test(e2e): assert the Factory Worker task-record contract, not shadow task data

[Soph]

https://entire.io/gh/entireio/cli/trails/1120

E2E Tests has failed on every push to main since ed9c31c0d (#2032, "capture background subagent work durably") — seven consecutive runs. One job (e2e-tests (factoryai-droid)), two tests, both at expected task checkpoint within 30s, and both failing again on gotestsum's automatic re-run, so this is deterministic rather than agent nondeterminism.

It bisects cleanly: run 32411315115 on ed34d4db5 (#2018) had Factory green, and ed34d4db5 is ed9c31c0d's only parent.

This is not a bug in #2032

waitForTaskCheckpoint polled the shadow branches for a tasks/ path — exactly what #2032 stopped writing. Under the new model a Worker's turn lands as a completed session.TaskRecord on the parent, and the transcript is materialized into the parent's checkpoint under tasks/<toolUseID>/ at condensation time.

#2032 already moved everything else onto that contract — its own integration test now asserts the inverse ("a Worker's turn must write a task record, not shadow data"), and CLAUDE.md says "records never live on the shadow branch". These two E2E tests were the only holdout, and they could not object during review because E2E Tests runs on push-to-main and never on a PR.

What this does

Moves both tests onto the shipped contract, in both directions:

• waitForCompletedTaskRecord polls .git/entire-sessions/*.json for a record with a non-empty completed_at. omitempty drops the zero value, so presence is the completion signal. Same strength as the existence check it replaces.
• assertNoShadowTaskData pins the inverse: no shadow branch carries a tasks/ path. Narrowed to task data deliberately — a parent session legitimately has shadow branches of its own, because shadow pinning keys on StepCount and says nothing about task content.

TestFactoryTaskCheckpointExistsBeforeCommit → TestFactoryTaskRecordExistsBeforeCommit. There is no pre-commit checkpoint under the new model, but the durability intent in the name is still what's tested: the Worker's work is captured before the user commits.

Also updates the stale cross-reference and pre-#2032 wording in factoryai_worker_session_test.go's header comments, which named the renamed test.

Scope note

Whether dropping the pre-commit shadow copy weakens durability is #2058's question, not this PR's. These tests assert what shipped; they do not settle whether it should have. Flagging explicitly so the red build isn't mistaken for that question being answered.

Verification

• mise run fmt && mise run lint — clean (0 issues)
• mise run test:ci — integration ok, canary 56/56 and 4/4, no failures
• go vet -tags=e2e ./e2e/tests/ — clean
• The changed E2E tests themselves are not run here: they need a real Factory Droid agent, so E2E Tests on merge to main is the actual verification. Prompts are untouched (Vogon regexes unaffected) and both tests skip on non-Factory agents (canary unaffected).

Also worth fixing separately

E2E Tests being push-to-main-only is what let this land. The Worker task path has no pre-merge coverage at all, and the canary is free and deterministic — extending it to cover the Worker task record would have caught this in review.

🤖 Generated with Claude Code

---

Note

Low Risk
Test-only assertion changes; no production behavior. Residual risk is E2E-only verification on merge to main.

Overview
Aligns Factory Droid E2E coverage with the post-#2032 contract: a Worker turn is a completed parent TaskRecord before commit, not tasks/ metadata on a shadow branch.

waitForTaskCheckpoint now polls .git/entire-sessions/*.json for a record with tool_use_id and completed_at, and assertNoShadowTaskData fails if any shadow tree still has a /tasks/ path. The pre-commit test is renamed to TestFactoryTaskRecordExistsBeforeCommit. Integration-test comments are updated to match.

^{Reviewed by Cursor Bugbot for commit 33d5866. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit 33d5866. Configure here.}

[Copilot]

Pull request overview

This PR updates Factory Droid coverage to assert the post-#2032 “task record → condensation materialization” contract: a Worker’s work should appear as a completed session.TaskRecord in session state before the user commits, and must not be written as tasks/ data on any shadow branch.

Changes:

• Rename and update the Factory E2E test to wait for a completed task_records[] entry in .git/entire-sessions/*.json rather than polling shadow-branch tasks/ paths.
• Add an E2E assertion that no shadow branch contains /tasks/ metadata paths.
• Refresh integration-test header comments to match the task-record model and updated E2E test naming.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
e2e/tests/factory_hooks_test.go	Moves E2E assertions from shadow tasks/ metadata to completed task-record state + adds a “no shadow task data” assertion.
cmd/entire/cli/integration_test/factoryai_worker_session_test.go	Updates comments to reflect that Workers are attributed via task records on the parent session.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.