LUNA Lab
EN
Help
Navigation
Choose language
Print viewPDFMarkdown (.md)Text (.txt)Export this page’s content and sources.

Actions and follow-up: recommend, compose, approve and review

Find a task, follow the steps, and check the result. Guidance for researchers and administrators, in both languages.

Reviewed 2026-10-12 · 50 topics

Search these guides in Korean or English. This does not search private research materials.

Choose another topic

Match destinations, compare AI options, approve actions and review results.

Reviewed · 2026-10-12

On this page · 11

1. Prepare an action from a plan

  1. Act on Today’s research opens a side panel beside your plan; Full page opens the dedicated screen. Items in the Follow-ups list open as full pages. The top Follow-ups link returns to the previous filters, page and scroll position. For manual writing, select a section and inspect the linked plan and evidence.
  2. Check the method and destination. A cited issue may suggest commenting on that issue. Expand destination content/recent status to confirm the original you intend.
  3. Write the content or explicitly request an AI wording draft. Saving a draft prepares privately and does not send.
  4. Review the actual account, destination and body in Preview before execution, then approve. Report/event handoffs still require final saving in their forms.
  5. Record results and the next check afterwards. Later review drafts can use this action and confirmed observations as evidence.
Back to search and contents ↑

2. Match evidence and verify the destination

  1. Linked from evidence · No AI used means citations, links or exact mentions produced candidates. It is not an AI-confirmed decision or completed delivery.
  2. Choose a method, then type a name, title, issue number or repository in Destination. Up to six suggestions appear without a Search button; More shows the next page. Before typing, only up to three exact evidence recommendations appear. Use arrow keys and Enter or click to select. The selection collapses to one line; Change preserves the title and body. Search uses no AI and distinguishes errors from no matches.
  3. A cited source is not necessarily the place to publish. A comparison report does not automatically become the destination for an instruction. Decide whom you need to ask and why.
  4. Inspect destination content/recent status for the body, assignees and available later comments. Freshness depends on connection type, access and read limits; check timestamps and errors.
  5. If candidates are absent or ambiguous, select manually or open the original. Do not mistake a private personal record for an external destination.
Back to search and contents ↑

3. Choose among AI action recommendations

  1. AI recommendations prepares up to three alternatives from current evidence and permitted destinations. Adjust the direction toward an existing issue, direct inquiry, report/discussion or personal work, or add instructions.
  2. Compare purpose, method, destination and reference numbers. Alternatives might ask in an existing issue, message a relevant researcher or discuss evidence below a report.
  3. Recommendation batches are saved. After using one, expand the recommendations to create another draft. Continue draft opens existing wording; View result opens the completed action. Reuse consumes no AI allowance, and each action still requires review before posting.
  4. New recommendations preserve earlier batches in a collapsed area. Check creation time and current destination context. Inaccessible destinations are unavailable; repeating an action requires explicitly choosing Create another action.
  5. Use Change to correct an unsuitable destination manually. Multiple cited sources do not mean the same text should be sent everywhere.
  6. AI recommendations and AI wording are separate requests: the first chooses a method/destination; the second drafts content for the selected destination.
Back to search and contents ↑

4. Know where each method leads

  1. GitHub comments post to an existing issue; new GitHub issues post to the selected repository. Chat messages go to the selected channel/direct conversation, and discussions go below the selected report/material. Verify access and destination.
  2. PI check-in sharing prepares a report for its recipient in LUNA. Personal work/check-in is a private record with no external publication.
  3. Manual/AI report creation and event creation prepare metadata and open the existing form. Recheck scope, period, dates, times, participants and audience there before saving.
  4. Continue code work links the repository and task context. Opening a repository does not complete local changes, commits or pull requests.
  5. For Notion or another supported publication, review the connection, actual destination and preview too. Publication does not imply overwriting the original or approving the research result.
Back to search and contents ↑

5. Compose text, mentions, citations and attachments

  1. Select a method and destination, then write the title and body. Optional AI wording is prepared for that destination and applied after review. Per-request instructions can specify questions and length.
  2. Expand Edit & attach for Markdown headings, lists, tables, @mentions, [[document citations]], file selection, pasted images and drag/drop. Check upload limits and errors.
  3. Select researchers from mention autocomplete. GitHub delivery uses explicitly linked GitHub IDs; a LUNA username is not assumed to be the same GitHub identity.
  4. Check whether recipients can read cited documents. Links and attachments affect what is shared, so inspect the actual outgoing body and links.
  5. Preview headings, lists, links and attachments before saving. If the source or destination changes, review a fresh preview instead of reusing earlier approval.
Back to search and contents ↑

