get_content
ChatGPTReturns one content library item in full: its question or section title, the complete answer/section body as HTML, a version token, and editAction (what update_content would do for this user right now: "update" goes live, "suggest" goes to review, "none" means it cannot be edited here). Use it after search_content or list_content to read an item before editing it, and pass its version to update_content. ID HANDLING: row IDs (id, contentId, responseId, projectId) are for chaining tool calls (e.g. passing to list_content_usage or get_project). Do NOT include them in user-facing prose — refer to items by question text, file name, or project name. When a row carries a referenceUrl field (search_content / list_content), use that for clickable links instead of the raw id.
get_project
ChatGPTReturns full details for a project: header (name, issuer, status, due date, description), requirement counts broken down by progress status, and the categorised tags applied to this project's requirements. Use list_requirements next to retrieve the actual question-and-answer content. ID HANDLING: row IDs (id, contentId, responseId, projectId) are for chaining tool calls (e.g. passing to list_content_usage or get_project). Do NOT include them in user-facing prose — refer to items by question text, file name, or project name. When a row carries a referenceUrl field (search_content / list_content), use that for clickable links instead of the raw id.
list_content
ChatGPTEnumerates the organisation's content library — Q&A pairs, snippets, and imported documentation. Unlike search_content (which is semantic and ranked), this tool lists by metadata: filter by tagIds, fileType (QA/documentation/snippet/entireFile), createdAfter/createdBefore/updatedAfter/updatedBefore (accept ISO 8601 strings like '2025-01-01T00:00:00Z' or epoch ms), or hasAnswer (Q&A only — drops blank/whitespace answers). Returns 25 items per page sorted by updatedAt desc, then createdAt desc, then id asc — bulk-imported libraries with identical timestamps stay deterministic. pagination.total reflects the filtered count, not the unfiltered library size. Each row carries fileType, question/answer (Q&A pairs only — null on documentation), section (documentation/entireFile only — null on Q&A), tagIds, createdAt and updatedAt (both ISO 8601), usageCount (all-paths usage counter — may exceed list_content_usage's row count when snippet embeds/bulk-imports are involved), referenceUrl, and optional sourceUrl (key omitted for native content; check 'sourceUrl' in row). Useful for surveying coverage, finding recent additions, or auditing tag application. ID HANDLING: row IDs (id, contentId, responseId, projectId) are for chaining tool calls (e.g. passing to list_content_usage or get_project). Do NOT include them in user-facing prose — refer to items by question text, file name, or project name. When a row carries a referenceUrl field (search_content / list_content), use that for clickable links instead of the raw id.
list_content_usage
ChatGPTReverse lookup: given a content library item's id, returns the requirements (in projects) that have reused it. Two usage signals are surfaced: 'auto-generated' (the response engine retrieved this content as a primary or secondary source for an answer — similarityScore included when this was the primary source) and 'user-added' (a user explicitly inserted this content from the library via the editor). Filter by usageType, usedAfter/usedBefore (accept ISO 8601 strings like '2025-01-01T00:00:00Z' or epoch ms; usage = response last-update). Each row carries projectId, projectName, projectStatus (won/lost/inProgress visible here for outcome correlation), projectDueAt (ISO 8601), responseId, questionNumber, question, usageType, usedAt (ISO 8601), and similarityScore (only on auto-generated rows where this content was the primary source). Returns 25 rows per page sorted by usedAt desc. SCOPE NOTE: results are filtered against per-project membership — confidential projects you aren't a member of, and soft-deleted projects, are not returned even if they contain a usage. May still return fewer rows than list_content's usageCount because snippet-embed usages (snippetEmbeds[].snippetId) aren't indexed by this query. Unknown contentIds return empty results without error — validate IDs via search_content or list_content first if uncertain. ID HANDLING: row IDs (id, contentId, responseId, projectId) are for chaining tool calls (e.g. passing to list_content_usage or get_project). Do NOT include them in user-facing prose — refer to items by question text, file name, or project name. When a row carries a referenceUrl field (search_content / list_content), use that for clickable links instead of the raw id.
