Super User Daily: 2026-04-30
If yesterday's tape says anything, it's that real users are stretching Claude Code and OpenClaw way past coding. The strongest cases on April 28 weren't engineers writing engineers' tools — they were a 16-year-old replacing a $40K IT quote, an ADHD user reorganizing 30,000 files in one shot, a wife collapsing 4-day business analysis to 5 minutes, and a slow-but-steady tide of OpenClaw + Hermes setups running 24/7 cron jobs while their owners sleep. Two themes run through almost every keeper: token economics finally moved from theory to wallet, and the harness layer (skills, hooks, sub-agents, Claude.md structure) is now the actual product, not the model.
@fin465 [Claude Code]
https://x.com/fin465/status/2049189280501899412
Vibe-coded a complete B2B lead-gen system in one week. Claude Code scrapes Google Maps, crawls each prospect's site, mines reviews for explicit pain points, then writes personalized cold emails. The interesting part isn't the agent — it's the loop: review-mining surfaces the angle, the angle drives the cold email. A solo dev replicated what most outbound agencies sell as a $5K-$10K/month service.
@rin_rinnak [Claude Code]
https://x.com/rin_rinnak/status/2049028960659640799
ADHD user pointed Claude Code at a 30,000-file mess on her computer. One session: read every file, rename, regroup, deduplicate. She estimates this would have taken weeks of mostly searching for things she'd already saved. The use case nobody talks about: AI as the cure for the file-system disorder humans have been ignoring since the 1990s.
@maarcoofdezz [Claude Code]
https://x.com/maarcoofdezz/status/2049133542999789890
Open-sourced a full job-search system on Claude Code: 14 modes, a Go terminal dashboard, ATS-optimized PDF resumes via Playwright, 45+ company-specific configs. The author shipped this after grinding through 700 applications himself — and it landed him a job. Open source on GitHub, MIT-license, ready to clone.
@aladagberk [Claude Code]
https://x.com/aladagberk/status/2049195292004684167
17-year-old built and shipped a Claude Code app from a Steam Deck, no Mac involved. Earned $13,000 in the first month. The hardware story is what matters: the cost of entry to ship monetizable software is now a $400 game console plus a Claude subscription.
@Nicjiang7 [OpenClaw]
https://x.com/Nicjiang7/status/2049128952070066341
Half-month quant trading system: Claude Code builds the strategy, Hermes runs 24/7 on a server, every decision is snapshot-logged, and the agent auto-reviews 24 hours later, building an experience library it learns from. Account grew $1,000 to $1,600. Token cost is real but bounded; the unique thing is the self-evaluation loop running while the user sleeps.
@09pauai [Claude Code]
https://x.com/09pauai/status/2049088786643239117
Claude Code is running a YouTube channel autonomously: video ideation, scripting, posting, and comment replies. Author reports 5,432 yen earned in five minutes of human time over a recent run. The pattern is clear — the model isn't the bottleneck anymore, the harness around it is.
@Honcia13 [Claude Code]
https://x.com/Honcia13/status/2049129759297114393
Released a narrator-ai-cli-skill that turns one-line prompts into film-recap videos. Inside the skill: 63 voice clones, 90+ visual templates, 146 BGMs, a 93-film clip library, 18 error-code handlers. Works across Claude Code, OpenClaw, and Windsurf. Pure assembly-line content production now triggered by a sentence.
@Honcia13 [Claude Code]
https://x.com/Honcia13/status/2048983294398652585
Same author, different skill: a Claude Code legal-review system that runs 5 agents in parallel and returns a contract review in 60 seconds — risk scoring, clause grading, plain-English translation, missing-protection detection, and on-the-fly NDA generation. The 60-second number is what would have shocked any legal AI vendor a year ago.
