Skip to content

feat(eve): own subagent workflow tools in the framework - #2255

Closed
ruiconti wants to merge 1 commit into
mainfrom
rui/workflow-tool-subagents
Closed

ruiconti wants to merge 1 commit into
mainfrom
rui/workflow-tool-subagents

Conversation

@ruiconti

Copy link
Copy Markdown
Contributor

Summary

Related to #1084. With experimental.tasks, local and remote subagent calls become framework-owned defineTool(...) definitions whose execute values are separately compiled Workflow functions.

Before → after: task-mode delegation ran inside one parent dispatch step → task dispatch now selects the local or remote framework tool definition and starts its compiler-assigned Workflow before re-entering the existing durable task lifecycle.

The definitions live under runtime/framework-tools/subagent; the Workflow executors stay under execution, the compiler-scanned domain. That split matters: combining defineTool and a workflow directive in one source module embeds Zod and framework registration code in every Workflow function payload (~2.09 MB per function when measured on the original spike).

The model-visible subagent contract is unchanged. Local and remote executors stay separate through transport selection, then share task admission. The durable task run remains the sole lifecycle writer — local HITL over the child continuation hook, remote HITL over the task callback capability.

Rebased onto main, and what that changed

This replaces #2167, which targeted a frozen base off tasks-v4/03-runtime (37314a9f, not an ancestor of main). Replaying onto main picked up 76 commits of drift and settled four things worth a reviewer's eye:

  • task-event-callback.ts keeps main's version. The spike's forwardTaskCallbackEvent emitted locally and forwarded; main's forwardTaskEventToSessionCallback suppresses local emission so two TUIs can't both present the same HITL request. Main's behavior wins; the spike's fireTaskEventCallbackStep test is kept as added coverage main was missing.
  • Shared sandbox provisioning moved. Main provisions from the plan inside prepareRuntimeActionDispatch, but this change splits planning out of preflight. Provisioning is now a helper called once per batch — deliberately not on per-entry rehydration, which would create a second sandbox (caught by opens a shared parent sandbox for a background-task child).
  • sandboxSessionId is threaded into the workflow tool's startSubagent call, and executeTaskControlAction keeps main's adapter/serializedContext parameters.
  • tasksEnabled stays boolean on the park result rather than narrowing to true.

The 12 spike commits are consolidated into one, since several superseded each other (prototype → inline → framework-owned).

Known residual

The wrapper Workflow start() is at-least-once if the outer dispatch step retries after it succeeds. Task and child identities stay deterministic, so admission is replay-safe, but no wrapper-run idempotency key is persisted.

Validation

Unit (652 files, 6841 passed, 1 skipped), integration (91 files, 661 passed), typecheck, build, lint, format, and guard:invariants all pass. SPIKE.md records the runtime path and boundaries.

Checklist

  • I linked an issue with prior discussion confirming this change is wanted
  • I ran the relevant checks from CONTRIBUTING.md
  • 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)

With experimental.tasks, local and remote subagent calls are framework-owned
defineTool definitions whose execute values are separately compiled Workflow
functions. Task dispatch selects the local or remote definition and starts
its compiler-assigned Workflow before re-entering the existing durable task
lifecycle; the definitions live under runtime/framework-tools/subagent while
the executors stay under execution, keeping definition-time schema and
branding code out of every compiled Workflow program.

The model-visible subagent contract is unchanged, and the durable task run
remains the sole lifecycle writer. Consolidates the 12 spike commits into
one commit because several superseded each other.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Signed-off-by: Rui Conti <ruiconti@gmail.com>
@vercel

vercel Bot commented Aug 18, 2026

Copy link
Copy Markdown
Contributor

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

Project Deployment Actions Updated (UTC)
eve-docs Ready Ready Preview Aug 18, 2026 7:02pm
eve-docs-4759 Ready Ready Preview, v0 Aug 18, 2026 7:02pm
eve-pkg Ready Ready Preview Aug 18, 2026 7:02pm

@github-actions

Copy link
Copy Markdown
Contributor

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

Key takeaways

  • No notable deltas vs main (d397775).

