archive_space
ChatGPTArchive a Space with all its documents and disconnect any repository sync in it (organization owners only). Delete a Space with this, not delete_folder.
bind_repository
ChatGPTTurn on two-way sync between a Space and a repository path, then import it. Poll get_two_way_sync_run with the returned taskId.
calendar_create_event
ChatGPTCreate a calendar event for the user, one-off or recurring. Pass recurrence instead of creating several events when the user asks for a repeating meeting or names a cadence. On the web app and in Slack this usually produces a confirmation card the user approves first, so the result says "pending_confirmation" until they do. It is written straight away on the API and for users who turned on auto-accept - trust the result status rather than assuming. Resolve dates and times from the conversation before calling (use the current date from the system prompt). Use calendar_find_free_time first when the user asks you to pick a time.
calendar_delete_event
ChatGPTDelete (cancel) a calendar event. Find the event with calendar_list_events first and make sure it is the one the user means. The user normally confirms the deletion before it happens, unless they turned on auto-accept - trust the result status.
calendar_find_free_time
ChatGPTFind free time slots on the user's calendars. Use it to answer "when am I free", to pick a time for a new meeting, or to check whether a proposed time is open. Busy time from all connected calendars is merged; the slots returned fall inside the working hours you pass.
calendar_list_calendars
ChatGPTList the calendars available on the user's connected calendar accounts (Google Calendar today). Each entry carries the connectionId and calendarId that other calendar tools take. Call this before creating an event when the user has several accounts, or when they name a specific calendar.
calendar_list_events
ChatGPTList the user's calendar events in a date range. Use it for questions like "what's on my calendar tomorrow", "when is my next meeting with X", or to find the eventId before updating or deleting an event. Results are sorted by start time across all connected calendars and include each event's connectionId, calendarId, and eventId.
calendar_update_event
ChatGPTChange an existing calendar event: reschedule it, rename it, edit its details or attendees, or change how it repeats. Find the event with calendar_list_events first. Only pass the fields that change. Instances of a repeating event carry a recurringEventId: pass it as eventId to change the whole series, and the instance id to change that occurrence only. Recurrence changes apply to the series, so they must be sent with the recurringEventId. The user normally confirms the change before it is written, unless they turned on auto-accept - trust the result status.
cancel_docs_build
ChatGPTCancel a build that is still waiting in the queue.
check_docs_custom_domain
ChatGPTCheck whether the custom domain is verified. ready means it can be enabled in Falconer settings.
clear_docs_site_password
ChatGPTRemove the docs site password so anyone can read the published site (owners only).
code_search
ChatGPTSearch for code elements by keyword. This tool performs lexical (BM25) matching only; no semantic or embedding-based search is performed. Only 'CodeElement' types are returned. Pass elementType to filter to a specific subtype ('file', 'function', 'class', or 'method'); omit it to search across all subtypes. To read the full content of a specific element by its ID, use the read_file tool and pass the element ID as fileId.
code_tree
ChatGPTDisplay a repository tree. Provide repo, optional level (depth), and optional path as tree root.
complete_upload
ChatGPTFinish an upload started with create_upload after the file bytes were sent to its uploadUrl. Returns the same result as upload_media, including the snippet to insert into document content and, for a placed file, its fileNodeId, navItemId and parentFolderId. Safe to call again for the same uploadToken.
convert_folder_to_space
ChatGPTTurn a top-level company folder into a Space, keeping its contents (organization owners only).
create
ChatGPTCreate a Falconer document. Without parentFolderId or parentDocumentId it lands in your private section, which other people cannot see. When the user names a Space or folder, pass it here instead of moving the document afterwards. Custom instructions may govern the new document: pass customInstructionsHash, the rulesHash get_folder returns for parentFolderId or read returns for parentDocumentId, or for a private document the organization rulesHash from the server instructions. Preserve existing Falconer document structures and reference identities when editing. Before creating or editing document content, call falconer_knowledge with topic "document-format". Load "document-components", "html-components", or "diagrams" before using those features. Load each relevant topic once per conversation. Upload new images, videos and file attachments (PDF, Office, text and code files) and insert the returned snippet. When you can make HTTP requests, use create_upload, send the file to its uploadUrl, then call complete_upload; otherwise use upload_media for small files. Prefer placement library for a file the user should find again: it lands in their Artifacts folder, or under libraryFolder as Artifacts > {Subfolder}. Use placement navigation only when the user names a folder or Space. A placed file is listed, moved, read and deleted like a document, and its snippet still inserts it into content.
create_docs_site
ChatGPTSet up a public docs site on a Space, seeded with starter pages unless syncWithGithub is true. Returns the existing site when there is one.
create_docs_view_folder
ChatGPTCreate a container in a Space's Docs view. Where the Docs view follows the repo files it is also a directory.
create_docs_view_page
ChatGPTCreate a page in a Space's Docs view, with its file in the Files view. Custom instructions may govern the new page: pass customInstructionsHash, the rulesHash get_folder returns for its parent container (the Space's folderId at the top level) or read returns for a parent page.
create_document_revision
ChatGPTCreate an explicit revision of the current Falconer document content.
create_file_view_directory
ChatGPTCreate a folder in a Space's Files view. Its parent folder must already exist.
create_file_view_page
ChatGPTCreate a .md or .mdx page at an exact path in a Space's Files view, optionally with a markdown body. The page shows only in the Files view until publish_file_view_page places it in the site's navigation. Until then only the organization's custom instructions govern it: pass their rulesHash from the server instructions as customInstructionsHash.
create_folder
ChatGPTCreate a Falconer folder.
create_repository_branch
ChatGPTCreate a branch in a connected repository from a base branch, with your personal GitHub account, to bind a Space to.
create_space
ChatGPTCreate a Space (organization owners only). A docs site or a repository sync is set up on a Space.
