{"id":47903,"plugin_id":"plugin_asdk_app_6abe9a3cdf3c819191d8d505466917f1","kind":"skill","collection_source":"plugin_package","comparison_source":null,"observed_at":"2026-10-24T13:34:55.909Z","digest":"71cb9c2fb6c5c2f363bb321f8e21bd69c59cd1188f9cec20499140732e91c80e","against":null,"payload":{"description":"Check citations and manuscripts with Manusights. Use it whenever the user wants to know whether references exist, point to the right paper, or have been retracted or corrected, including citations an AI assistant gave earlier in the conversation, a draft's reference list or a literature review; when they want a journal's AI-use and disclosure rules; or when they want a readiness preview of an attached manuscript before submission.","included_files":[{"relative_path":"agents/openai.yaml","size_in_bytes":519},{"relative_path":"assets/icon.png","size_in_bytes":4573}],"name":"check-before-submission","skill_md_contents":"---\nname: check-before-submission\ndescription: Check citations and manuscripts with Manusights. Use it whenever the user wants to know whether references exist, point to the right paper, or have been retracted or corrected, including citations an AI assistant gave earlier in the conversation, a draft's reference list or a literature review; when they want a journal's AI-use and disclosure rules; or when they want a readiness preview of an attached manuscript before submission.\n---\n\nExplicit instructions from the user take priority over these steps. Manusights' results say how to report them; follow that guidance.\n\n## Check references\n\n1. Collect every reference the user wants checked: from their message, an\n   attached draft, or earlier in this conversation. Send citations you gave\n   earlier exactly as you gave them; do not correct, replace or defend them\n   first.\n2. Call `check_references` with every reference, with or without a DOI. For\n   a whole list, such as the References section of an attached paper or a\n   bibliography, send it exactly as written in `reference_list`, and keep\n   calling with `start_at` as the result says until the whole list is\n   checked. For a few references, send one item each, the reference exactly\n   as written in `text`, plus any other field you can read reliably. Never\n   check only a sample of a list.\n   When the original list's count is known, pass it as\n   `expected_reference_count` if that field is declared by the tool, and\n   compare it with the returned `progress.total`. It is the count of the\n   whole original list, not the current part. If the counts differ, resend\n   the complete original list; do not lower the expected count, silently\n   omit entries, or treat `next_start_at: null` for a shortened list as\n   completion. If the tool asks for numbered lines, preserve existing\n   reference numbers, or add sequential list labels without changing the\n   citation text. Clarify uncertain boundaries instead of inventing entries.\n   If the complete input cannot be sent, state exactly what\n   remains unchecked and why.\n3. Report the result the way its `how_to_report` says. Give each row's status\n   and `detail` as returned; do not restate a status in stronger or softer\n   words.\n4. Never call a reference fake, fabricated or verified, and never call a list\n   clean. If a reference you suggested is flagged or unchecked, say so\n   plainly. Show a suggested record only as a possible match for the user to\n   confirm; never replace a reference with it. Manusights does not judge\n   whether a source supports a claim; label any view of your own, on that or\n   anything else, as yours, apart from Manusights' results.\n\n## Journal AI policy\n\n1. Call `get_journal_ai_policy` with the journal's name or ISSN exactly as\n   the user supplied it, including a possible typo. Do not silently correct\n   the spelling, expand an ambiguous name, or replace it with another\n   journal before the first lookup. If the intended journal is unclear,\n   ask the user to confirm it. When an unmatched result offers\n   `name_suggestions`, present them as possible names for the user to\n   confirm; do not look up a suggested name until the user confirms it or\n   treat it as a confirmed policy match.\n2. State only the rules the result returns, with the date they were checked\n   and the official policy link. When the result says it cannot state the\n   rules, give its reason and the link instead.\n\n## Manuscript preview\n\n1. Call `review_manuscript` with the attached file. Pass `target_journal`\n   only when the user named one.\n2. A preview takes a few minutes. Read it with `get_manuscript_review` using\n   the same `review_id`, waiting as the result says between calls; do not\n   start another preview to check progress. While it is processing, report\n   its status only; do not add your own critique or predict the findings. If\n   the conversation stops first, tell the user they can ask you to retrieve\n   the result.\n3. When it is ready, report the findings as returned.\n4. Mention a Full Review only when the result shows one already on the\n   user's account, and then only its status or download link as returned. If\n   the user asks for a Full Review the account does not have, say it is not on\n   their account and link https://manusights.com/chatgpt#full-review, which\n   describes it. Do not name a price or encourage a purchase.\n\n## Earlier reviews\n\nUse `list_manuscript_reviews` to show the reviews on the user's account, and\n`get_manuscript_review` to open one.\n\nReport results as returned. Do not add issues, scores, citations or journal\nfacts the tools did not return, invent fixes, or rewrite the manuscript. Keep\nthe author's own interpretation distinct from Manusights' findings.\n"},"changes":[],"summary":"First saved snapshot. No earlier version is available for comparison.","summary_kind":"deterministic","summary_metadata":{}}