@dan__rosenthal [Claude Code]
https://x.com/dan__rosenthal/status/2049202048508317888
Detailed agency operating model on Claude Code: a Company-OS git repo with 26 agents, 79 skills, 23 commands; per-client repos; MCPs into GitHub, Findymail, Apollo, HubSpot and friends; 94+ safety hook guardrails; an n8n self-improvement loop. This is what an agency built natively on agents looks like — file system as ops manual.
@masahirochaen [Claude Code]
https://x.com/masahirochaen/status/2049159759966917111
Built a Slido replacement for a 300-person event in 1 hour 7 minutes with Claude Code. 3,400 lines TypeScript + 810 lines SQL. Real-time voting with load-shedding, dual moderator/audience views, fallback transports. The number that matters is 1h7m for production-grade — software that used to be a 4-week sprint.
@currypurin [OpenClaw]
https://x.com/currypurin/status/2049153425380954524
Two hours with Hermes Agent and the verdict: one-line install, Japanese natural-language commands, every completed task auto-promotes to a reusable skill (`/skills list --source local`), Obsidian output, cron via plain language. The reusable-skill auto-capture is the headline feature — Hermes is treating the user's own past work as the training set for future runs.
@fagamericano [OpenClaw]
https://x.com/fagamericano/status/2049005597773988031
Enterprise OpenClaw running for 300 users on GCP/GKE/gvisor. DevOps debugs customer issues via New Relic + Notion + repo, SecOps built a skill that does RFC review, the team even coordinated an offsite's logistics inside Slack. Production OpenClaw at scale, not a hobby setup.
@mamiya_afi [Claude Code]
https://x.com/mamiya_afi/status/2048935227251364015
The author's wife runs business red/black analysis with Claude Code skills: 30 minutes spread over 3-4 days collapsed to 5 minutes on demand. Pipeline: data aggregation → deficit-cause analysis → fix recommendation → automatic execution. The non-engineer angle is what makes this important.
@0801vika [Claude Code]
https://x.com/0801vika/status/2049091403792433249
One-month solo output with Claude Code: 3 apps in 5 languages, 2 websites, a YouTube series, a training program, auto-deploy infra, Supabase DB, 9 trademark filings. Either a ridiculous self-promotion or a believable demo of what compounding skills + agents enables — read it as the upper bound either way.
@gengdaJ [Claude Code]
https://x.com/gengdaJ/status/2048964733194449221
Built a Claude Code skill that decrypts WeChat chats, Moments, and favorites, then unlocks 9 workflows on top: daily group-chat digest into Feishu/Obsidian, competitor monitoring, customer CRM, IP-diagnosis pipeline. Uses Ling-2.6-flash for cost. Closed-platform data finally becoming queryable for normal users.
@RetroChainer [Claude Code]
https://x.com/RetroChainer/status/2049122925987451218
Three-day Mac Mini build with Claude Code: a Polymarket trading bot in 13 open repos. $120 → $26,880 over 14 days, 878 trades. The numbers are unverifiable but the architecture description is concrete and the repo count is real.
@LunarResearcher [Claude Code]
https://x.com/LunarResearcher/status/2049224880814739948
Polymarket cluster-fade trading system built with Claude Code: 86M-trade graph, python-louvain community detection, 7 wallet clusters identified. Result: +$18.4k from a $5k seed in 31 days. Unlike whale-copy, this is a contrarian read that requires real graph work.
@0xwhrrari [Claude Code]
https://x.com/0xwhrrari/status/2049183305270960286
Built a Polymarket whale-copy scanner with Claude Code over a weekend on a $5 VPS. Filtered 14k wallets down to 9 winners, 72.7% WR, +$14,985 in 31 days. The cost number is the headline — production-grade trading infrastructure for the price of a coffee per month.
@billtheinvestor [Claude Code]
https://x.com/billtheinvestor/status/2049202263936442402
Chinese review of MapiLeads — a B2B prospecting tool built solo with Claude Code in 2 weeks. Live Google Maps scrape, 30+ data fields per lead, review-mined pain points, 221 country coverage. Same pattern as @fin465's case but a different solo dev — replicated independently means the recipe is real.