Delta vs main (d397775)

Area Metric Baseline Current Delta
Package Packed tarball 8.23 MB 8.23 MB +2.9 kB ⚠️
Package Unpacked publish size 30.90 MB 30.91 MB +13.6 kB ⚠️
Package Installed footprint 74.43 MB 74.44 MB +13.6 kB ⚠️
Package Published files 3335 3347 +12
Package Installed files 7043 7055 +12
Package Installed package instances 35 35 0
Package Distinct installed package names 34 34 0
Package Installed dependency edges 52 52 0
Package Installed optional peer edges 47 47 0
Runtime Unique function payloads 2 2 0
Runtime Total function bytes 17.98 MB 18.00 MB +26.5 kB ⚠️
Runtime Public routes 16 16 0
Changed function payloads vs main (d397775) (2)
Function Status Baseline Current Delta Route changes
functions/__server.func changed 8.99 MB 9.00 MB +13.3 kB ⚠️ none
functions/.well-known/workflow/v1/flow.func changed 8.99 MB 9.00 MB +13.3 kB ⚠️ none

eve init install

Metric Baseline Current Delta
Installed footprint 112.84 MB 112.85 MB +13.6 kB ⚠️
Installed packages 121 121 0
dependencies 4 4 0
devDependencies 2 2 0
Dependency package bytes 45.40 MB 45.41 MB +13.6 kB ⚠️
devDependency package bytes 5.04 MB 5.04 MB 0 B ➖
Build Metadata
  • Preset: vercel
  • Nitro: nitro@3.0.260610-beta
  • Output directory: apps/fixtures/weather-agent/.vercel/output
  • Build metadata timestamp: 2026-08-18T19:03:44.938Z
  • Route aliases: 16 public, 1 internal (17 total aliases)
  • Vercel routes in config: 19
  • Severity legend: 🔴 dominant/large, 🟠 notable, 🟡 watch, ⚪ small
Package Drill-Down

Package Details

  • Package: eve@0.39.0
  • Package directory: packages/eve
  • Tarball: 8.23 MB (eve-0.39.0.tgz)
  • Unpacked payload: 30.91 MB across 3347 published files
  • Installed footprint: 74.44 MB across 7055 installed files
  • Installed root package: 29.56 MB
  • Installed dependencies: 44.88 MB
  • Installed package instances: 35
  • Distinct installed package names: 34
  • Installed dependency edges: 52
  • Installed optional peer edges: 47
  • Runtime dependencies: 2
  • Peer dependencies: 5 (4 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: 29.56 MB (39.7%)
  • @rolldown/binding-linux-x64-gnu: 19.41 MB (26.1%)
  • ai: 6.69 MB (9.0%)
  • zod: 5.07 MB (6.8%)
  • undici: 3.50 MB (4.7%)
Publish payload breakdown
Published file size
🔴 dist/src/compiled/shadcn-registry/index.js       [################........] 10.20 MB 33.0%
🟠 dist/src/compiled/@photon-ai/chat-adapter-ime... [####....................] 2.44 MB 7.9%
🟠 dist/src/compiled/experimental-ai-sdk-code-mo... [##......................] 1.51 MB 4.9%
🟡 dist/src/compiled/_chunks/node/undici-CUIybhQ... [#.......................] 508.8 kB 1.6%
🟡 dist/src/compiled/_chunks/workflow/undici-BBG... [#.......................] 508.8 kB 1.6%
🔴 Other published files                            [########################] 15.75 MB 50.9%
Installed footprint breakdown
Installed package size
🔴 eve                             [########################] 29.56 MB 39.7%
🔴 @rolldown/binding-linux-x64-gnu [################........] 19.41 MB 26.1%
🔴 ai                              [#####...................] 6.69 MB 9.0%
🔴 zod                             [####....................] 5.07 MB 6.8%
🟠 undici                          [###.....................] 3.50 MB 4.7%
🟠 nitro                           [##......................] 2.41 MB 3.2%
🔴 Other installed packages        [######..................] 7.81 MB 10.5%
Runtime dependencies (2)
Package Range Notes
nitro 3.0.260610-beta
undici 8.9.0
Peer dependencies (5)
Package Range Notes
@opentelemetry/api ^1.0.0 optional peer
ai catalog:
braintrust ^3.0.0 optional peer
just-bash ^3.0.0 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: 112.85 MB across 8931 installed files
  • Installed packages: 121 total (115 transitive-only)
  • dependencies: 4 direct packages totaling 45.41 MB
  • devDependencies: 2 direct packages totaling 5.04 MB
  • Other transitive package files: 62.40 MB

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

Heavy installed dependencies

  • eve: 29.56 MB (26.2%)
  • @typescript/typescript-linux-x64: 27.95 MB (24.8%)
  • @rolldown/binding-linux-x64-gnu: 19.41 MB (17.2%)
  • zod: 9.02 MB (8.0%)
  • ai: 6.69 MB (5.9%)
Installed footprint breakdown
Installed package size
🔴 eve                              [########################] 29.56 MB 26.2%
🔴 @typescript/typescript-linux-x64 [#######################.] 27.95 MB 24.8%
🔴 @rolldown/binding-linux-x64-gnu  [################........] 19.41 MB 17.2%
🔴 zod                              [#######.................] 9.02 MB 8.0%
🔴 ai                               [#####...................] 6.69 MB 5.9%
🟠 undici                           [###.....................] 3.50 MB 3.1%
🔴 Other installed packages         [##############..........] 16.72 MB 14.8%
dependencies (4)
Package Range Installed size Share
@vercel/connect 0.4.3 141.0 kB 0.1%
ai ^7.0.58 6.69 MB 5.9%
eve file:eve-0.39.0.tgz 29.56 MB 26.2%
zod 4.4.3 9.02 MB 8.0%
devDependencies (2)
Package Range Installed size Share
@types/node 24.x 2.54 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     [########################] 9.00 MB 50.0%
🔴 functions/__server.func                         [########################] 9.00 MB 50.0%

