[OwenKephart]

Summary

The v27 migration needs proof that committed facts, resumed execution, and rendered results agree through failure and suspension. This PR exercises those boundaries and retains the runtime fixes found there, including cancellation during admission and instrumentation hearing session creation. It now also tests real idle/stale cancellation receipts and schema-4 delivery attribution through eve's OTel provider. Stacked on #4854; #4855 follows. Related to #4930.

Coverage includes steering, approvals, child relays, interrupted sign-ins, compaction, call-ID collisions, and eval ordering/accounting. A cancellation after admission consumes the selected message into its cancelled turn, so retained history and the stream agree. Hook cancellation does not escape admission and retry that message.

Scope boundaries:

• Failure placement, child-delivery attribution, shared activity selection, and acceptance-aware frontend tests belong to refactor(eve)!: let the session stream own the frontend lifecycle #4854; the child-delivery golden integration case remains here.
• Repeated-call-ID protection covers IDs retained in the operational view or model prompt, not IDs discarded by retention or clear/compaction.
• Native fact-to-span tests replace relevant old tracing coverage; this does not restore every deleted tracing suite.
• The existing expected failure remains: a failed session closes its open turn as cancelled, so the turn span has no session error.
• Reactions feat(eve)!: compute dynamic capabilities with functions and memos #4616 is untouched. Most release notes remain collected here; the cancellation and instrumentation notes now accompany their own lower PRs.

It also checks the stream's promises on random streams, and logs a turn step that disagrees with waiting() under EVE_CHECK_SESSION_EVENTS. Waits are announced from the tables alone, and only the writer closes and scopes facts.

Production code: 524 more lines (+729/−205). Growth comes from runtime fixes uncovered by coverage: Teams' private sign-in reveal, call-ID collision protection, and cancellation-safe publication. The dev-only step-decision check adds 41 lines; announcing from the tables and one cascade site remove 10. Excludes tests/fixtures, generated copies, docs, and tooling.

Validation

At assembled head b1c123f55 (runtime unchanged from the validated D8 snapshot):

• Sixteen explicitly selected unit files: 269 passed, one pre-existing expected failure, in 5.15 seconds. No whole test tier ran.
• Package TypeScript check/declaration build, docs, invariant guard, targeted lint, formatting, and diff checks passed.
• Unknown-tool/sign-in and task-cancel-origin regressions fail on the prior owners, then pass with the fixes.
• pnpm --filter eve build:js resolved the earlier config-import blocker. Two real idle/stale receipt cases pass (17.94 seconds) and fail against the prior owners on absent settlements. The actual SDK retry/per-call metadata case passes (11.11 seconds), as does the concurrent-approval requester case (16.44 seconds).
• Raw workflow fixtures now supply dispatch identities at their sender boundaries, including their initial contexts. They do not inject admission or settlement facts. The stranded-reset assertions remain CI-owned.
• No full integration/scenario/e2e or browser run. Capability reports were not regenerated during this restack; final generation is coordinated against the assembled cleanup source.

Latest type-only follow-through at assembled head 390f9e5b3: AgentMessageResult.error.message
is restored to string. Package typecheck/declaration build, targeted lint, formatting, and diff
checks pass. Runtime tests were not rerun for this annotation-only correction. The invariant guard
reports only the unreleased tool/v93 contract hash; generation remains deferred to the agreed
final assembled-source pass. No reports or baselines were rewritten.

Follow-through at 77e487f18: four commits announce waits from the tables alone, cascade only in the writer, log step decisions that disagree with waiting(), and check the stream's promises on random streams. The step check labels deferred dispatch and doesn't log it. The property test runs 20 seeded streams in the unit tier (about 0.6 seconds), and EVE_STREAM_PROPERTY_RUNS raises the count. It fails if the reply retention change in #4567 is reverted. At this head, the package typechecks, and src/protocol, src/harness, and src/execution unit tests have one pre-existing failure (v26-transcript-lines). The scoped integration suites show the same 4 failures with and without the commits, and no new checker rule fires. At #4855's head, lint, formatting, and pnpm docs:check pass, and the invariant guard reports only the existing tool/v93 hash.

