Let nested Factory automations poll, save settings safely, and show investigated Activity

[enzoames]

Summary

• Nested Factory Slack, GitHub, and Sentry automations can poll and run, including via the generic run-now path for jobs nested under jobs/.
• Create and Settings share the same source cards and a sticky Save/Discard bar; Save disables the form and does not overwrite edits made while the request is pending.
• Activity shows investigated items (held, dispatched, failed) with a collapsed Trace, including failed Slack dispatch from factory_runs.
• Migration 24 reconciles leftover default-factory rows, clears every extra factory sharing a Slack channel, then creates the unique index through the migration SQL executor.

Test plan

• Create or open a non-default factory, enable Slack polling, and click Run now on factory-slack-feedback.
• Confirm poll-slack-channel is no longer rejected with “not invoked by a governed Factory automation.”
• Change a Settings field and confirm the sticky Save/Discard bar appears; save and confirm the Automations list stays visible while a run is in progress.
• Switch factories with unsaved Settings edits and confirm Factory B does not inherit Factory A values.
• Open Activity after a poll: investigated items instead of raw check counts; held items show summaries; a failed Slack dispatch shows the error.
• Two factories on the same Slack channel: boot should clear the extra assignment instead of failing unique-index creation.

[github-actions]

Here's a visual recap of what changed:

Open the full interactive recap

[enzoames]

Addressed all three open Builder review items in ae5170d (also merged origin/main).

• Required factory_config reconciliation: collision-safe id rewrite in migration 21; merge source settings and poll cursors before deleting leftovers in 25.
• Required self-dispatch: restored the local fallback port to 3000; still uses inbound Host when present.
• Optional create-factory Slack member id: same schema as Settings.

No items skipped. PR not merged.

[enzoames]

Addressed all four open Builder review items in 300059e.

Required — migration 24 reconciles duplicate factory_config rows before creating the Slack-channel unique index; 25 still reconciles for workspaces that already ran the old index-only 24.

Optional — Settings remounts/resets hydration on factory switch; Automations clears queuedRuns on factory switch; Settings save baselines every trimmed field.

No items skipped. Branch is 1 commit behind main (unrelated Clips/Slides sessionStorage fix); no merge conflicts. PR not merged.

[enzoames]

Addressed all five open Builder review items in 4237cf9, and landed outcome-first Activity in the same commit.

Required — migration 24 clears every duplicate Slack-channel assignment (not only leftover/product-feedback) before creating the unique index.

Optional — unique-index DDL goes through migration sql; Slack channel+checkpoints merge as one identity; Settings disables controls while Save is pending and does not apply the snapshot over a newer edit; generic run-now accepts a nested path.

Activity now lists investigated items (held/dispatched/failed) with a collapsed Trace, including failed Slack dispatch from factory_runs.

No items skipped. Branch is 1 commit behind main (unrelated Clips/Slides); no merge conflicts. PR not merged.

[enzoames]

Addressed both open Builder review items in b359136.

Required — eventItemIds type predicate now returns a boolean (value.length > 0). This was the CI Typecheck failure (TS2322); Factory typecheck passes locally.

Optional — slack_workspace is taken from the same row that supplies the adopted Slack channel, so leftover secondary channels are not polled as primary.

No items skipped. Branch is 6 commits behind main (deploy/docs/desktop/plugin work, merge-tree clean, no Factory overlap); did not merge main. PR not merged.

[builder-io-integration]

Builder reviewed your changes and found 1 potential issue 🟡

Review Details

Incremental Code Review Summary

The latest revision fixes both previously open issues: the audit report’s item-ID predicate now returns an explicit boolean, and Slack workspace selection stays coupled to the adopted channel during reconciliation. I resolved those stale review threads before this review.

The new Activity projection and audit reporting are otherwise well structured, with factory/org scoping, investigated-item aggregation, provider failure visibility, and focused regression coverage. One failure-state gap remains in the new reporting path. This remains high risk because the change touches shared automation execution and operational failure reporting.

New Findings

• 🟡 MEDIUM — Activity only treats failed as a failed run, hiding persisted timed_out and reconciliation_required outcomes as successful/held activity without the stored error.
• 🟡 LOW — Collapsed scan traces emit a raw English Loaded … item(s). string instead of localized copy.

The migration ordering/conflict fixes, nested run-now support, and Settings state protections remain covered by the updated tests. 🧪 Browser testing: Will run after this review (PR touches UI code).

[builder-io-integration]

🟡 Show all terminal run failures in Activity

triageRuns can persist timed_out and reconciliation_required states with an error, but this projection only emits a dispatch error when status is exactly failed. Activity therefore presents callback/provider failures as successful or held outcomes with no error; treat all error-bearing terminal states as failed, including timed_out and reconciliation_required.