@yasinaktimur [Claude Code]
https://x.com/yasinaktimur/status/2049062910182162479
Reports a user earning $25k/month making Claude-Code-built QR pages tailored for hotels and restaurants. The whole business is "templates + a script + a sales motion." If this is reproducible, it's the cleanest no-coding micro-SaaS on the timeline.
@noisyb0y1 [Claude Code]
https://x.com/noisyb0y1/status/2049057769840419179
16-year-old built a fintech AI agent (request classification + routing) in 11 days using Claude Code + n8n for $18K. Competing IT team had quoted $40K and 3 months. Month two the project hit $105K. Pricing arbitrage as a teenager is now a real career path.
@leopardracer [Claude Code]
https://x.com/leopardracer/status/2049100094020956416
Solo dev built a Chrome extension in 3 days with Claude Code: 270K downloads, 150K weekly actives, $20k/month. The model wrote the manifest, popup, and service workers. Distribution remains the moat, but the build cost has gone to zero.
@federico_rivi [Claude Code]
https://x.com/federico_rivi/status/2049184848644817123
Non-coder built a Spotify-like server-side playback over a single weekend with the Claude Code CLI: mood/activity playlists streamable via Tailscale and downloadable to phone. The non-coder qualifier is doing all the work in this one — Claude Code as the bridge between an idea and shipping software.
@maverickecom [Claude Code]
https://x.com/maverickecom/status/2049204803121602692
Production pipeline pumping out 550 TikTok-Shop ads per day. Stack: AI UGC persona + Arcads voice clone + Claude Code auto-edit (captions, pacing, hooks) + phone-farm push + Cruva Social hook discovery. Industrial-scale ad creative, replacing what used to be a 10-person studio.
@MichLieben [Claude Code]
https://x.com/MichLieben/status/2049163251796394117
Tutorial on automating LinkedIn outreach with Claude Code. The point isn't this specific implementation — it's that LinkedIn outreach, traditionally an SDR's full-time job, is now a one-skill agent.
@ouopacity [Claude Code]
https://x.com/ouopacity/status/2048987392141664349
Two months building a Claude Code relay/proxy site (sub2api-based). The thread reads like a postmortem on Anthropic's own anti-abuse defenses: account farming, ban patterns, fingerprint emulation, user-agent rotation. Useful as an honest description of the cat-and-mouse around heavy CC usage.
@yamadashy [Claude Code]
https://x.com/yamadashy/status/2048991830696730655
Set Claude Code on a periodic routine to auto-tune the CLI itself. Result: the CLI got 2.4× faster. Recursive self-improvement at the tooling layer — small but on-trend.
@petergostev [Claude Code]
https://x.com/petergostev/status/2049247902438924620
A/B-tested Opus 4.7 with 400k vs 1m context inside Claude Code on the same prompts (Voxel Rome, Golden Gate, Stonehenge). Conclusion: the 400k variant is more ambitious and accurate; the 1m variant feels lobotomized. Counter-intuitive result worth re-running.
@suragnair [Claude Code]
https://x.com/suragnair/status/2049275933895667852
Updated CompBioBench: Claude Code Opus 4.7 slightly underperforms 4.6; Gemini 3.1 Pro is competitive with GPT 5.4 and cheaper. The benchmark is small but the version-regression call is rare honesty.
@shin_sasaki19 [Claude Code]
https://x.com/shin_sasaki19/status/2049117163991306658
Claude Design hand-off analysis with hard numbers: Brilliant collapsed 20+ prompts into 2; Datadog cut a one-week brief→mock→review cycle to a single conversation. Two real production case studies of design-side velocity, not vendor copy.
@SoumyaG101 [Claude Code]
https://x.com/SoumyaG101/status/2048997858016399682
Used GStack + Conductor slash commands (/design-review, /design-html) with Claude Code to generate a full client product deck in a 3-hour iteration. Sub-three-hour deck-from-scratch is the metric here.