Checklist

• I added tests and documentation where relevant
• I added a changeset if this touches the published eve package
• DCO sign-off passes for every commit (git commit --signoff)

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
eve-docs	Error	v0	Oct 20, 2026 5:34pm UTC
eve-pkg	Ready	Preview, v0	Oct 20, 2026 5:34pm UTC

[github-actions]

Bundle + Package Summary: apps/fixtures/weather-agent

Key takeaways

• No notable deltas vs v27/observable-session (67e1dc2).

Delta vs v27/observable-session (67e1dc2)

Area	Metric	Baseline	Current	Delta
Package	Packed tarball	7.65 MB	7.65 MB	+286 B ⚠️
Package	Unpacked publish size	26.90 MB	26.90 MB	+4.1 kB ⚠️
Package	Installed footprint	74.97 MB	74.98 MB	+4.1 kB ⚠️
Package	Published files	4198	4200	+2
Package	Installed files	8207	8209	+2
Package	Installed package instances	34	34	0
Package	Distinct installed package names	32	32	0
Package	Installed dependency edges	51	51	0
Package	Installed optional peer edges	9	9	0
Runtime	Unique function payloads	2	2	0
Runtime	Total function bytes	24.53 MB	24.54 MB	+9.4 kB ⚠️
Runtime	Public routes	17	17	0

Changed function payloads vs v27/observable-session (67e1dc2) (2)

Function	Status	Baseline	Current	Delta	Route changes
functions/__server.func	changed	12.27 MB	12.27 MB	+4.7 kB ⚠️	none
functions/.well-known/workflow/v1/flow.func	changed	12.27 MB	12.27 MB	+4.7 kB ⚠️	none

eve init install

Metric	Baseline	Current	Delta
Installed footprint	113.26 MB	113.26 MB	+4.1 kB ⚠️
Installed packages	97	97	0
dependencies	4	4	0
devDependencies	2	2	0
Dependency package bytes	45.04 MB	45.04 MB	+4.1 kB ⚠️
devDependency package bytes	5.11 MB	5.11 MB	0 B ➖

Build Metadata

• Preset: vercel
• Nitro: nitro@3.0.260903-beta
• Output directory: apps/fixtures/weather-agent/.vercel/output
• Build metadata timestamp: 2026-10-14T23:56:15.143Z
• Route aliases: 17 public, 1 internal (18 total aliases)
• Vercel routes in config: 20
• Severity legend: 🔴 dominant/large, 🟠 notable, 🟡 watch, ⚪ small

Package Drill-Down

Package Details

• Package: eve@0.79.1
• Package directory: packages/eve
• Tarball: 7.65 MB (eve-0.79.1.tgz)
• Unpacked payload: 26.90 MB across 4200 published files
• Installed footprint: 74.98 MB across 8209 installed files
• Installed root package: 26.90 MB
• Installed dependencies: 48.08 MB
• Installed package instances: 34
• Distinct installed package names: 32
• Installed dependency edges: 51
• Installed optional peer edges: 9
• Runtime dependencies: 2
• Peer dependencies: 7 (6 optional)

Installed footprint is measured from an isolated temporary npm install of the packed tarball.
Graph metrics read only package.json files in package directories directly beneath a node_modules boundary, including nested boundaries. Each directory is one package instance; distinct names come from those manifests. Dependency edges count each unique name in dependencies or optionalDependencies per instance; optional peer edges count peerDependencies marked optional.

