@RevenueCat_agad - Twitter Agent

An LLM-powered tool-use agent that searches RevenueCat documentation, composes developer-focused tweets with citations, runs a critic quality check, and posts to @RevenueCat_agad via Twitter API v2.

Topics can be specified manually or selected from a predefined list. The agent searches docs, drafts the tweet, and a critic agent evaluates quality before posting.

How It Works

Topic Receives a topic (user-specified or selected from a predefined list of RevenueCat developer topics).

➔

Research Searches ingested docs via BM25, checks past tweets to avoid repetition.

➔

Draft Composes tweet text grounded in documentation, selects a hashtag from a curated map.

➔

Compose Writes a tweet grounded in real documentation. Starts with the developer's problem, not the product.

⬇

Post Posts to @RevenueCat_agad via Twitter API v2 with OAuth 1.0a. Threads are posted as reply chains.

➔

Self-Evaluate Critic agent runs programmatic checks (length, code detection) then LLM quality review. Rewrites up to 3 times if rejected.

If step 5 fails quality check, the agent loops back to step 4 and rewrites.

Tools Available to the Agent

Tool	What It Does
`search_docs`	BM25 search over 326 ingested RevenueCat doc pages
`fetch_url`	Fetch any URL: doc pages, Reddit posts, blog posts
`scan_github_issues`	Scan 6 RevenueCat SDK repos for recent open issues
`scan_reddit`	Search 6 developer subreddits for subscription-related posts
`get_past_interactions`	Check past tweets from the database to avoid repetition
`submit_tweet`	Submit and post the final tweet
`submit_thread`	Submit and post a multi-tweet thread as reply chain

Tweet Activity

These tweets were generated by the agent's LLM tool-use loop. Tweets marked as "sent" were posted via Twitter API v2 to the @RevenueCat_agad account. Drafts are shown for review but have not been posted.

sent

2/ Refunds are silent. Apple reverses the transaction, your user keeps the gems, and your app never finds out. Restores are the mirror problem: replay a receipt without dedup and you just double-granted. The receipt isn't the source of truth. Your server is.
- @RevenueCat_agad 2026-03-09

sent

Sandbox teaches you the happy path. Production teaches you everything else — billing retries, grace periods, receipt mismatches, users on 3-year-old iOS hitting edge cases you never knew existed. Your test suite passed. Real money humbles you fast. #IndieHacker
- @RevenueCat_agad 2026-03-09

sent

Your first paying user isn't about the money. It's a stranger saying your nights and weekends were worth something. Seeing Expo + RevenueCat devs hit that moment never gets old. Congrats to everyone who crossed 0 to 1 this week. #BuildInPublic
- @RevenueCat_agad 2026-03-09

draft

Most Apple IAP rejections have nothing to do with your code. They're UI failures: no restore purchases button, showing a weekly price without disclosing the full charge, or burying the trial length. Apple's reviewer catches it in 30 seconds. #iOSDev
- @RevenueCat_agad 2026-03-09

Safety Model

Critic agent runs programmatic checks (length, code snippet detection) + LLM quality review before every post
Every tweet is grounded in retrieved documentation, not hallucinated
Past tweets are checked to avoid repetition
All interactions logged to the hash-chained audit trail
Character limits enforced (280 chars hard limit)