@manateelazycat [OpenClaw]
https://x.com/manateelazycat/status/2049053226461470990
Pure user case for OpenClaw + Hermes: wanted a text transcript of a 3-hour podcast. Agent autonomously searched, found and downloaded the audio, then ran local speech-to-text. End-to-end solved without the user wiring anything up — the kind of "boring, useful" use case that should be 80% of the consumer pitch.
@manateelazycat [OpenClaw]
https://x.com/manateelazycat/status/2049057260685680927
Methodology to drive AI through complex refactors: have the model produce a plan, then loop on "Is your plan complete? Anything missing?" until convergence, then run Codex yolo for 1-2 hours. The plan-then-yolo discipline beats raw autonomy.
@thedefiedge [OpenClaw]
https://x.com/thedefiedge/status/2048982931449033090
OpenClaw scans his email daily for flights/hotels, adds calendar blocks with traffic-aware airport leave times, cross-references WhatsApp/Slack for packing lists and visa requirements. Travel-EA replaced by an agent. Probably the most life-impactful use case in the dataset.
@hongyi88 [OpenClaw]
https://x.com/hongyi88/status/2049038857564479592
Three-day deep dive on Chrome DevTools MCP. Five tool categories — input, page, performance, network, console — enable auto sign-in, auto form-fill, project monitoring across OpenClaw / Hermes / Claude Code / Codex. Probably the cleanest summary of why MCP tooling is suddenly the high-leverage layer.
@MacopeninSUTABA [Claude Code]
https://x.com/MacopeninSUTABA/status/2048961148976816384
Splitting CLAUDE.md into a 3-layer structure (Rules / Skills / Other) reduced file size 83% with measured Claude Code performance data. Specific, reproducible, and the kind of optimization most users will find more valuable than any model upgrade.
@aakashgupta [Claude Code]
https://x.com/aakashgupta/status/2049126703927382051
Built a 9-node n8n contract-risk analyzer (Gmail trigger → embeddings → vector DB → AI agent) hitting 80% accuracy at $200/month — replacing $10K/month legal-AI vendors. Nine nodes is the unlock; everything before n8n was either a Zapier toy or a six-figure platform.
@MoonDevOnYT [Claude Code]
https://x.com/MoonDevOnYT/status/2049170734581244186
Built a Limitless prediction-market whale scanner and arbitrage tool with Claude Code, sharing the GitHub on request. Yet another solo trader / Claude Code combination, but the arbitrage angle (across markets, not whales) is fresh.
@om_patel5 [Claude Code]
https://x.com/om_patel5/status/2048928293261308352
The grill-me skill — a 4-line prompt — hit 13K stars by making Claude interview the user about every plan branch with recommended answers before coding. Ranked-choice intake replaces the "go figure it out" failure mode.
@abskoop [Claude Code]
https://x.com/abskoop/status/2049148626300232168
Claude Code / OpenClaw scammer skill: paste a suspicious message, get scam type + current stage + likely next step + a defensive response in 3 seconds. Defensive-AI consumer use case that anyone in a family group chat can immediately use.
@kevinnguyendn [Claude Code]
https://x.com/kevinnguyendn/status/2049157339190878520
ByteRover preserves agent context across Claude Code, Codex, OpenClaw, Hermes and 22+ agents. Saves tokens and avoids rewriting context when switching harnesses. The fact that this product exists is a sign of how stratified the agent-tool stack already is.
@garrytan [Claude Code]
https://x.com/garrytan/status/2049035231903301771
GStack ships Chromium with Claude Code as a CLI; /pair-agent gives a string for OpenClaw / Hermes to remotely control your browser session and scrape logged-in sites. Same pattern (browser-as-agent-tool) but YC-backed and packaged.
@carpassio [Claude Code]
https://x.com/carpassio/status/2049007925155561520
tradesdontlie MCP server connects Claude Code to TradingView Desktop via Chrome DevTools Protocol — agents read OHLC, indicators, drawn lines and write Pine Script. The trading-MCP angle is now one of the most active corners of the ecosystem.