Heavy installed dependencies

• eve: 26.90 MB (35.9%)
• @rolldown/binding-linux-x64-gnu: 19.81 MB (26.4%)
• ai: 8.20 MB (10.9%)
• zod: 6.15 MB (8.2%)
• undici: 3.56 MB (4.8%)

Publish payload breakdown

Published file size
🟠 dist/src/compiled/shadcn-registry/index.js       [#####...................] 3.85 MB 14.3%
🟠 dist/src/compiled/@photon-ai/chat-adapter-ime... [###.....................] 2.29 MB 8.5%
🟠 dist/src/compiled/@ai-sdk/code-mode/index.js     [#.......................] 1.03 MB 3.8%
🟡 dist/src/compiled/@vercel/blob/index.js          [#.......................] 604.9 kB 2.2%
🟡 dist/src/compiled/_chunks/workflow/undici-D-I... [#.......................] 521.9 kB 1.9%
🔴 Other published files                            [########################] 18.61 MB 69.2%

Installed footprint breakdown

Installed package size
🔴 eve                             [########################] 26.90 MB 35.9%
🔴 @rolldown/binding-linux-x64-gnu [##################......] 19.81 MB 26.4%
🔴 ai                              [#######.................] 8.20 MB 10.9%
🔴 zod                             [#####...................] 6.15 MB 8.2%
🟠 undici                          [###.....................] 3.56 MB 4.8%
🟠 nitro                           [##......................] 1.89 MB 2.5%
🔴 Other installed packages        [########................] 8.46 MB 11.3%

Runtime dependencies (2)

Package	Range	Notes
nitro	3.0.260903-beta
undici	8.10.2

Peer dependencies (7)

Package	Range	Notes
@opentelemetry/api	^1.0.0	optional peer
ai	catalog:
braintrust	^3.0.0	optional peer
chat	^4.41.0	optional peer
dd-trace	^6.13.0	optional peer
just-bash	^3.1.0	optional peer
microsandbox	^0.5.0	optional peer

eve init install drill-down

eve init install details

• Command: eve init my-agent
• Package manager: npm
• Installed footprint: 113.26 MB across 10104 installed files
• Installed packages: 97 total (91 transitive-only)
• dependencies: 4 direct packages totaling 45.04 MB
• devDependencies: 2 direct packages totaling 5.11 MB
• Other transitive package files: 63.11 MB

Installed footprint is measured from an isolated temporary eve init my-agent using the current packed eve tarball.

Heavy installed dependencies

• @typescript/typescript-linux-x64: 27.95 MB (24.7%)
• eve: 26.90 MB (23.7%)
• @rolldown/binding-linux-x64-gnu: 19.81 MB (17.5%)
• zod: 9.76 MB (8.6%)
• ai: 8.20 MB (7.2%)

Installed footprint breakdown

Installed package size
🔴 @typescript/typescript-linux-x64 [########################] 27.95 MB 24.7%
🔴 eve                              [#######################.] 26.90 MB 23.7%
🔴 @rolldown/binding-linux-x64-gnu  [#################.......] 19.81 MB 17.5%
🔴 zod                              [########................] 9.76 MB 8.6%
🔴 ai                               [#######.................] 8.20 MB 7.2%
🟠 undici                           [###.....................] 3.56 MB 3.1%
🔴 Other installed packages         [###############.........] 17.09 MB 15.1%

dependencies (4)

Package	Range	Installed size	Share
@vercel/connect	2.2.0	188.7 kB	0.2%
ai	^7.0.137	8.20 MB	7.2%
eve	file:eve-0.79.1.tgz	26.90 MB	23.7%
zod	4.5.4	9.76 MB	8.6%

devDependencies (2)

Package	Range	Installed size	Share
@types/node	24.x	2.61 MB	2.3%
typescript	7.0.2	2.50 MB	2.2%

Function Drill-Down

Payload Size Graph

Unique function payload size and share of total
🔴 functions/.well-known/workflow/v1/flow.func     [########################] 12.27 MB 50.0%
🔴 functions/__server.func                         [########################] 12.27 MB 50.0%

Top Function Payloads

🟠 functions/.well-known/workflow/v1/flow.func • 1 public route • 12.27 MB

Metric	Value
Public routes	/.well-known/workflow/v1/flow
Runtime	nodejs24.x
Handler	index.mjs
Payload	12.27 MB
Function files	12.27 MB across 122 files
Traced dependencies	0 B
Signal	🟠 Bundled file _chunks/vercel.web.mjs is 4.26 MB (34.8%)

🟠 🔎 Dependency Analysis

📦 Bundled files:

Bundled file size
🟠 _chunks/vercel.web.mjs                         [######################..] 4.26 MB 34.8%
🟠 _libs/undici.mjs                               [#####...................] 1.01 MB 8.2%
🟡 _chunks/sandbox3.mjs                           [####....................] 815.4 kB 6.6%
🟡 _chunks/compiled-artifacts-instrumentation.mjs [####....................] 733.9 kB 6.0%
🟡 _chunks/token-util-Bla5Z0G9.mjs                [####....................] 704.7 kB 5.7%
🟠 Other bundled files                            [########################] 4.75 MB 38.7%

🧾 Vercel Config

{
  "handler": "index.mjs",
  "launcherType": "Nodejs",
  "shouldAddHelpers": false,
  "supportsResponseStreaming": true,
  "runtime": "nodejs24.x",
  "maxDuration": "max",
  "experimentalTriggers": [
    {
      "type": "queue/v2beta",
      "topic": "__eve776561746865722d6167656e74_wkf_workflow_*",
      "consumer": "default",
      "retryAfterSeconds": 5,
      "initialDelaySeconds": 0
    }
  ],
  "environment": {
    "WORKFLOW_PRECONDITION_GUARD": "1"
  }
}

---

🟠 functions/__server.func • 16 public routes, 1 internal alias • 12.27 MB

Metric	Value
Public routes	/ /.well-known/workflow/v1/webhook/[token] /eve/v1/callback/[token] /eve/v1/connections/[name]/callback/[attemptId]/[token] /eve/v1/connections/[name]/callback/[token] /eve/v1/health /eve/v1/info /eve/v1/session /eve/v1/session/[parentSessionId]/subagents/[callId]/[childSessionId]/stream /eve/v1/session/[sessionId] /eve/v1/session/[sessionId]/cancel /eve/v1/session/[sessionId]/clear /eve/v1/session/[sessionId]/compact /eve/v1/session/[sessionId]/reset /eve/v1/session/[sessionId]/stream /eve/v1/session/[sessionId]/stubs
Internal aliases	/__server
Runtime	nodejs24.x
Handler	index.mjs
Payload	12.27 MB
Function files	12.27 MB across 122 files
Traced dependencies	0 B
Signal	🟠 Bundled file _chunks/vercel.web.mjs is 4.26 MB (34.8%)

🟠 🔎 Dependency Analysis

📦 Bundled files:

Bundled file size
🟠 _chunks/vercel.web.mjs                         [######################..] 4.26 MB 34.8%
🟠 _libs/undici.mjs                               [#####...................] 1.01 MB 8.2%
🟡 _chunks/sandbox3.mjs                           [####....................] 815.4 kB 6.6%
🟡 _chunks/compiled-artifacts-instrumentation.mjs [####....................] 733.9 kB 6.0%
🟡 _chunks/token-util-Bla5Z0G9.mjs                [####....................] 704.7 kB 5.7%
🟠 Other bundled files                            [########################] 4.75 MB 38.7%

🧾 Vercel Config

{
  "handler": "index.mjs",
  "launcherType": "Nodejs",
  "shouldAddHelpers": false,
  "supportsResponseStreaming": true,
  "runtime": "nodejs24.x"
}

Build Timing: e2e/fixtures/agent-tools-sandbox

This is an informational timing measurement inside eve build, from preflight through publication. Output-size measurement and profile writing are excluded.

Build mode: deployable Vercel build with sandbox template prewarm included.

• Build pipeline: 3.14 s -> 3.07 s (-72.4 ms) vs v27/observable-session (67e1dc2).
• Timing is informational: shared GitHub runners are too variable for a hard timing budget.

