fix(frontend): send committed revision reference only on clean agent runs

[ardaerzin]

Context

Follow-up to the run-context plumbing in #4892, per Mahmoud's coordination note on direct-call tools: when a variant is committed, the playground should send the committed revision reference with the run request (not just the inline config) so the service can mark the run non-draft and let a self-targeting tool bind its own revision/variant via runContext.

Problem

The canonical playground run builder (buildAgentRequest, used by AgentChatPanel) forwarded the full reference family — application + application_variant + application_revision — on every run whenever the entity carried a UUID, ignoring whether the left-panel config had unsaved edits.

The agent service derives draft-ness purely from a resolved committed-revision reference (services/oss/src/agent/tracing.py → _run_context_workflow: is_draft = revision is None), and the SDK resolver (/applications/revisions/retrieve with resolve: True) re-resolves a bare variant ref to its latest revision. So a dirty (edited) run still got marked non-draft, and a self-targeting tool would bind a revision whose config differs from what is actually running.

What changed

web/packages/agenta-playground/src/state/execution/agentRequest.ts — forward the reference family only when the run targets a committed, unmodified revision:

const isCommittedRevisionRun =
    !isDirty && typeof fullReferences?.application_revision?.id === "string"
const references = isCommittedRevisionRun ? fullReferences : null

• Clean committed revision → full references → service marks the run non-draft; a self-targeting tool binds the exact revision/variant.
• Dirty (unsaved edits) or uncommitted local draft → references: null — the service's documented "unsaved inline config carries no revision reference, so is_draft is True" case.
• application_id URL query is derived independently from the full identity, so a draft run stays associated with its app.

Tests

web/packages/agenta-playground/tests/unit/agentRequest.test.ts — added clean→refs, dirty→null, and uncommitted-local-draft→null coverage (replacing the old partial-ref local-draft case). 24 tests pass, tsc --noEmit clean, source lint clean.

Note / coordination item

Dropping all references for draft runs assumes draft-run traces still surface via application_id/OTel rather than body references. Expressing "draft run bound to a variant" (variant bound but still draft) would need a backend change — an explicit is_draft flag, or not auto-resolving variant → latest revision. This PR is the FE half and is forward-compatible either way.

[vercel]

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

Project	Deployment	Actions	Updated (UTC)
agenta-documentation	Ready	Preview, Comment	Jun 28, 2026 11:55am

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: a40f4b97-97b6-484b-8fa1-cd218e248eeb

📥 Commits

Reviewing files that changed from the base of the PR and between 27228f1 and 8bf2c8c.

📒 Files selected for processing (2)

• web/packages/agenta-playground/src/state/execution/agentRequest.ts
• web/packages/agenta-playground/tests/unit/agentRequest.test.ts

---

📝 Walkthrough

Summary by CodeRabbit

• Bug Fixes
  • Improved request handling so run references are only sent when the run is clean and tied to a committed revision.
  • Kept application scoping in the request URL even when references are omitted for drafts or dirty runs.
  • Updated coverage to verify references are included for committed runs and excluded for dirty or uncommitted drafts.

Walkthrough

buildAgentRequest now conditionally sets references to null when isDirty is true or when no committed application_revision.id exists, using a new fullReferences intermediate. The application_id URL parameter is sourced from fullReferences instead of the gated references. Tests are updated to mock isDirty and cover clean, dirty, and uncommitted draft scenarios.

Conditional references and isDirty gating

Layer / File(s)	Summary
Conditional references and URL app scoping src/state/execution/agentRequest.ts	fullReferences is built unconditionally from the entity; references is set to fullReferences only when isDirty is false and a committed application_revision.id exists, otherwise null. application_id in the URL query is sourced from fullReferences to preserve app scoping in all cases.
Test mock, seed, and scenario coverage tests/unit/agentRequest.test.ts	isDirty selector added to the workflowMolecule.selectors mock; seed helper extended to accept and set isDirty (default false). Tests replaced/added for clean committed (references present), dirty committed, and uncommitted local draft (both expect references: null with application_id still in the URL).

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly captures the main change: gating committed revision references on clean agent runs.
Description check	✅ Passed	The description accurately explains the request-body reference gating and added test coverage.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fe-feat/playground-commit-revision-runcontext

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[ardaerzin]

@coderabbitai review

[coderabbitai]

✅ Action performed

Review finished.

Note: CodeRabbit is an incremental review system and does not re-review already reviewed commits. This command is applicable only when automatic reviews are paused.