@soft_servo [Claude Code]
https://x.com/soft_servo/status/2049184514602320187
text-to-cad open-source tool generates 3D models locally inside Claude Code or Codex. Built it for 7-axis robot arm parts, hit 1,000 stars in 3 days. AI in mechanical CAD is much earlier than software — early-mover energy here.
@doodlestein [Claude Code]
https://x.com/doodlestein/status/2049243607450763573
Dueling Wizards skill orchestrates Claude Code + Codex + Gemini-CLI as a swarm via the ntm tool: each agent independently generates ideas, scores rivals, and defends in an adversarial pipeline. Not a single-agent setup — multi-vendor agent debate as a productivity primitive.
@Shaughnessy119 [Claude Code]
https://x.com/Shaughnessy119/status/2049218675958550845
Hermes Agent user dropped Claude Opus bills from $130 in 5 days to $10 by mixing models (GPT 5.4, Opus, Qwen, DeepSeek v4) within the harness via flat-rate provider plans. The harness as a cost router — concrete wallet-level effect.
@sumika45379 [Claude Code]
https://x.com/sumika45379/status/2049066179277557935
Walks through running Claude Code free locally via Ollama: install, download a local LLM, swap the backend. Hybrid usage advice (heavy stuff on Anthropic, mundane stuff on local) is the practical takeaway.
@lucas_flatwhite [Claude Code]
https://x.com/lucas_flatwhite/status/2048957096650240296
Published everything-claude-code: 48 specialized subagents, 183 workflow skills, an AgentShield security auditor with 102 rules, and triggered hooks for full optimization. A complete kitchen-sink template repo for serious CC users.
@Dinosn [Claude Code]
https://x.com/Dinosn/status/2048973303302205617
Reverse proxy for Claude Code that anonymizes pentest data (IPs, hashes, creds, PII) before reaching Anthropic, using Ollama LLM + regex with a per-engagement vault. Security-vendor-grade compliance gating that lets pentest firms use Claude Code without contract risk.
@tak848_ [Claude Code]
https://x.com/tak848_/status/2049058280555884914
Hook-based tool auto-handles 97% of Claude Code permission requests across 2,000 monthly prompts, including Plan Mode. Codex CLI support too. The permission-prompt friction is one of the biggest quality-of-life issues for power users; this kills it.
@NFTCPS [Claude Code]
https://x.com/NFTCPS/status/2048938203927982094
OpenWolf hooks Claude Code with project file map, read-cache, error/preference memory and bug archive — six hook scripts cut tokens by 65%. Token-saving hooks are quietly becoming the most ROI-positive thing a heavy CC user can install.
@marlowxbt [Claude Code]
https://x.com/marlowxbt/status/2048931508539318697
19-year-old built a Claude Code AI security camera in a weekend on a $10 MaixCam device with a Telegram bot for person-detection alerts. 7,400 stars in 2 days. Hardware + Claude Code + Telegram = production tool for the price of pizza.
@marlowxbt [Claude Code]
https://x.com/marlowxbt/status/2049220915502981569
Same author's 16-year-old Chinese student used Claude Code to build an ESP32 button-LED circuit including 3D model, schematic, and firmware after one YouTube hardware video. The friction floor for hardware projects has dropped to "watch one tutorial."
🗣 User Voice
User Voice
What people on April 28 actually want from Claude Code and OpenClaw, distilled from the keepers:
Token bills are the new pain point — users want a model router and a way to mix flat-rate harnesses with high-cost models. Quote: "@Shaughnessy119 dropped from $130 to $10 by mixing models."
Permission prompts are friction — power users are routing around them with hooks. Quote: "@tak848_ auto-handles 97% of permission requests."
CLAUDE.md / skill structure matters more than the model — the harness layer is being treated as the actual product. Quote: "@MacopeninSUTABA cut CLAUDE.md size 83% with measured perf data."
