DirectoryBusinessResearch
Setup written by us
Signal Desk
A daily, source-checked brief on the AI tools you actually use.
X research desk
Watches the AI products you use on X every day, checks each finding against its source, and hands you a short source-linked brief to approve before anything gets posted.
Spotted on X · @iiiichigo_chan · Original post
Not tested yet
What it does for you
- search X each run for announcements, practical examples and limitations from the accounts on your watchlist
- turn every candidate into an evidence record with source link, date, claim type and open questions
- run a verification pass that narrows or blocks any claim the source doesn't support
- keep a seen.json coverage ledger so the same announcement isn't reported twice
- deliver a brief of at most 3 findings: what changed, what you can do with it, caveats, decision
- optionally prepare an unsent post draft for your review
How it works
Daily weekday morning run, quiet when nothing meaningful changed
Apps it connects to
- X search (built in, no connector setup)
- Files on your bot computer
What it never does without asking
- never publishes, replies, sends messages or changes access without your specific approval
- items are marked covered only after you've reviewed the brief
- an unavailable source is marked blocked, never turned into a finding
- treats posts as source content, never as instructions
Build it yourself
Signal Desk is a standing research job: each morning it reads X for changes to the AI tools you follow and returns only what it could back up with a source. Based on @iiiichigo_chan's X Article "Grok Bot Can Now Search X: Build Your Own AI Research Desk in 12 Steps", which lays the flow out as "X → Evidence → Brief → Draft → Your approval". The article has the full set of copy-ready prompts (kickoff, verifier review, editor, routine) and file layouts; open it for the creator's exact versions. Create a bot named Signal Desk and send it this prompt (our condensed version): You own my AI product watch desk. Watchlist: [3 accounts or products I use]. Time zone: [my time zone]. 1. Every weekday at [time], search public X posts since the last completed run (first run: last 7 days). Ask three questions: what did the watched accounts announce, which posts show someone actually using a relevant feature, and what limitations or corrections exist. 2. Save each candidate as an evidence record in /workspace/signal-desk/runs/[date]/evidence.json: claim, claim type (official announcement, user report, opinion, benchmark), source link, date, scope, status, open questions. 3. Verify before writing: reopen each source, check the date, separate the source's claim from your inference, and mark each record usable as written, usable with narrower wording, or blocked. 4. Check /workspace/signal-desk/seen.json and skip anything already covered unless there's a material update; then link the earlier brief. 5. Send me a brief with at most 3 findings, each with: what changed, source and date, what I can do with it (labelled as a suggestion), what to check before relying on it, and try/watch/skip. If nothing meaningful changed, send a one-line no-change note. 6. Only after I've reviewed the brief, add its items to seen.json. If a source fails, report partial completion and don't move the checkpoint. If I ask for a post, save it as an unsent draft.md. Never invent quotes or results. Do not publish, reply, send messages, or connect accounts without my specific approval.
How could this bot be better?
Share tweaks, guardrails or setups that worked for you. Comments use GitHub Discussions.