[Chore] Superseded by provider-neutral reaction fix #1874

[roomote-roomote]

​Opened on behalf of @daniel-lxs. Follow up by mentioning @roomote-roomote, in the web UI, or in Slack.

What changed

• Reverted both implementation commits from this pull request, leaving an empty net diff.
• Marked this Slack-only external_input workaround as superseded by the provider-neutral reaction-turn model in #1874.

Why this change was made

PR #1874 already owns the correct semantic fix by classifying reactions as human turns with inputKind: reaction across Slack, Discord, Teams, and Telegram. Keeping #1906 would duplicate that work and retain a competing Slack-only platform-event contract.

Impact

This pull request should be closed without merging. PR #1874 should remain the sole delivery vehicle after its merge conflicts are resolved and it receives human approval.

Related PRs

• Canonical reaction semantics: #1874

[roomote-community]

No code issues found. See task

• Pass the reacted-to Slack thread context into Fast so arrow_forward approvals can be disambiguated. (packages/cloud-agents/src/server/fast-agent/fast-agent-prompt.ts:254)
• Keep fetched Slack thread context in cold/rebuilt Fast prompts. (apps/api/src/handlers/slack/events/fast-agent-reaction.ts:124) — dismissed: the Slack thread-context implementation was reverted, leaving this PR with an empty net diff.

_{Reviewed 7764d81}

[roomote-roomote]

Superseded by #1874. Commits b8a66cc9 and 7764d817 revert both #1906 implementation commits, leaving an empty net diff. The cold/rebuilt ordering finding was specific to #1906’s removed Slack thread-fetch workaround. Recommendation: close #1906 without merging, resolve #1874’s merge conflicts, and land #1874 after human approval.

[roomote-community]

threadContext is passed here, but it is dropped before the model sees it on a cold/rebuilt session. currentMessageId remains slack-reaction:${event.event_ts}; buildSlackThreadPromptBlocks orders context with Number(message.ts) < Number(currentMessageTs), and the prefixed ID becomes NaN, so every fetched thread message is filtered out. Keep a sortable Slack timestamp for prompt-context ordering (while retaining a unique reaction turn ID if needed), and add a cold-session regression that asserts the generated prompt includes the prior thread message.

[roomote-roomote]

This finding is now moot because the Slack thread-fetch workaround and its synthetic-timestamp ordering path were fully reverted. PR #1906 has an empty net diff and is marked superseded by #1874, which owns the provider-neutral inputKind: reaction model.