss regions never refresh #1459 merged to main through the pre-merge gate, including the client-bundle regression test its acceptance requires
• Fresh Streams browser resolver cannot see Aspire VITE service references #1548 merged to main through the pre-merge gate, with the resolver decision recorded rather than left implicit
• cut-trace.md extended with each merge captured from git log origin/main after the fact
• Retrospective addendum written, encoding no policy the reopen's own evidence does not support

[rickylabs]

[CHECKPOINT] — consumer-visible merge, for the fixes/release coordinator

Reporting a merged consumer-visible change per the 0.0.6 intermediary-canary strategy. This lane does not cut releases; this is a report, not a request to publish.

Landed

Merged	Commit	Issue	Consumer-visible effect
2026-08-12T11:40:14Z	5db37e7bb	#1457 (p1)	createNetScriptChatStreamProxy() now forwards durable State-Protocol query parameters (offset, live, handle, cursor) upstream; id remains dropped as application routing metadata

Why it matters to a consumer: durable chat subscribers lost resume and live-polling semantics because the proxy silently dropped the query string. rickylabs/eis-chat PR #169 carries a thin adapter that re-appends every parameter except id; that adapter can be removed once this ships.

Surface change: additive only — a new optional query mapping option on NetScriptChatStreamProxyOptions. Existing callers compile unchanged; their behaviour changes only in that the query is now forwarded, which is the fix.

Evidence

• IMPL-EVAL PASS (automatic dispatcher, run 31590876488).
• Seven-check pre-merge gate: slices/pre-merge-gate-1556.md. scaffold-runtime (postgres + sqlite), scaffold-static, code-quality, quality, check-test, surface-diff, deps-report, close-gate all pass.
• deno task --cwd packages/fresh test → 223/223; reverting the forwarding → 217/6, restored → 223/223.

Caveat worth carrying into any canary note

The package-quality verdict for packages/fresh rests on an explicit target scan, not on the repo quality:gate — its configured roots omit several published packages (#1542). Same caveat this lane recorded on #1528 and #1536.

Still in flight in this lane

• Fresh Streams browser resolver cannot see Aspire VITE service references #1548 (p1, browser stream discovery) — implemented, IMPL-EVAL running.
• bug(fresh): DeferComponent is not hydrated, so partial-miss regions never refresh #1459 (p1, deferred coordinator hydration) — PLAN-EVAL PASS, implementation next. Its client-bundle regression criterion is split to test(fresh): no capability to assert the deferred coordinator reaches the client bundle #1557, so bug(fresh): DeferComponent is not hydrated, so partial-miss regions never refresh #1459's PR will carry Refs, not a closing keyword, until that lands.

I will report each of these as its own checkpoint when merged.

[rickylabs]

[CHECKPOINT] — consumer-visible merge, for the fixes/release coordinator

Second consumer-visible checkpoint from this lane. This lane does not cut releases; reporting only.

Landed

Merged	Commit	Issue	Consumer-visible effect
2026-08-12T12:03:55Z	59e435c5d	#1548 (p1)	The zero-configuration browser stream factory now resolves Aspire's canonical Vite service reference. createNetScriptStreamDB({ streamPath, schema }) connects without a baseUrl override instead of throwing "Durable streams URL not found…"

Why it matters to a consumer: with @netscript/fresh@0.0.5, a Fresh/Vite island using the documented factory threw before connecting even though the AppHost injected both VITE_services__streams__http__0 and VITE_STREAMS_URL. Consumers worked around it by passing baseUrl: import.meta.env.VITE_… ?? …, duplicating service discovery at every call site. That workaround remains supported but is no longer required.

Surface change: none. mod.ts is untouched — the fix is internal to the resolver, plus a narrow ImportMeta.env type declaration.

Evidence

• IMPL-EVAL PASS, run 31593936968 (sole authoritative evaluator for head ccfa5407; exactly-once confirmed by root).
• Seven-check pre-merge gate: slices/pre-merge-gate-1559.md. All named gates pass with pending = 0, including both scaffold-runtime tiers.
• 38/38 package tests. Regressing the reader to a computed index — behaviourally identical — fires the shape guard (36 passed, 2 failed).

Two caveats that belong in any canary note

1. No test proves Vite's define substitution fires in a real build. The source-shape guard is an explicit surrogate accepted by PLAN-EVAL. End-to-end confirmation needs a real Fresh/Vite build — worth exercising in a canary against a scaffolded app.
2. packages/sdk/src/discovery/browser-env.ts:65 has the same defect and is not fixed here. If the SDK's browser discovery is in the canary's blast radius, expect the same symptom there. To be filed once this shape is proven in a real build rather than fixed blind.

Also carried: packages/plugin-streams-core's quality verdict rests on an explicit target scan, not the repo quality:gate, whose roots omit it (#1542).

Remaining in this lane

#1459 (p1, deferred coordinator hydration) — PLAN-EVAL PASS, implementation next. Its client-bundle regression criterion is split to #1557, so its PR will carry Refs, not a closing keyword.

[rickylabs]

Closing as a non-merge orchestration record.

This control PR carried orchestration evidence only and never had a closing keyword. Its lane is complete, so it is closed rather than merged, at owner direction, so it does not remain an open item against the 0.0.6 milestone.

What it recorded

The narrow reopen of the 0.0.6 runtime / public-surface lane. Everything it coordinated has landed through its own leaf branch and PR against main:

PR	Issues	Merge
#1593	#1583	f542f31cb
#1595	#1589	4dc4d8262
#1607	#1577	1f9efb4d
#1600	#1569	6aee2b414
#1602	#1576, #1568	1ed78f508
#1605	#1562	bfcf4ed11

Plus, earlier in the run: #1457, #1459, #1548, #1571, #1580, #1227, and the terminal-green Canary.4.

Every referenced issue is closed and carries status:shipped. The runtime / public-surface lane is empty.

Verified before closing

$ git diff --name-only origin/main...4560ef9e5 | grep -vc '^\.llm/runs/'
0

72 commits, zero non-run-artifact files — it is what it claimed to be, an evidence-only control branch, and merging it would add nothing to main but history.

One consequence, stated so it is not discovered later

The artifacts now live only on chore/release-0.0.6-runtime-reopen.

This lane's first control PR, #1525, was merged at 0f0b6b6a3, so that evidence reached main. This one is not, so the branch is the sole copy of the reopen's record: drift entries D-1 through D-14, every slice brief, every pre-merge gate record, and the lane retrospective in cut-trace.md.

Please do not prune this branch as a stale chore/ branch. Deleting it destroys the entire evidence trail for this reopen — including the corrections this lane made to its own reasoning, which are the parts most likely to matter to whoever runs the next milestone.

Handover

No release was dispatched from this lane. Canary.5's blockers are both cleared — #1599 merged carrying canary:0.0.6-canary.5, and #1605 merged at bfcf4ed11. 0.0.6 holds one open issue, docs-owned #1531, whose PR #1608 is non-draft.

Release ownership hands to ns006-fixes.