Top Function Payloads

🟠 functions/.well-known/workflow/v1/flow.func • 1 public route • 9.00 MB
Metric Value
Public routes /.well-known/workflow/v1/flow
Runtime nodejs24.x
Handler index.mjs
Payload 9.00 MB
Function files 9.00 MB across 50 files
Traced dependencies 0 B
Signal 🟠 Bundled file index.mjs is 2.02 MB (22.4%)

🟠 🔎 Dependency Analysis

📦 Bundled files:

Bundled file size
🟠 index.mjs                     [#################.......] 2.02 MB 22.4%
🟠 _chunks/runtime-artifacts.mjs [###############.........] 1.75 MB 19.4%
🟡 _libs/undici.mjs              [########................] 980.5 kB 10.9%
🟡 _chunks/sandbox.mjs           [#######.................] 781.9 kB 8.7%
🟡 _chunks/frames-DoexhtxJ.mjs   [######..................] 672.6 kB 7.5%
🟠 Other bundled files           [########################] 2.80 MB 31.1%

🧾 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 • 15 public routes, 1 internal alias • 9.00 MB
Metric Value
Public routes /
/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/task-input/[token]
Internal aliases /__server
Runtime nodejs24.x
Handler index.mjs
Payload 9.00 MB
Function files 9.00 MB across 50 files
Traced dependencies 0 B
Signal 🟠 Bundled file index.mjs is 2.02 MB (22.4%)

🟠 🔎 Dependency Analysis

📦 Bundled files:

Bundled file size
🟠 index.mjs                     [#################.......] 2.02 MB 22.4%
🟠 _chunks/runtime-artifacts.mjs [###############.........] 1.75 MB 19.4%
🟡 _libs/undici.mjs              [########................] 980.5 kB 10.9%
🟡 _chunks/sandbox.mjs           [#######.................] 781.9 kB 8.7%
🟡 _chunks/frames-DoexhtxJ.mjs   [######..................] 672.6 kB 7.5%
🟠 Other bundled files           [########################] 2.80 MB 31.1%

🧾 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: 2.57 s -> 2.12 s (-450.8 ms) vs main (d397775).
  • Timing is informational: shared GitHub runners are too variable for a hard timing budget.
Detailed phase timings vs `main (d397775)`
Phase Baseline Current Delta
extension.check 0.8 ms 0.9 ms +0.1 ms
project.resolve 0.2 ms 0.3 ms +0.1 ms
workspace.create 0.5 ms 0.6 ms +0.1 ms
host.prepare 251.4 ms 237.1 ms -14.3 ms
vercel.service-prefix.resolve 1.6 ms 1.6 ms 0.0 ms
nitro.create 185.4 ms 186.4 ms +1.0 ms
sandbox.prewarm 268.2 ms 528.9 ms +260.7 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 0.5 ms 0.5 ms 0.0 ms
nitro.prerender 0.4 ms 0.4 ms 0.0 ms
nitro.bundle 1.07 s 1.11 s +31.1 ms
nitro.cache.write 0.2 ms 0.2 ms 0.0 ms
vercel.workflow-function.materialize 592.2 ms 48.3 ms -543.9 ms
agent-summary.emit 0.5 ms 0.4 ms -0.1 ms
nitro.close 0.1 ms 0.1 ms 0.0 ms
output.publish 187.7 ms 2.5 ms -185.2 ms
workspace.remove 1.6 ms 1.4 ms -0.2 ms

Comment on lines +67 to +69
bundle: prepared.bundle,
parentStepIndex: prepared.batch.event.stepIndex,
parentTurnId: prepared.batch.event.turnId,

@vercel vercel Bot Aug 18, 2026

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

dispatchTaskStep omits adapter and serializedContext when calling executeTaskControlAction, so every task_update control call fails with "This session is not owned by a parent task."

Fix on Vercel

@ruiconti

Copy link
Copy Markdown
Contributor Author

Closing as a duplicate of #2260, which owns the Layer 1 framework subagent-tool work in its separate workstream. The per-call background and task_join layer remains in #2258 and will be stacked on #2260; this PR should not independently own or duplicate Layer 1.

@ruiconti ruiconti closed this Aug 18, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant