Meta MuseUndisclosed

A friends-only feed from his own DMs: the author had Muse work out who to include and build a dashboard of their posts and stories, though images could not be fetched yet

Shivam BharukaX (Twitter) · 2026-09-14

Read original

I deleted Instagram from my phone a couple of years ago because so much of the feed felt irrelevant and distracting. But one thing I missed was seeing updates from my friends and family. So I asked @Muse to build me a personal feed just for that. It used my DMs to figure out who to include and put together a dashboard of their posts and stories. (It couldn't fetch the actual images yet due to limitations in the Instagram capability). But overall I am starting to really like the ability to create these mini-apps. I also had it build a dashboard of everything its keeping an eye for me. We have seen a few products that let people create mini-apps. But I think putting that capability inside a personal assistant makes it much more useful. It already has the context to suggest what I might actually use.

Business model

Shivam Bharuka writes that he deleted Instagram from his phone a couple of years ago because most of the feed felt irrelevant and distracting, but he missed seeing updates from friends and family. So he asked Muse to build a personal feed just for that: it used his DMs to work out who to include, then put together a dashboard of their posts and stories. He flags one limitation himself - the images could not be fetched yet because of limits in the Instagram capability. He also mentions a second dashboard of everything it keeps an eye on for him, and says he is starting to like making these mini-apps. The post does not say how many people were included, what rule decided inclusion, how often it refreshes, what permissions were granted, where the data lives, what the dashboard shows when an image is missing, and there is no link or screenshot. What it shows is that the dashboard was built and the list was inferred from DMs - not that it refreshes correctly, and not that the list is accurate. All of this is his account; AgentFlow connected no social account and never saw the dashboard.

The point is not another feed but moving the question of who counts as close back from the algorithm to the person, using evidence that already exists - DMs - rather than a model's hunch. What can be delegated is inferring candidates, fetching public content and assembling the page. What cannot is finalising the list, because a wrong inclusion puts someone's posts in front of the wrong reader, and the platform limits mean the dashboard may only ever be text.

Workflow

  1. Set the scope and the exclusions first: which relationship groups to keep (family, close friends, colleagues, watch-only) and who must never appear. Write the rule, not the names, so the model does not invent a list.
  2. Infer candidates from existing DMs and present them as candidates only: for each person, the evidence - last interaction, number of exchanges, whether replies went both ways. Every row is marked for your confirmation.
  3. You go through the candidates one by one and keep or remove each. Anyone removed leaves the fetching scope immediately, with no cached content kept. The list is only final after that pass.
  4. Build the read-only dashboard from the final list: person, content, time, original link. No engagement data, no likes, no comments, no messages, private by default.
  5. Mark what cannot be fetched. If the platform will not return images, the row shows text and a link with images not available - never a placeholder, an old image or an inferred caption.
  6. Refresh, deduplicate and allow exits: merge multiple posts per person, sort by time, log each refresh with successes and failures, and once someone is removed they stop appearing from the next refresh and an exit record is kept.

Adaptation plan

The post does not say how many people were included, what rule decided inclusion, how often it refreshes, what permissions were granted, where the data is stored, what the dashboard shows when an image is missing, or whether it is still running. There is no link and no screenshot. AgentFlow connected no social account, read no DMs and never saw the dashboard.

Prompt

AgentFlow draft · Untested.

You are my friends-and-family feed assistant. You propose candidates, fetch public content, assemble the dashboard and label what is missing. I finalise the list, the permissions and any removal.

INPUT - WHAT I WILL PROVIDE
1) The relationship groups I want to keep (family, close friends, colleagues, watch-only) and anyone who must never appear.
2) The read-only scope I authorise: public posts and stories only, no engagement.
3) The fields the dashboard should show (person, content, time, original link) and how often to refresh.
4) What I prefer when an image cannot be fetched: text only, or a blank row marked as such.

EXECUTION STEPS
1. From existing DM history, find people with frequent two-way exchanges inside the groups I named, and output candidates with evidence: last interaction, number of exchanges.
2. Every candidate is marked for my confirmation. Anyone I remove leaves the fetching scope immediately.
3. After the list is final, fetch public posts and stories per person and keep only four fields: person, content, time, original link.
4. Do not collect engagement data, like, comment, message or follow. The dashboard is private by default and gets no shareable link.
5. Where a field cannot be fetched, mark it not available and show text plus link. Never substitute a placeholder image or old content.
6. Merge multiple posts per person, sort by time, and log each refresh with the time, the number of items and the failures.
7. Keep an exit record: who was removed and from which refresh it took effect.

OUTPUT FORMAT
- Candidates: person / evidence (last interaction, exchanges) / status (to confirm, kept, removed)
- Dashboard: person / content summary / time / original link / missing fields
- Missing list: item / what is missing / how it is handled
- Refresh log: time / items / failures
- Exit record: person / effective from
- For me alone: finalise the list / adjust permissions / decide removals

HUMAN CONFIRMATION
I confirm every name, decide whether the dashboard stays private, and decide removals. You never treat candidates as a final list, never publish or share the dashboard, never interact on my behalf, and never turn could not fetch into nothing new.

FAILURE AND MISSING DATA
If the DMs are not enough to judge, keep the person as a candidate marked insufficient evidence rather than guessing. On rate limits, log the failure and retry at most twice. If images are unavailable, show text only. If a removed person still sits in a cache, delete it immediately and record it in the log.

Running costs and pitfalls

The post does not say how many people were included, what rule decided inclusion, how often it refreshes, what permissions were granted, where the data is stored, what the dashboard shows when an image is missing, or whether it is still running. There is no link and no screenshot. AgentFlow connected no social account, read no DMs and never saw the dashboard. No independent cost benchmark is available. Check subscriptions, permissions and external actions before running a workflow.

Author-reported results are not independently audited.

Original text and media remain attributed to their authors.