create_upload
ChatGPTStart uploading an image, video or file attachment of any supported size without sending its bytes through this conversation. Returns an uploadUrl, method and headers: send the raw file bytes there yourself with every returned header exactly as given (for example curl -X PUT -H "Content-Type: ..." -H "Content-Disposition: ..." --data-binary @<file> "<uploadUrl>"), then call complete_upload with the uploadToken to get the snippet to insert. Pass placement library to also place a PDF, Office, image or video file in the navigation, in the Artifacts folder or Artifacts > libraryFolder, where it is listed, moved, read and deleted like a document; the placement is checked here and applied by complete_upload. Requires the ability to make HTTP requests; use upload_media otherwise. uploadUrl expires quickly and must never be persisted or shown in documents.
db_read
ChatGPTRun a read-only SQL query against one of the organization's connected databases (PostgreSQL or MySQL) and get the rows back as a table. The query runs as a restricted database user that can only read the tables (and columns) the organization chose to expose. Before writing a query, if you do not already know the exact table and column names from earlier in this conversation, discover them by running: SELECT table_schema, table_name, column_name, data_type FROM information_schema.columns WHERE table_schema NOT IN ('pg_catalog', 'information_schema', 'mysql', 'performance_schema', 'sys') ORDER BY table_schema, table_name, ordinal_position Reference tables as table_schema.table_name. Only the listed columns are readable, so name columns explicitly when SELECT * is denied. If a table or column you need is not listed, the organization has not exposed it - say so instead of guessing. Rules: - Exactly one SELECT statement per call (WITH ... SELECT and VALUES are fine). Writes, DDL, and multiple statements are rejected. - At most 200 rows are returned (fewer when rows are large), so aggregate (COUNT, SUM, GROUP BY) or filter in SQL instead of pulling raw rows. - Queries time out after 20 seconds. When to use: - Answering questions with live data from a connected database (metrics, lookups, recent records) When NOT to use: - Finding documents, code, or knowledge (use the search tools)
delete
ChatGPTArchive a Falconer document and documents beneath its navigation placements, remove those navigation items, and delete contained library files. A page already published to a synced GitHub repository is instead queued for removal and stays readable until the next push; the response reports which happened. Requires the document ID. Given a file id or file URL instead, deletes that file placed in the navigation together with its nav item.
delete_docs_build
ChatGPTDelete a finished build and its files (owners only). The live build and running builds cannot be deleted.
delete_document_content
ChatGPTDelete exact Markdown text from a Falconer document. Read the document first and copy exact text into old_string. Pass customInstructionsHash when custom instructions govern the document. Preserve existing Falconer document structures and reference identities when editing. Before creating or editing document content, call falconer_knowledge with topic "document-format". Load "document-components", "html-components", or "diagrams" before using those features. Load each relevant topic once per conversation. Upload new images, videos and file attachments (PDF, Office, text and code files) and insert the returned snippet. When you can make HTTP requests, use create_upload, send the file to its uploadUrl, then call complete_upload; otherwise use upload_media for small files. Prefer placement library for a file the user should find again: it lands in their Artifacts folder, or under libraryFolder as Artifacts > {Subfolder}. Use placement navigation only when the user names a folder or Space. A placed file is listed, moved, read and deleted like a document, and its snippet still inserts it into content.
delete_document_revision
ChatGPTDelete a single Falconer document revision.
delete_file_view_file
ChatGPTDelete a file or folder and everything under it from a Space's Files view. Pages in a synced Space are archived and stay until a push removes them from GitHub; elsewhere pages nothing else places are deleted.
delete_folder
ChatGPTRemove a folder and its descendant navigation items, archive documents without placements outside that subtree, and delete contained library files. Documents with other placements are preserved. Disconnects repository sync for folders in the subtree.
detect_repository_docs_config
ChatGPTFind the docs config (such as docs.json) at a base path, which bind_repository with setUpDocsite seeds the site navigation from.
disable_docs_custom_domain
ChatGPTStop serving and billing the custom domain but keep its setup, so it can be enabled again.
disconnect_repository
ChatGPTTurn off two-way sync for a Space. Documents keep their content; pending sync reviews are discarded.
download_skill_package
ChatGPTDownload a native skill as a complete ZIP with SKILL.md and every attached file. Call read first and pass its packageHash. If the package changed, reread the skill. Download the temporary resource link, extract it, and run scripts in your own environment.
edit
ChatGPTLegacy exact replacement for Falconer document content. New clients should prefer replace_document_content, insert_document_content, or delete_document_content. Must read the document first to get exact text to match. Pass customInstructionsHash when custom instructions govern the document. Preserve existing Falconer document structures and reference identities when editing. Before creating or editing document content, call falconer_knowledge with topic "document-format". Load "document-components", "html-components", or "diagrams" before using those features. Load each relevant topic once per conversation. Upload new images, videos and file attachments (PDF, Office, text and code files) and insert the returned snippet. When you can make HTTP requests, use create_upload, send the file to its uploadUrl, then call complete_upload; otherwise use upload_media for small files. Prefer placement library for a file the user should find again: it lands in their Artifacts folder, or under libraryFolder as Artifacts > {Subfolder}. Use placement navigation only when the user names a folder or Space. A placed file is listed, moved, read and deleted like a document, and its snippet still inserts it into content.
export_space_to_repository
ChatGPTWrite a Space bound for export into its repository path in one commit, as you. Re-runnable.
falconer_knowledge
ChatGPTList Falconer knowledge topics by omitting topic, or read a topic by its exact name or page path. Includes document formatting, components, HTML, diagrams, and product documentation. Load document-format before authoring Falconer content and the relevant component topics before using their syntax. Does not require an organization.
get_docs_broken_links
ChatGPTList broken links found by the latest succeeded build of a public docs site.
get_docs_build
ChatGPTGet a public docs build, including its status, search indexing, and publish state.
get_docs_custom_domain
ChatGPTGet the custom domain of a docs site, its status, and the DNS or proxy setup it needs.
get_docs_preview
ChatGPTGet the status and URL of a preview you requested. Stop polling once status is succeeded or failed.
get_docs_redirects
ChatGPTList a docs site's redirects, with warnings for entries the next build would drop.
get_docs_site
ChatGPTGet one public docs site with its publish status and latest build.
get_docs_site_config
ChatGPTRead a docs site's branding, navbar, SEO, social, analytics and footer settings (owners only).
get_docs_view
ChatGPTRead the Docs view of a Space with a docs site or two-way sync: containers and pages with their file paths, plus kind, access, hidden and live URL on a docs site. rootNavItemId reorders the top level. Spaces with neither have no Docs view; use get_nav_tree there.
get_file_conflict
ChatGPTGet both sides of a sync conflict, its block hunks, and the allowed resolutions.
get_file_diff
ChatGPTGet a unified diff of what pushing a synced file would change in the repository.
get_file_view_download_url
ChatGPTGet a URL that downloads a file from a Space's Files view. Stored files get a signed URL that expires; text files come back as a data: URL.
get_folder
ChatGPTGet folder metadata with visible one-level children. When custom instructions govern documents created in the folder, the result opens with them and the rulesHash to pass to create.
get_github_connection
ChatGPTCheck whether your personal GitHub account is connected with the repo scope that pushing synced files needs, and get the URL to connect it.
get_media_download_urls
ChatGPTGet short-lived download URLs for Falconer images and videos the authenticated user can access. Download returned URLs immediately; never embed or persist them.
get_nav_tree
ChatGPTGet a visible Falconer navigation tree.
get_two_way_sync_file_history
ChatGPTGet a synced page's last pushed and last pulled commits and its open pull request.
get_two_way_sync_file_status
ChatGPTGet the repository sync status of one file.
get_two_way_sync_run
ChatGPTGet the status of a bind or retry run, with how many markdown files it has linked so far.
get_two_way_sync_settings
ChatGPTGet the repository, branch, base path and merge strategy of a synced Space.
github_pull_request_get
ChatGPTFetch detailed information about a single GitHub pull request to answer questions about what changed and when. Use this tool AFTER you already know the exact repository and PR number (for example, after calling github_pull_request_search). You MUST provide: - repository: the full repo name, e.g. "FalconerAI/falconer" for PRs like https://github.com/FalconerAI/falconer/pull/2642 - prNumber: the numeric PR id, e.g. 2642 or 2626 The tool returns a JSON string with rich PR details including timestamps (created/updated/closed/merged), who merged it, comment and review comment counts, repository and author info, and a list of files with patches and change stats. Use this when you need to investigate a specific PR’s changes, timeline, and context (for example to understand why something changed, when it shipped, or which files were touched).
github_pull_request_search
ChatGPTSearch the connected GitHub repositories' pull requests live, by repository, state (open, merged, closed), draft status, author, reviewer, labels, base branch, a date range on created, updated, merged, or closed, and words in the title or description. Returns the exact number of matches plus the newest ones with their timestamps, labels, and code change sizes. Use repositories plus state: "open" to count or list a repository's open pull requests. Use authoredBy/reviewedBy with a GitHub username or email, not a person's name. Convert relative dates to absolute YYYY-MM-DD dates in 2026 before calling.
gmail_create_draft
ChatGPTCreate a draft in the user's own Gmail. It never sends: the user reviews and sends it from Gmail. Pass connectionId (from gmail_search results) when several Gmail accounts are connected; the call fails and lists them otherwise. The body is Markdown and is sent as both plain text and HTML.
gmail_read_attachment
ChatGPTRead the contents of a Gmail attachment found via gmail_search. Supports PDFs, Word documents (docx, docm, doc), PowerPoint presentations (pptx, pptm), Excel spreadsheets (xlsx, xlsm, xls, ods; returned as CSV per sheet), OpenDocument files (odt, odp), forwarded emails (eml, msg; rendered with headers and body), RTF, and text files (txt, md, html, csv, tsv, json, xml, yaml, source code, and other text/* types), returned as extracted text, and images (jpeg, png, gif, webp), returned as viewable image content. Identify the attachment by messageId plus the filename from gmail_search results, or by partId when multiple attachments share a filename. Both may be omitted when the message has exactly one attachment. Message IDs are scoped to the Gmail connection that returned them; pass that connectionId whenever the user has more than one account connected. Text output is paginated in pages of 20000 characters: each result includes offset, totalChars, and nextOffset. When nextOffset is not null, call again with offset=nextOffset to read the next page.
gmail_search
ChatGPTSearch Gmail messages with rich filters and thread-aware retrieval. Supports filtering by sender, recipient, date range, labels, category, subject, body text, read status, thread ID, and attachment constraints. Accounts: - The user may have several Gmail accounts connected. Omit connectionId to search all of them; pass it to search one mailbox. - Every thread carries the connectionId and accountEmail it came from. Thread IDs, message IDs, and page tokens are only valid inside that connection, so pass its connectionId when fetching a thread, reading an attachment, or requesting the next page. Thread behavior: - messageSelection='last': only latest message in each thread - messageSelection='all': every message in each thread - messageSelection='range': arbitrary slice via rangeOffset + rangeLimit Results list each message's attachments (partId, filename, mimeType, sizeBytes). To read an attachment's contents, call gmail_read_attachment with the message id, its connectionId, and filename, or partId when multiple attachments share a filename.
google_doc_edit
ChatGPTEdit a Google Doc in place with text replacements. Read the doc with google_doc_read first. textToMatch is copied exactly from google_doc_read's markdown, including backslash escapes, and must match exactly one place in the doc. To insert text, match the preceding text and repeat it at the start of newText. newText is markdown: headings, bullet and numbered lists, bold, italic, code and links. Escape literal *, _, ` and [ with a backslash, and write & as \&. Each line of newText is one paragraph, and a blank line inserts an empty paragraph. A line that should stay plain text but starts with #, -, >, |, < or 1. must escape that marker with a backslash, for example \# Notes or 1\. Notes. An empty newText deletes the matched text. Tables, images and block quotes are not supported. The user approves every edit unless they turned on auto-accept. A requires_confirmation result means nothing changed: say what the edit will do and stop. When your next turn carries the approval, resend exactly the same call. Never claim success before an applied result.
google_doc_find_by_label
ChatGPTList the Google Docs that carry one of your organization's Falconer labels, such as every doc marked Deprecated, searched live in Google Drive. Returns up to 50 docs the connected Google account can see, most recently changed first, with hasMore set when there are more. incomplete means Google searched only some of the shared drives, so tell the user the list may be missing docs. Read any of them with google_doc_read.
google_doc_read
ChatGPTRead a Google Doc live from Google as markdown, with its title and current revisionId. Always read a doc with this tool before editing it with google_doc_edit. The markdown is backslash-escaped (for example snake\_case and \> 5 users). When you call google_doc_edit, copy each textToMatch exactly from this markdown, including the backslashes. Placeholder lines such as [table] stand for content that cannot be edited.
google_doc_update_labels
ChatGPTAdd or remove Google Drive labels on a Google Doc, for example to mark it deprecated. Only the doc's labels change, never its content or title. Use the names of your organization's Falconer labels; google_doc_read lists the ones already on the doc. An unknown name fails with the list of available labels. The user approves the change unless they turned on auto-accept. A requires_confirmation result means nothing changed: say what will change and stop. When your next turn carries the approval, resend exactly the same call. Never claim success before an applied result. If the result says no labels are set up or a label can't be applied, pass on the steps it gives: a Google Workspace admin gives the named account the Applier role, and adds a labeled doc's link under Labels on the Integrations page.
graph_explore
ChatGPTDiscover the organization's knowledge graph: pull requests, issues, projects, teams, people, code, external and Falconer documents, and how they connect. - No arguments: the node and edge types that hold data here (the "graph" skill lists the full schema). - types: properties (with real example values) and edges of those types, needed to write graph_query filters. - nodeIds: what is connected to specific nodes, grouped by edge, with up to 5 neighbors per edge. Use it to look around an entity before writing a query. Only returns what the current user is allowed to see.
graph_query
ChatGPTQuery the knowledge graph by following edges between nodes. Returns one row per node of the returned alias, only for nodes the user may see. Shape: start (one node pattern) -> steps (each follows an edge from the previous node, or from "from") -> return. - Node pattern: { as, type, ids?, me?, nameContains?, where?: [{ prop, op, value }] }. start.type may be omitted when ids are given. - Step: { edge, dir: out|in|any, as, type?, where?, mode?: match|optional|absent, edgeAs?, edgeWhere?, minHops?, maxHops? }. dir "in" walks an edge backwards, e.g. from a person to the issues ASSIGNED_TO them. mode "absent" keeps rows with no such neighbor. maxHops walks hierarchy edges (HAS_PARENT, PARENT_ISSUE, CONTAINS). - ops: eq, neq, in, notIn, contains, startsWith, gt, gte, lt, lte, exists, notExists. Dates accept ISO strings or "-30d", "-12h", "-2w". - A predicate can compare with an earlier alias instead of a value: { prop: "id", op: "neq", ref: "author.id" }. - Find a pull request or issue by number with where [{ prop: "number", op: "eq", value: 9646 }], or by title with nameContains. - return: { alias?, fields?: [props], include?: ["alias.prop"], metrics?: [{ fn: count|sum|min|max|avg, of: "alias" | "alias.prop", as }] }. Metrics group by the returned node, for "who / which has the most" questions. - orderBy: [{ by: metric name or "alias.prop", dir }]. limit up to 50. Example, pull requests I merged in the last 30 days with their reviewers: {"start":{"as":"me","type":"User","me":true},"steps":[{"edge":"IS_USER","dir":"in","as":"account"},{"edge":"AUTHORED_BY","dir":"in","as":"pr","where":[{"prop":"merged_at","op":"gte","value":"-30d"}]},{"from":"pr","edge":"REVIEWED_BY","as":"reviewer","mode":"optional"}],"return":{"alias":"pr","fields":["number","merged_at"],"include":["reviewer.name"]}} Edge names and properties are listed in the "graph" skill; use graph_explore to check real property values or what surrounds a node.
insert_document_content
ChatGPTInsert Markdown content before or after exact anchor text, or at the start/end of the document. Use an anchor copied from the document for before/after. Pass customInstructionsHash when custom instructions govern the document. Preserve existing Falconer document structures and reference identities when editing. Before creating or editing document content, call falconer_knowledge with topic "document-format". Load "document-components", "html-components", or "diagrams" before using those features. Load each relevant topic once per conversation. Upload new images, videos and file attachments (PDF, Office, text and code files) and insert the returned snippet. When you can make HTTP requests, use create_upload, send the file to its uploadUrl, then call complete_upload; otherwise use upload_media for small files. Prefer placement library for a file the user should find again: it lands in their Artifacts folder, or under libraryFolder as Artifacts > {Subfolder}. Use placement navigation only when the user names a folder or Space. A placed file is listed, moved, read and deleted like a document, and its snippet still inserts it into content.
issue_create
ChatGPTCreate an issue in a connected tracker, filed in the requesting person's own name. Supported trackers: github. For Linear, use linear_create_issue. Use it when the user asks to file, open or create an issue, bug or task. - Creation needs the person's own account on that tracker. If a connection card or link is returned, end the turn and wait for them to connect. - For GitHub, the repository must be one the organization syncs. If it is unclear which one, ask, or use the options the tool returns. - Only pass labels and assignees the user asked for. If one does not resolve, the tool lists the valid options; ask the user rather than guessing. - Never call issue_create to change an issue that already exists, including one you just created. To assign, label, retitle, edit or close it, use issue_update. - If the person filed an issue with the same title in the same place in the last few minutes, the tool returns that issue instead of creating another. Share its link, or change it with issue_update; pass allowDuplicate only if the user confirms they want a second one.
issue_get
ChatGPTRead one issue in full from any connected tracker (Linear, Jira, Asana, GitHub Issues): description, status, assignees, team or project, labels, parent and sub-issues, blocking and related issues, linked pull requests, comments, attachments and custom fields. Pass the issue key (e.g. "ENG-123", "PAY-42", "owner/repo#7"), its URL, or an id from issue_search. Use issue_search to find issues. For a GitHub issue, linkedPullRequests lists the merged pull requests that closed it ("closed_issue"), are linked to close it ("closes") or mention it ("mentions"), newest first. Check them before saying what was delivered or whether an issue was closed correctly. Open pull requests are not listed; search pull requests for "#<number>" to find them.
issue_search
ChatGPTSearch issues and tasks across every connected tracker (Linear, Jira, Asana, GitHub Issues) with filters for text, status, assignee, team or project, label, priority, type and dates. Use statusCategory to filter by state in a way that works across trackers: todo, in_progress, done, canceled. Use status only for an exact tracker status name. Only Linear and Jira track in_progress. GitHub issues are only open or closed and Asana sections are free form, so open GitHub issues and unfinished Asana tasks always come back as todo. For active, open or in-flight work, pass statusCategory ["todo", "in_progress"] so every tracker is included. If the user asks specifically for in-progress work, say that GitHub and Asana do not track that state. The query parameter does basic text search, which is unreliable for finding issues by person. Use the assignee parameter with an email address (or a GitHub username for GitHub issues). If you only have a person's name, ask the user for their email first. Use startDate/endDate for time filtering in YYYY-MM-DD format; by default they match any of createdAt, updatedAt and completedAt. The current year is 2026. For a changelog, search for issues completed, created or updated in the window. Use issue_get to read one issue in full, with comments, sub-issues and links.
issue_update
ChatGPTChange an existing issue in a connected tracker, as the requesting person: retitle it, replace its description, close or reopen it, add or remove labels, and assign or unassign people. Supported trackers: github. Use it whenever the user wants to change an issue that already exists, including one just created with issue_create. Never create a new issue to make a change. - If the user names the issue by title or description instead of its key, find it with issue_search first. If more than one issue matches, list each one's key and repository and ask which to change before calling this tool. Never change more than one issue unless the user asked for that. - Before closing or reopening an issue, or suggesting that, read it with issue_get and base the change on its current state and linkedPullRequests, not on older summaries or comments. - Pass only the changes the user asked for; everything else stays as it is. - Changes need the person's own account on that tracker. If a connection card or link is returned, end the turn and wait for them to connect. - If a label or assignee does not resolve, the tool lists the valid options; ask the user rather than guessing.
keep_file
ChatGPTCancel the queued delete of a non-page file so the repository keeps it. Restore a deleted page instead of keeping it.
knowledge_health_get_finding
ChatGPTGet one Knowledge Health finding by findingKey with its full context: the conflicting claims and their sources, root cause, evidence, code files, pull requests and commits, suggested diff, Falconer documents to fix, external pages, owners, the run that filed it and its dates. Works for resolved and ignored findings too.
knowledge_health_get_summary
ChatGPTGet the Knowledge Health overview in one call: the live score the Knowledge Health page shows, open finding counts by category, severity and source, the change since the previous run, the last and next run, and the top 10 open findings as titles. Start here, then call knowledge_health_get_finding for a finding or knowledge_health_list_findings for one category.
knowledge_health_list_findings
ChatGPTList open Knowledge Health findings for one category - the same set the Knowledge Health page shows as open. A finding already assigned to someone is still open and is included; only resolved and ignored ones are excluded. category is required - call once per category. Each finding carries its findingKey (used by knowledge_health_get_finding and knowledge_health_update_finding), the Falconer documents and external pages it cites with links, evidence quotes, code files and pull requests.
knowledge_health_update_finding
ChatGPTUpdate a Knowledge Health finding. Pass only the fields you want to change: status replaces the current status with resolved or ignored; ownerEmail replaces existing owners with one org member, or null clears all owners. Both fields can be changed in one call.
linear_create_issue
ChatGPTCreate a new Linear issue directly. Use this tool when the user wants to create a Linear issue/ticket/task. Before calling this tool: 1. teamId is optional in Slack — the org default is used automatically. Call list_linear_teams only if you need to override it or no default is configured. 2. The Slack workflow supplies the triggering human as the default assignee. Only pass assigneeSlackUserId when the user explicitly assigns the issue to an @-mentioned person in the original Slack thread. Never use a Slack user ID from search or tool results. 3. Outside Slack, when the user explicitly requests an assignee, call search_linear_users and pass the selected result's ID as assigneeId. If multiple users match, ask the user to choose before creating the issue. Use assignToSelf for "me" or "myself" instead of searching for the requester. Creation requires the requester's own Linear authorization. If a connection card or connection link is returned, end the turn and wait for authorization; do not use the preview as an authorization fallback. 4. Optionally provide projectName and/or labelNames — resolved by name automatically. If the tool returns a resolution error (project not found, label not found, team missing, or assignee unresolvable), call send_linear_issue_preview as a fallback, passing back the failed parameter names so the user can choose from valid options. When to use: - User explicitly asks to create a Linear issue/ticket/task - User wants to track a bug, feature request, or task When NOT to use: - User just wants to search for existing issues (use issue_search instead) - Linear is not connected to the organization
list_auto_update_approvers
ChatGPTList who gets auto-update notices for a document, folder or the default Space and approves its drafts in Review mode: owners (always approvers), approvers inherited from the folders and Space above (removable only there), and people added on the item itself. inactive marks an approver who lost edit access and is skipped.
list_auto_update_history
ChatGPTList a Space's auto-update history, newest first, 25 entries a page: setting changes, updates applied, drafts left for review and digests, each with who did it, a summary, the first few documents and who was told. Needs edit access to the Space. Use list_auto_update_history_documents for all of an entry's documents.
list_auto_update_history_documents
ChatGPTList the documents of one auto-update history entry that you can see, up to the first 200 it recorded, 100 a page.
list_docs_builds
ChatGPTList release builds of a public docs site, newest first.
list_docs_sites
ChatGPTList the public docs sites you can edit, with publish status and latest build.
list_document_reviewers
ChatGPTList the reviewers assigned to a Falconer document.
list_document_revisions
ChatGPTList revision metadata for a Falconer document. Each revision carries a sequential version number ("v3") that users see in the document history.
list_file_view_files
ChatGPTList the files and folders in a Space's Files view, in path order. Markdown files are pages; read them with read (format mdx for the repository file).
list_github_repositories
ChatGPTGet a list of all available GitHub repositories in your organization. Use this to discover which repositories contain code before searching.
list_integrations
ChatGPTList every integration this user could have: organization sources (GitHub, Linear, Jira, Notion, Confluence, Slack, ...), personal connections (Gmail, Google Calendar, Granola, and personal Slack, GitHub, and Linear sign-ins), and MCP servers (Sentry, Asana, PostHog, ...). Each entry includes saved accounts, known MCP authentication status, an authorized connect or reconnect link, whether a setup step in Falconer is still owed (with its link), and the latest indexing run for sources that index content. A saved connection does not guarantee usable credentials. Call it when the user asks what is connected, needs data from a source, or wants to connect or reconnect something.
list_linear_teams
ChatGPTList all available Linear teams in the connected workspace. Use this tool to discover team IDs before creating a Linear issue with linear_create_issue. Returns a list of teams with their IDs and names.
list_nav_children
ChatGPTList visible child folders, documents and files for a nav location. A type file item carries file.id, the file id that read, delete and move_nav_item work with.
list_organizations
ChatGPTList Falconer organizations available to this MCP connection.
list_repository_branches
ChatGPTList branches of a connected repository.
list_repository_folders
ChatGPTList one level of a connected repository: folders, markdown files and docs config files, to pick a base path.
list_spaces
ChatGPTList the Spaces you can view, with whether each is repository-synced or has a docs site.
list_sync_repositories
ChatGPTList the repositories the organization connected through the GitHub app, which bind_repository can sync. installUrl adds more.
list_two_way_sync_commits
ChatGPTList repository commits pulled into a synced Space, newest first, read with your personal GitHub account.
list_two_way_sync_conflicts
ChatGPTList synced files that were changed or deleted in the repository while they had Falconer edits.
list_two_way_sync_pull_requests
ChatGPTList open Falconer pull requests from a synced Space that pushes through pull requests.
list_two_way_sync_status
ChatGPTList files in a repository-synced Space with their sync status (modified, added, conflict, pending_rename, ...). Unchanged files are left out unless includeInSync is true.
meeting_notes_search
ChatGPTSearch meeting notes (from Granola and other meeting sources) using semantic and keyword search. Use this tool when the user asks about meetings, decisions made in meetings, action items, what was discussed with specific people, or any information that would come from meeting transcripts or notes. Supports optional date range filters to narrow results to a specific time period.
move_docs_view_items
ChatGPTMove up to 50 items to one place in a Space's Docs view, in the order given; their files move with them. Each result is moved or failed. To move content to another Space use move_nav_items.
move_document
ChatGPTMove a document nav placement by document ID, including into another Space or out of the private section (section company). Use sourceNavItemId when a document has multiple placements.
move_file_view_file
ChatGPTMove or rename a file or folder in a Space's Files view. toPath is the full destination path; its parent folder must exist. Pages keep their navigation placement.
move_nav_item
ChatGPTMove a folder, document or file nav item, including into another Space or between the company and private sections. Provide parentFolderId null to move to the root.
move_nav_items
ChatGPTMove several folders, documents or files to one place in a single call, for example private documents into a Space. Each item reports moved or failed on its own; a failure does not undo the others.
overwrite
ChatGPTOverwrite a Falconer document with complete new Markdown content. Requires the document ID. This rewrites the whole document body and can cause existing comments to lose the text they reference. Use edit for targeted changes when comments should stay anchored. Pass customInstructionsHash when custom instructions govern the document. Preserve existing Falconer document structures and reference identities when editing. Before creating or editing document content, call falconer_knowledge with topic "document-format". Load "document-components", "html-components", or "diagrams" before using those features. Load each relevant topic once per conversation. Upload new images, videos and file attachments (PDF, Office, text and code files) and insert the returned snippet. When you can make HTTP requests, use create_upload, send the file to its uploadUrl, then call complete_upload; otherwise use upload_media for small files. Prefer placement library for a file the user should find again: it lands in their Artifacts folder, or under libraryFolder as Artifacts > {Subfolder}. Use placement navigation only when the user names a folder or Space. A placed file is listed, moved, read and deleted like a document, and its snippet still inserts it into content.
people_search
ChatGPTSearch people and groups in the organization and the memberships between them. Resolves names, emails, Slack user IDs, and GitHub usernames to organization members with their title, profile team, role, and activity. A name that matches several people returns them ordered by closeness to the current user, with the evidence under "proximity" and "ambiguous" when no single match stands out. Lists identity provider (SCIM) groups, their members, and the groups a given person belongs to. Examples: {"query": "Lilian"} finds a person by name; {"group": "Design"} lists members of the Design group; {"target": "groups", "email": "me"} lists the current user's groups; {"target": "groups", "query": "platform"} finds groups by name.
publish_docs_build
ChatGPTMake a succeeded build live (owners only). Pass an older build id to roll back; a build with staleSubdomain still points at the pre-rename host.
publish_document
ChatGPTPublish a private document into company navigation. Use parentFolderId null for company root.
publish_file_view_page
ChatGPTPlace a markdown file that is only in the Files view into the docs site's navigation, at the site root unless parentNavItemId names a container from get_docs_view.
push_all_changes
ChatGPTPush every pending change in a repository-synced Space, including renames, deletes, and non-markdown files. Can take a while on large Spaces.
push_file
ChatGPTPush one file to its repository as you, using your connected GitHub account: a page, or the queued write, rename or delete of any other file.
read
ChatGPTRead a Falconer document by document ID, title, or URL. Attached files are returned as temporary resource links. Pass format mdx to read a docs page as the file two-way sync writes to the repository. Given a file id or file URL, reads a file placed in the navigation instead: its metadata, snippet and a temporary download link. The body of a <Dynamic> or <DynamicInline> widget is live data refreshed in the background; a status="refreshing" attribute on its tag means the value is stale and being reloaded. When custom instructions govern the document, the result opens with them and the rulesHash to pass on writes.
read_agent_tasks
ChatGPTRead cloud coding agent tasks by their chip ids: provider, status (refreshed from the provider when it can report one), link, error, prompt, repository, branch, who started it, and when. Use it to answer how tasks are going or which failed.
read_document_revision
ChatGPTRead the Markdown content for a Falconer document revision.
read_file
ChatGPTRead a specific document or code file when its identifier is already known. Accepts a Falconer /editor/... or /view/... URL, an element ID, or a full path (format: "repo:path/to/file.ts"). Prefer this tool over document search for direct Falconer links and IDs. Returns a paginated excerpt and the total character count; use offset and limit to paginate long content.
read_file_view_file
ChatGPTRead a text or config file (CSS, JSON, YAML, snippets) from a Space's Files view, with the etag to pass to write_file_view_file. Pages are read with read.
read_skill_resource
ChatGPTRead a skill file by its package-relative path, including SKILL.md. Call read first and pass its packageHash; if the package changed, read the skill again. Text is returned in bounded portions. Binary or large files return a temporary download link. Run downloaded scripts in your own environment.
remove_docs_custom_domain
ChatGPTRemove the custom domain from a docs site. A 503 means CloudFront is still releasing it; call again in a few minutes.
remove_docs_view_item
ChatGPTRemove an item and everything under it from a Space's Docs view. mode unlist keeps its files (not where the Docs view follows the repo files); mode delete also deletes its file or directory, archiving synced pages and deleting other pages nothing else places.
reorder_docs_view_children
ChatGPTSet the order of the items under a Docs view container, or under rootNavItemId for the top level. List every child once.
reorder_folder_children
ChatGPTSet the full direct-child order for a Falconer folder. navItemIds must include every direct child nav item exactly once; this is not a partial reorder.
reorder_nav_children
ChatGPTSet the full direct-child order for a folder or document nav item. navItemIds must include every direct child nav item exactly once; this is not a partial reorder.
replace_document_content
ChatGPTReplace exact Markdown text in a Falconer document. Read the document first and copy exact text into old_string. Defaults to the first match; use occurrence: "all" for every exact match. Pass customInstructionsHash when custom instructions govern the document. Preserve existing Falconer document structures and reference identities when editing. Before creating or editing document content, call falconer_knowledge with topic "document-format". Load "document-components", "html-components", or "diagrams" before using those features. Load each relevant topic once per conversation. Upload new images, videos and file attachments (PDF, Office, text and code files) and insert the returned snippet. When you can make HTTP requests, use create_upload, send the file to its uploadUrl, then call complete_upload; otherwise use upload_media for small files. Prefer placement library for a file the user should find again: it lands in their Artifacts folder, or under libraryFolder as Artifacts > {Subfolder}. Use placement navigation only when the user names a folder or Space. A placed file is listed, moved, read and deleted like a document, and its snippet still inserts it into content.
request_docs_preview
ChatGPTBuild a private preview of the docs site from its current content (owners only). Poll get_docs_preview, with the same documentId, until status is succeeded or failed; previewUrl opens in a browser signed in to Falconer.
resolve_file_conflict
ChatGPTResolve a sync conflict read with get_file_conflict. apply_merged takes a local or remote side per changed block hunk.
resolve_file_conflicts
ChatGPTResolve up to 50 sync conflicts by keeping or taking whole files. Returns one result per item.
restore_document_revision
ChatGPTRestore a Falconer document to a specific revision. Pass customInstructionsHash when custom instructions govern the document.
retry_docs_build_index
ChatGPTQueue search indexing again for a succeeded build whose indexing failed or stalled.
retry_two_way_sync
ChatGPTRun a full import of a synced Space again. Poll get_two_way_sync_run with the returned taskId; a null taskId means a run already queued will cover it.
search
ChatGPTSearch Falconer documents by keyword or semantic query.
search_linear_users
ChatGPTSearch active users in the connected Linear workspace. Use this tool before linear_create_issue when a user explicitly requests an assignee outside Slack. Pass the selected result's id as assigneeId. If multiple results are plausible, ask the user to choose instead of guessing. Returns at most 10 users with their Linear ID, name, display name, and email address.
search_memories
ChatGPTSearch team memories — facts, decisions, and knowledge that team members have asked Falcon to remember. Use this when answering questions to check if the team has stored relevant context. Each result carries the memoryId that delete_memory takes.
send_notification
ChatGPTSend a notification to a single recipient, either as an in-app Falconer notification or as a Slack message. Recipients: - channel='falconer': one organization member, identified by recipientEmail. Delivered to their Falconer inbox (and email/Slack if they enabled those in their notification preferences). - channel='slack': slackChannel (post to one public channel) or recipientEmail (notifies that organization member, reaching Slack when they allow it). Set one, not both: when both are given the message goes to the channel only. Add slackThreadTs to reply in the thread of a message in that channel (for an automation's Slack event, its data.threadRootTs), and omit title for a plain reply. Notifying multiple users or channels requires separate calls - only do so when the user explicitly asked for each recipient. Always include a Markdown link to the most relevant destination in the title when one exists (a Falconer document, a PR, a conversation) so recipients can jump straight to it. Only send notifications the user asked for or clearly expects; never use this tool to message people speculatively.
set_docs_redirects
ChatGPTReplace the redirect list. Pass the list you read as base so edits made in between are kept (owners only).
set_docs_site_password
ChatGPTRequire a shared password to read the docs site, or change it. Visitors already signed in stay signed in (owners only).
set_docs_subdomain
ChatGPTChange the docs site subdomain (owners only). The live site moves to it with the next published build.
set_document_status
ChatGPTSet a document’s review-workflow status: draft, in_review, approved, authoritative, or outdated.