6. Drafts, approval and uncertain delivery

  1. Save draft prepares an action. An AI draft marked ready means wording is ready, not that publication succeeded.
  2. Review the sending account, destination, body, attachments and changes, then choose Confirm and execute. Complete any separate creation form in that form.
  3. For success, inspect the result link and history. Sent does not mean read, answered, approved or research completed.
  4. Execution needs checking or an unknown response can mean the external write succeeded without a receipt. Inspect the original and history before sending again.
  5. If access, source or content changed, recheck the destination and use a new preview. Errors do not require deleting personal notes or original issues.
Back to search and contents ↑

7. Inspect linked replies and later activity

  1. Open action details → Results and next check → Check linked replies. Supported destinations are read for replies and activity after the action within bounded limits.
  2. Directly linked replies and later activity on the same issue are labeled differently. A new issue comment is not necessarily an answer to your question; inspect content, author and time.
  3. Background checking is supported but is not unlimited real-time monitoring. Recent actions are checked in bounded batches; access and connection limits can cause delays or omissions.
  4. Check observation timestamps, errors and original links. No found reply does not prove inactivity. Inspect the original when needed.
  5. Separate sending an instruction, receiving a reply and verifying an outcome. AI reviews preserve those distinctions and do not invent completion or approval.
Back to search and contents ↑

8. Record personal results and share them

  1. Personal work/check-in records something you do yourself. It is private and is not automatically published to the lab or a researcher.
  2. Save directly verified results and next steps in Results and next check. A follow-up date schedules your own review; it is not an imposed recipient deadline or an automatic message schedule.
  3. Share result creates a new sharing draft from the saved outcome. Choose an available chat, discussion, GitHub or report route, then preview and approve again. The original private record is not automatically published.
  4. I finished my follow-up clears your follow-up candidate. It is different from external task completion, experimental success or PI approval.
  5. Your results are self-reports; linked replies are observed evidence for AI review. They feed newly generated reviews/plans without automatically overwriting earlier saved text.
Back to search and contents ↑

9. History, counts, cancellation and deletion

  1. Use the number beside a plan or Follow-ups → View all to inspect actions. Event-history counts can include state changes and differ from confirmed execution counts beside plans.
  2. Cancel a pre-execution draft with Cancel this action. This does not undo an existing GitHub comment, issue or chat. Check results first for processing or uncertain actions.
  3. Expand Delete record in the list/details, review its irreversible scope and confirm. Draft, cancelled and posted records can be removed, but processing work needs outcome checks first.
  4. Deleted personal actions are excluded from lists, detail views and AI evidence. Minimal receipts needed to prevent duplicates or authorize shared attachments may remain.
  5. Deleting a personal execution record does not delete external issues, comments, chats, reports or original attachments. Use the original service's controls and permissions for those.
Back to search and contents ↑

10. Recurrence, scheduling and GitHub access

  1. Expand recurrence/scheduling only when needed. It starts disabled; recurring draft preparation and scheduled fixed-text delivery are different.
  2. Scheduled delivery requires the complete actual destination/body preview and separate confirmation. Free-text instructions do not authorize arbitrary recurring external actions.
  3. Changed sources, content or permission can stop a schedule or require renewed review. Check status/history and recurrence when cancelling or deleting.
  4. In personal GitHub connection settings, connect only your account's required repositories. This does not grant access to other researchers' private scopes.
  5. Disconnecting blocks later execution but leaves existing publications. Linking to repository work is distinct from actually committing code or opening a PR.
Back to search and contents ↑

11. English display and translated drafts

  1. In English mode, action lists, pickers and editors translate titles, destination labels, linked sources, recommendations and expanded context with the existing free translator on supported devices. Destination IDs and links stay unchanged and no personal AI allowance is used.
  2. Korean Title, Content and Follow-up fields first show an English preview. Choose Edit original to revise the source or Use English draft to apply the translation to the field. Viewing a preview or saving a draft alone does not replace the original with English.
  3. View original restores display text. It does not undo edits already applied with Use English draft. Late translations never overwrite text you are editing.
  4. Mentions, document-citation syntax, link URLs, code, filenames, dates and Markdown markers are preserved. Review names, terminology and context before applying. The final execution review shows the actual saved text that will be submitted and is not automatically translated.
  5. Translations of a saved action's title, body and follow-up criteria are reused as private copies. Changed source versions or access invalidate reuse; deleting the action clears its copies. Unsaved edits are translated only in the current page.
  6. Choose Enable translation if the device needs preparation. Unsupported devices show a saved copy or the original, with no automatic switch to paid AI. Full pages and the action drawer in Today's research use the same behavior.
Back to search and contents ↑

Still need help?

Include the page, time, steps and error message. Leave out passwords, tokens and private research text.

Contact support →