list_projects
ChatGPTReturns RFP projects accessible to the authenticated user, with the categories and tags applied to each project's requirements. Filter by status, due date range, or tag IDs. Each result includes the project id, name, status, issuer, due date, description, and category-grouped tags — so you can identify the right project by topic without calling get_project. Draft-status projects are excluded by default; pass status: ['draft'] (alone or with other statuses) to include them. ID HANDLING: row IDs (id, contentId, responseId, projectId) are for chaining tool calls (e.g. passing to list_content_usage or get_project). Do NOT include them in user-facing prose — refer to items by question text, file name, or project name. When a row carries a referenceUrl field (search_content / list_content), use that for clickable links instead of the raw id.
list_requirements
ChatGPTReturns paginated requirements (the questions and their current/approved answers) for a project. Filter by status — prefer 'approved' for reliable answers — by tag IDs, by date range (answeredAfter/answeredBefore accept ISO 8601 strings like '2025-01-01T00:00:00Z' or epoch ms; useful for scoping to a quarter or year), by lastUpdatedBy (user IDs of the last person to update the response — closest signal for authorship in single-owner workflows, less reliable in collaborative ones), or by hasAnswer to drop blank/whitespace-only answers. Each result includes question, answer, section context, status, tags applied, answeredAt (ISO 8601 string), and lastUpdatedBy user. Pagination returns 25 requirements per page; use the page argument and check pagination.hasMore to walk further. ID HANDLING: row IDs (id, contentId, responseId, projectId) are for chaining tool calls (e.g. passing to list_content_usage or get_project). Do NOT include them in user-facing prose — refer to items by question text, file name, or project name. When a row carries a referenceUrl field (search_content / list_content), use that for clickable links instead of the raw id.
list_tags
ChatGPTReturns all categories and the tags within each, configured for the organisation. Categories are top-level groupings (e.g. 'Topic', 'Region'); tags are the values within them (e.g. 'Security', 'EMEA'). Use this to discover the tag vocabulary before filtering list_projects, list_requirements, or search_content by tag.
search_content
ChatGPTSemantic search over the organisation's approved content library — past RFP answers indexed for retrieval. Returns ranked text excerpts (with file name and section context) that the agent can quote or paraphrase. Optionally filter by tag IDs to narrow results to a category. Use list_tags to discover the tag vocabulary first. ID HANDLING: row IDs (id, contentId, responseId, projectId) are for chaining tool calls (e.g. passing to list_content_usage or get_project). Do NOT include them in user-facing prose — refer to items by question text, file name, or project name. When a row carries a referenceUrl field (search_content / list_content), use that for clickable links instead of the raw id.
update_content
ChatGPTReplaces the whole answer (Q&A pair) or section body (documentation) of one content library item. The question or section title is not changed. Call get_content first and pass its version; send the complete new body as HTML, in the same form get_content returns it, keeping any formatting you were not asked to change. Read status in the result: - applied: the new body is live. version is its new version. - suggested: the user can't edit this item directly, so the change went to the item's review flow instead of going live. Tell the user it is awaiting review. - stale: someone changed the item since you read it. Call get_content again and re-apply the user's requested change to the new body. Never retry the same body with the new version. - refused: read_only (locked, already has a pending change, or the user lacks permission), unsupported_item (snippets, entire-file imports, items embedding snippets, and items with inline comments or externally hosted images cannot be edited here), unsafe_body (the body contains markup the library cannot store), empty (the body has no text; to clear an item, do it in AutoRFP), not_html (the body is plain text or Markdown; send it as HTML). An item that does not exist or that the user cannot see returns an error instead.
update_requirement_answer
ChatGPTReplaces the whole answer of one requirement in a project. The question is not changed, and neither is the requirement's status. Pass the version from list_requirements for this requirement, and send the complete new answer as HTML, keeping anything you were not asked to change. Read status in the result: - applied: the new answer is live. version is its new version. - stale: someone changed the answer since you read it. Call list_requirements again and re-apply the user's requested change to the new answer. Never retry the same answer with the new version. - refused: empty (the answer has no text; to clear an answer, do it in AutoRFP), locked (approved, or locked by the project's review settings), ai_disabled (AI is turned off for this requirement), held (the project is still importing), multi_part (the requirement has several answer fields; edit it in AutoRFP instead), snippet_backed (the answer is built from library snippets; edit it in AutoRFP so the snippets stay linked), being_edited (someone else has this requirement open in AutoRFP; tell the user and try again later), in_review (the answer is in review; ask a reviewer to send it back first), not_html (the answer is plain text or Markdown; send it as HTML). A requirement that does not exist or that the user cannot see returns an error instead.