Turnfeed

Use this when the user asks to like or unlike a Turnfeed post. It changes the user's like state for the selected post; likes are Turnfeed social signals that can be visible in the app and notifications. Human-readable target fields must identify the post for confirmation; a raw post id is only a fallback disambiguator.

like_post

Use this when the user asks to like or unlike a Turnfeed reply. It changes the user's like state for the selected reply; likes are Turnfeed social signals that can be visible in the app and notifications. Human-readable target fields must identify the reply for confirmation; a raw reply id is only a fallback disambiguator.

like_reply

Use this when the user makes a bare open/start/launch request such as 'Open Turnfeed'. It returns the live default Turnfeed feed immediately. Show the returned feed exactly as instructed and end after the feed; do not add a setup explanation, permission notice, viewer identifier, example prompts, question, offer, or other preamble/follow-up. Do not use it for inbox, thread, profile, search, or public writes.

start_turnfeed_chat

Read the signed-in user's Turnfeed inbox: notifications and notification activity, replies, mentions, likes, follows, what needs their attention, whether anyone wrote to them, or following/blocked lists. This is account activity and relationship controls, not private DMs. Use existing returned context when sufficient for a factual answer or summary. For fresh reads, notificationOrder=latest means chronological events; latest follows returns follow events. Ordinary inbox reads use attention order: replies and mentions first, then other activity. Preserve returned text for ordinary display; answer questions naturally from structured data, distinguishing total, visible/matching, and returned-page counts. Initial all-item requests authorize following exact nextCursor with the same filters/order until hasMore=false; disclose partial coverage and never infer missing pages. Use relationshipKind=following or blocked for a fresh viewer-bound targetRef when an authorized unfollow or unblock needs it. Hide targetRef and cursors; omit unsolicited notices or next-step offers.

open_turnfeed_inbox

Use this when the user asks to pin or unpin one of their Turnfeed posts. It changes which post appears pinned on the user's profile. Human-readable target fields must identify the post for confirmation; a raw post id is only a fallback disambiguator.

pin_post

Use this when the user explicitly asks to post or publish supplied final text on Turnfeed, including "Post this on Turnfeed: <text>". Publishes that exact text with visibility=public and the user's Turnfeed profile after ChatGPT's native approval. A fresh clientId identifies each new post; reuse that clientId when retrying the same post. For a quote post, first read the exact ordinary public source thread, then pass its postId and the returned human-readable source label, author, text, and timestamp in the quote fields so native approval identifies the source. Quote commentary is required. Group posts, polls, followers-only posts, and quote chains are unsupported. A source from a feed digest requires get_thread_context before quoting. Do not use for drafts, replies, intent-only requests without final post text, or text that expresses intent to kill or seriously harm someone. Post text is limited to 600 characters.

create_post

Use this when the user clearly requests publishing exact public reply text the user supplied to a readable Turnfeed post, especially after get_thread_context returned human-readable targetArguments. If the target came from get_feed_digest, call get_thread_context with the retained postId before using this tool. Requires an explicit user publish request with the exact target and text, plus any approval required by ChatGPT. Reply text is limited to 600 characters. The reply targets the supplied author/content and publishes the exact reply text with visibility=public under the user's public profile name/handle, visible to people using Turnfeed. Do not use this tool for intent-only commands such as "I want to reply"; open the reply composer or ask for the exact reply text instead. Do not use this tool for text that expresses intent to kill or seriously harm someone; refuse or help rewrite safely instead. Match the target from the human-readable fields; do not rely on raw ids. Requires clientId for retry-safe duplicate suppression.

publish_public_reply_to_post

Use this when the user clearly requests publishing exact public reply text the user supplied to a readable Turnfeed reply, especially after get_thread_context returned human-readable targetArguments for a visible reply. Requires an explicit user publish request with the exact target and text, plus any approval required by ChatGPT. Reply text is limited to 600 characters. The reply targets the supplied author/content and parent context, and publishes the exact reply text with visibility=public under the user's public profile name/handle, visible to people using Turnfeed. Do not use this tool for intent-only commands such as "I want to reply"; open the reply composer or ask for the exact reply text instead. Do not use this tool for text that expresses intent to kill or seriously harm someone; refuse or help rewrite safely instead. Match the target from the human-readable fields; do not rely on raw ids. Requires clientId for retry-safe duplicate suppression.

publish_public_reply_to_reply

Use this when the user asks to report a Turnfeed post for moderation review. It sends the selected reason and creates a private moderation record retained independently of routine target deletion; Turnfeed exposes no dedicated report-withdrawal tool. Human-readable target fields must identify the post for confirmation; a raw post id is only a fallback disambiguator.

report_post

Use this when the user asks to report a Turnfeed reply for moderation review. It sends the selected reason and creates a private moderation record retained independently of routine target deletion; Turnfeed exposes no dedicated report-withdrawal tool. Human-readable target fields must identify the reply for confirmation; a raw reply id is only a fallback disambiguator.

report_reply

Use this only when the user explicitly asks to reset their Turnfeed profile and activity while keeping the same Turnfeed/Auth0 identity. Do not use it to delete or unlink an account. For 'delete all my posts', use open_turnfeed_feed with preparePostDeletion=all_my_posts, display its scope and 'Delete these posts?', and end the turn. Only after a subsequent user message confirms that scope may delete_post use the returned arguments for ChatGPT's native approval; never reset the account for a posts-only request. Reset clears ordinary profile details, posts, replies, likes, follows, the user's own blocks and mutes, hidden words, notification settings, and report history. Blocks set by other people remain under those people's control. Some moderation, safety, legal, and backup records may remain as described in the privacy policy.

reset_me

Read targeted Turnfeed search/filter results, a specific public thread, a public profile, or the signed-in user's own profile and posts. Returns actual content and structured source data for factual questions, summaries, comparisons, translations and private drafts. targetText accepts the original request verbatim; targetKind=thread also supports authorName and postQuery for a member and the beginning or wording of a post. targetKind=profile with profileScope=self resolves profile fields from verified Turnfeed sign-in; authorScope=self selects that user's posts. Author, feedQuery, timeRange and focus can be combined. timeRange.basis distinguishes publication time (created) from visible thread activity (activity); pastHours is rolling, while since/until have explicit timezone boundaries. Full thread content supports analysis of truncated feed previews. hasMore and nextCursor describe pagination; continuation requires the exact cursor and original profile, author, topic, timeRange and focus selectors. Full/all requests cover all matching pages; incomplete retrieval remains partial. viewerOwnsTarget=true identifies verified ownership. Self-following is unsupported; selfFollowMessage explains that state. A visible recent post's postId can identify a handle-less member for a separate relationship action. An untargeted request such as 'Just show me the feed' returns the same real digest as get_feed_digest. Notifications and attention are available through open_turnfeed_inbox. preparePostDeletion=all_my_posts, without other selectors, returns the signed-in user's exact post/reply deletion scope and snapshot-bound deleteArguments. The preview leaves stored posts unchanged. The deletion workflow has separate preview and confirmation turns: the preview displays its exact scope and 'Delete these posts?', then a subsequent user message confirms that scope before the separate delete_post operation. The preview request and snapshot token do not represent that later confirmation. Publishing and deletion are performed by separate write tools.

open_turnfeed_feed

Use this when the user asks for an untargeted feed/post list such as 'Just show me the feed', catch-up, active, recent, interesting, reply-worthy, newest, latest, or chronological posts. Call it immediately in the same turn (focus=active by default; focus=latest only for explicit newest/latest/chronological), then show ordinary feed requests plainly in order. For time requests, set timeRange with pastHours or explicit since/until timestamps including timezone. Use basis=created for posts published during the interval and activity for conversations active during it. Calendar days require known timezone boundaries; ask only if those cannot be determined. Do not search an entire question as a topic. Pass targetText for natural-language fallback. When the user asks for more, call it again with the exact previous nextCursor and the same focus and original timeRange or activityWindow. Never display the cursor. An explicit all/complete request authorizes paging to hasMore=false; otherwise describe only the retrieved portion. For summaries, comparisons, translations or factual questions, answer naturally from structured data and cite the relevant authors/posts. Read full thread context before analyzing truncated text. Private drafting or reasoning over already available data needs no tool call or publication. For ordinary display, end after the returned feed content or empty state; do not append an unsolicited question, offer, or suggested next step. Never promise a later fetch, delay the read for confirmation, or answer from opening text or memory. Use open_turnfeed_inbox for personal notifications/attention.

get_feed_digest

Use this when the user explicitly asks to save exact supplied fields on their public Turnfeed profile. Call it exactly once with only the supplied displayName, handle, bio, websiteUrl, or explicitly requested avatarUrl plus visibility=public. ChatGPT controls permission checks and user approval. Omitted fields remain unchanged. Never use this for reading a feed, searching posts, or any request that does not explicitly ask to change the user's own profile. Supplied profile fields can be visible to people using Turnfeed and attached to the user's public posts, replies, likes, follows, and profile page.

set_profile