Detailed phase timings vs `v27/observable-session (67e1dc2)`

Phase	Baseline	Current	Delta
extension.check	0.4 ms	15.2 ms	+14.8 ms
project.resolve	0.2 ms	0.4 ms	+0.2 ms
workspace.create	0.4 ms	0.5 ms	+0.1 ms
host.prepare	650.4 ms	576.8 ms	-73.6 ms
vercel.service-prefix.resolve	1.2 ms	1.1 ms	-0.1 ms
nitro.create	625.3 ms	634.3 ms	+9.0 ms
sandbox.prewarm	319.6 ms	270.5 ms	-49.1 ms
nitro.cache.prepare	0.2 ms	0.2 ms	0.0 ms
nitro.prepare	0.6 ms	0.5 ms	-0.1 ms
nitro.public-assets	0.6 ms	0.6 ms	0.0 ms
nitro.prerender	0.4 ms	0.5 ms	+0.1 ms
nitro.bundle	1.40 s	1.43 s	+29.4 ms
nitro.cache.write	0.4 ms	0.3 ms	-0.1 ms
vercel.workflow-function.materialize	79.0 ms	76.8 ms	-2.2 ms
agent-summary.emit	0.5 ms	0.5 ms	0.0 ms
connect-manifest.emit	0.3 ms	0.2 ms	-0.1 ms
nitro.close	0.1 ms	0.1 ms	0.0 ms
output.publish	2.9 ms	2.7 ms	-0.2 ms
workspace.remove	2.2 ms	2.1 ms	-0.1 ms

[OwenKephart]

Latest-main restack: combined top green

Head: 2a23bff78f49855886060a2db6ed973cd41ed5d9
Base: 4210766b5c5449e5b59bbddd5bb80a42111666ed

All current-head workflows pass:

• Main CI: all ten jobs pass, including typecheck, full unit tests, Linux/Windows integration, scenarios, templates, Web Chat, TUI, framework fixtures, and lint. The duplicate main CI run also passes.
• Local E2E: passed.
• Postgres E2E: passed.
• Vercel E2E: passed.
• Bundle Analysis: passed.

The newer coverage commit is preserved, including its focused span, eval, framework-binding, dynamic-resolver, route, and channel-notice contracts. Main's managed connection callbacks, replay-stable session deadlines, retention rules, and empty-root-response fix are preserved. The v27 extension epochs now use tool 93, dynamic-tool 87, and connection 51 without rewriting main's published history.

The first CI run exposed a stale message.completed assertion in main's new Workflow retention eval. It now checks one native text content.completed reply, retaining its history-retention assertions. The fixture typecheck and fresh combined-top CI pass. Superseded workflows were cancelled; these are not runtime failures.

Local validation passed package and basic-runtime fixture typechecks, 179 unit tests, and nine scoped integration cases, plus lint/format, guards, extension contracts/history, docs, and template drift. One span-error-status test inherited from the newer coverage commit remains an explicit expected failure. Expired-history rejection, empty-root acceptance, and deadline preservation failed with their respective main fixes disabled and passed after restoration.

All 19 PRs and their order remain intact in native stack #4913, rooted at main. All 109 stack commits have verified SSH signatures and DCO trailers. The 19 release notes remain centralized in #4726, and the production LOC deltas for #4854 and #4855 are unchanged. Unrelated reactions/research edits were left untouched.

This current-head result supersedes the earlier pending-CI notes in the PR descriptions.

[AndrewBarba]

Reviewed with a fresh-context reviewer (tests per the test-audit gate, plus every stack changeset). Nothing critical; the breaking PRs are correctly minor. Approving.

Fix before release:

• .changeset/v27-turns-wait-on-people.md: "Text produced while waiting on tasks is narration" only holds with finalReplyOnly (subagents). Root sessions still get reply, as this PR's own held-turn.root.eval.ts change shows.
• tracing/eve/session-spans.test.ts:278 it.fails records that a failed session leaves invoke_agent without error status (it was set on main via turn.failed). Meanwhile instrumentation-vocabulary.md says the trace schema is unchanged. Fix it, or disclose it and link a tracking issue.

Minor: edge-cases.integration.test.ts:864 races on setTimeout(…, 500); per-read 20s timeouts vs the 30s test timeout; eve-events-entry.md describes intermediate states users never had; the changeset audit comment is stale (20 files, next minor is 0.82.0).

[vercel]

Additional Suggestion:

Registry channel files import a non-existent export messageToUserContent from eve/channels/chat-sdk, causing TS2724 build failures