sign_out_docs_site_visitors
ChatGPTSign every visitor of a password-protected docs site out; they enter the same password again (owners only).
slack_thread_search
ChatGPTSearch Slack messages using the requesting user's personal OAuth connection, or browse recent channel history. Use this when users ask to find a recent Slack discussion or message. - To search one channel: provide query and channel. - To search across accessible conversations: provide query without channel. By default personal search includes public channels, private channels, DMs, and group DMs permitted by the granted scopes. - For questions about what a person said or requested in a period: resolve their Slack ID with people_search, then set senderSlackUserId and startDate/endDate. Initially omit keyword guesses such as their name, raw ID, or "today". A sender filter finds messages they authored, not mentions of them. - Resolve relative dates in the requesting user's timezone. For today, set both dates to the current local date. Read the local timestamps of matches and context separately. - If nextCursor is returned, reuse the same filters and pass cursor to continue when needed. - To get recent top-level posts in a channel, newest first: provide channel without query. For the latest post, set limit to 1. - To list joined public channels: omit query and channel. - To read one selected thread: provide threadUrl. - If the user identifies a channel as private, set channelVisibility to 'private'. Search and recent-post requests automatically use the requesting user's personal Slack connection in both web chat and Slack, including channels Falcon has not joined. Nonpublic results are available in web chat and one-to-one Slack DMs with Falcon; shared Slack conversations search public channels only. Missing scopes can further limit coverage. If the connection is missing, the web user sees a connect card; follow the returned instructions and wait for the connection outcome. In Slack, provide the returned connection link. Do not ask the user to choose credentials or invite Falcon for these requests. Read coverage and limitations before answering. Say "I found no matching requests in the searched conversations" instead of asserting that someone made no requests or that these were their only messages. Disclose relevant excluded conversation types, remaining pages, and truncated evidence. Contextual messages are not filtered matches or a complete thread. Explain actual connection or operation limits; do not invent Slack API or privacy restrictions. Explicit thread reads still require Falcon to have joined the public channel and return at most 15 messages. Only read a thread when the request needs its replies; a search result is sufficient to identify a message. A thread-read limitation does not mean personal search is unavailable.
start_agent_tasks
ChatGPTStart cloud coding agent tasks (Claude Code, Codex, Cursor, or Falconer) as the person, one per item, and get back the chip for each. Use it when the person asks to start, run, or launch agent tasks, including one per table row. Each task's prompt is its instruction followed by its context lines. Write each started task's chip into its row afterwards with a table set_cells edit: {"ref": "agent_task", "id": taskId, "name": name}. Up to 3 tasks follow the person's auto-accept setting; more always need their approval. A requires_confirmation result means nothing started.
store_memory
ChatGPTRemember something for the user: a memory (fact, decision, information) or a skill (process, procedure, how to do something). Both need a scope (private, team, company); without one the tool returns needs_scope instead of storing. Skills are saved as skill documents with an activation rule.
trigger_docs_build
ChatGPTStart a release build; poll get_docs_build until it finishes. autoPublish (owners only) publishes on success.
unpublish_docs_site
ChatGPTTake a public docs site offline (owners only). Builds are kept.
update_auto_update_approvers
ChatGPTAdd or remove auto-update approvers on a document, folder or the default Space. Needs edit access to it (organization owner for the default Space), and each added person must be an organization member who can edit it. Returns the full list afterwards.
update_docs_site_config
ChatGPTChange docs site settings; null clears a field. Logo and favicon are file paths in the Space's Files view. Changes go live with the next build (owners only).
update_docs_view_item
ChatGPTRename a Docs view container, rename the file or directory behind an item, or set a docs site container's kind, options, access, or hidden flag (owners only for docs sites).
update_document_permissions
ChatGPTUpdate document permissions, including public access. Direct permission changes on inherited documents switch them to explicit mode automatically. User grants with sendNotification enabled may deliver email or Slack notifications that cannot be recalled by revoking access.
update_document_reviewers
ChatGPTAdd or remove document reviewers. Use add/remove with userId or email, or removeAll to clear. removeAll is applied before remove and add. Adding a reviewer grants comment access if needed and may send email or Slack notifications that cannot be recalled by removing the reviewer.
update_folder
ChatGPTUpdate a Falconer folder title or description.
update_space
ChatGPTSet auto-update or its mode for every document in a Space that does not restrict editing with its own sharing: true or false, or null to hand it back to the default. Documents can still set their own value. Needs edit access to the Space. Every auto-update change needs confirmed: true. Call without confirmed first: nothing changes and the result is code confirmation_required with a preview of how many documents turn on or off, change mode, keep their own value or are not reached, and how many people are notified. Show that preview to the user and call again with the same arguments and confirmed: true only after they agree.
update_two_way_sync_settings
ChatGPTChange how pushes from a synced Space land in the repository.
upload_file_view_file
ChatGPTUpload a file (image, font, any asset) to a path in a Space's Files view, base64 encoded, at most 5 MB. Pass replace to overwrite an existing file; otherwise an existing path is refused.
upload_media
ChatGPTUpload an image, video or file attachment to Falconer and return a snippet that can be inserted into document content. Attachments (PDF, Word, Excel, PowerPoint, text, Markdown, HTML, CSV, TSV, ZIP, JSON and JavaScript, TypeScript, Python or Java source files) return an <a data-id="referenceId"> reference that renders as a file pill and appears in the document read response attachments. Pass placement library to also place a PDF, Office, image or video file in the navigation, in the Artifacts folder or Artifacts > libraryFolder, where it is listed, moved, read and deleted like a document. The file travels as base64 inside this call, so only use it for small files or when you cannot make HTTP requests; otherwise use create_upload.
write_file_view_file
ChatGPTWrite a text or config file in a Space's Files view, creating it when the path is new. Pass ifMatch with the etag from read_file_view_file so you never overwrite an edit made since. In a synced Space the change is queued for the next push. Pages are edited with overwrite or the document edit tools.