Skip to content

[Improve] Recommend follow-up work in conversation - #2309

Draft
roomote-roomote[bot] wants to merge 1 commit into
developfrom
feature/conversational-recommendations-2fbl4w457vsaz
Draft

[Improve] Recommend follow-up work in conversation#2309
roomote-roomote[bot] wants to merge 1 commit into
developfrom
feature/conversational-recommendations-2fbl4w457vsaz

Conversation

@roomote-roomote

Copy link
Copy Markdown
Contributor

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

Related issue

No linked issue; explicitly requested internal Roomote work.

Why this PR exists

  • A maintainer explicitly invited this PR in the linked issue or discussion
  • I am a maintainer / this is internal Roomote work

Follow-up recommendations should be part of the conversation: explain useful work, ask what to start, and use the normal task-start flow after acceptance instead of emitting new suggested-task cards.

What changed

  • Remove structured suggestion arguments and emission from Fast and worker chat replies, including the sandbox suggestion API client.
  • Align shared, Fast, and sandbox prompts, automation tool descriptions, workflow guidance, and public docs around concrete prose recommendations and accepted task starts.
  • Override older saved instructions requesting cards at runtime without rewriting saved preferences or migrating data.
  • Preserve existing saved suggestion records, card UI, reaction/Start handlers, and schema so previously posted cards remain usable. Explicitly authorized autonomous automation act workflows remain autonomous.

New recommendations no longer depend on structured cards or hidden fields for their implementation context. Existing cards and authorized autonomous work retain their behavior.

How it was tested

  • Delivery rerun on the latest fetched develop: five Fast/shared-prompt/workflow test files passed (304 tests); three worker reply/API-client/tool-description test files passed (84 tests).
  • Earlier implementation validation passed: Fast 285, worker 91, shared prompts 50, and types 11 tests; cloud-agents/types/worker type checks; worker lint; scoped formatting and oxlint; Mintlify validation and broken-link checks.
  • Commit hook passed staged oxfmt. Push hooks passed pnpm exec oxlint --deny-warnings, pnpm exec turbo lint --log-order grouped --output-logs new-only, pnpm check-types:fast (27 tasks), and pnpm knip. Knip emitted only a non-failing configuration hint to remove mint from ignoreBinaries.
  • git diff --check passed. The implementation completion review reported no remaining gaps after a stale worker test was corrected.
  • Browser proof is not applicable to these nonvisual tool contracts, generation rules, and documentation changes; no screenshots or simulated chat proof are claimed. Full end-to-end live chat was not tested.

Checklist

  • The PR title follows the repo convention: [Fix], [Feat], [Improve], [Refactor], [Docs], [Chore] followed by a user-facing description
  • This PR is small and scoped to one change
  • pnpm lint and pnpm check-types pass locally
  • I added tests or included a clear manual validation note above
  • I removed secrets, tokens, private keys, and customer data from code, logs, and screenshots
  • If this change should appear in the changelog, I ran pnpm changeset

The exact full lint/type commands above were not run; the pre-push fast type checks, residual lint, oxlint, and implementation-scoped checks passed. No changeset was added in this delivery-only pass.

@roomote-community

roomote-community Bot commented Sep 7, 2026

Copy link
Copy Markdown
Contributor

1 issue outstanding. See task

  • Onboarding scans can no longer produce suggestions because they have no chat reply surface. (packages/cloud-agents/src/server/suggested-tasks-prompt.ts:161)

Reviewed d8b1dc6

- Do not return ideas that genuinely require cross-repository execution unless you can tie the launch target to one repository.
- Each suggestion should cover a different subsystem, flow, or file area when possible.
- Finish with one \`send_chat_reply\` call: put the concise report summary in \`message\`, set \`purpose\` to \`closeout\`, and put the final structured actions in \`suggestions\` so they appear beneath that report in the originating conversation.
- Finish with one \`send_chat_reply\` call: put the concise report and actionable recommendations in ordinary prose in \`message\` and set \`purpose\` to \`closeout\`.

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.

This prompt is also used by the web onboarding scan (apps/web/src/trpc/commands/task-suggestions/onboarding.ts), which launches a Scan with notifySlack: false and no chat context. In that runtime send_chat_reply is not registered, and this PR has removed the only path that persisted workItems(kind='suggestion'). The scan therefore completes without producing onboarding suggestions; ensureTaskSuggestions treats that completed run as empty and will not retry. Keep a non-conversational persistence/output path for onboarding, or split its prompt from chat automation reports.

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