People want the agent to act while they sleep — the high-engagement cases all involve cron / overnight / 24-7 / autonomous loops. Quote: "@Nicjiang7 has Hermes running quant trading 24/7 with self-review."
Non-coding cases want OS-level access and identity continuity (email, calendar, browser sessions, WeChat, Slack). Quote: "@thedefiedge has OpenClaw doing his travel admin daily."
What people on April 28 actually want from Claude Code and OpenClaw, distilled from the keepers:
Token bills are the new pain point — users want a model router and a way to mix flat-rate harnesses with high-cost models. Quote: "@Shaughnessy119 dropped from $130 to $10 by mixing models."
Permission prompts are friction — power users are routing around them with hooks. Quote: "@tak848_ auto-handles 97% of permission requests."
CLAUDE.md / skill structure matters more than the model — the harness layer is being treated as the actual product. Quote: "@MacopeninSUTABA cut CLAUDE.md size 83% with measured perf data."
People want the agent to act while they sleep — the high-engagement cases all involve cron / overnight / 24-7 / autonomous loops. Quote: "@Nicjiang7 has Hermes running quant trading 24/7 with self-review."
Non-coding cases want OS-level access and identity continuity (email, calendar, browser sessions, WeChat, Slack). Quote: "@thedefiedge has OpenClaw doing his travel admin daily."
📡 Eco Products Radar
Eco Products Radar
Tools mentioned 3+ times across the dataset, sorted by mention frequency:
Hermes / Hermes Agent — agent harness from Nous Research, the most-mentioned non-Anthropic harness today
Codex / Codex CLI — OpenAI's coding agent, the default comparison point in nearly every thread
Cursor — IDE-integrated agent, still cited as "best for IDE UX"
OpenCode — open-weight coding harness
Gemini CLI — Google's CLI, often used in multi-vendor swarms
Obsidian — the personal knowledge target for most CC/OpenClaw skill output
Claude Design — production design hand-off product cited with measurable enterprise wins
GPT-5.5 (G-55) — OpenAI's late-April release, widely used as the cost-effective alternative to Opus
DeepSeek V4 / V4 Pro — rising open-weight option in cost-routing setups
Kimi / Kimi K2 — used as a swap-in inside CC harnesses
GLM 4.7 — Zhipu's open model, also routing-friendly
Qwen 3.6 — Alibaba's local-runnable open model
Ollama — the default local-LLM runtime in hybrid setups
MCP — the connector standard underlying every advanced workflow above
Skills — Claude's primitive that almost every keeper depends on
Notion — destination for agent-generated docs
Polymarket — the trading platform bait for almost every Claude Code trading bot
TradingView — the chart layer agents are now reading via Chrome DevTools Protocol
Figma — design-handoff target for Claude Design and Claude Code workflows
n8n — workflow automation glue that pairs with CC for cheap production agents
Remotion — programmatic video pipeline used in agent content factories
ByteRover — context-preservation layer across multiple harnesses (3+ mentions)
ZenMux — a flat-rate provider often paired with Hermes / OpenClaw
Devin / Devin CLI — Cognition's offering, cited as the closest production-grade alternative for shut-laptop-and-walk-away workflows
Tools mentioned 3+ times across the dataset, sorted by mention frequency:
Hermes / Hermes Agent — agent harness from Nous Research, the most-mentioned non-Anthropic harness today
Codex / Codex CLI — OpenAI's coding agent, the default comparison point in nearly every thread
Cursor — IDE-integrated agent, still cited as "best for IDE UX"
OpenCode — open-weight coding harness
Gemini CLI — Google's CLI, often used in multi-vendor swarms
Obsidian — the personal knowledge target for most CC/OpenClaw skill output
Claude Design — production design hand-off product cited with measurable enterprise wins
GPT-5.5 (G-55) — OpenAI's late-April release, widely used as the cost-effective alternative to Opus
DeepSeek V4 / V4 Pro — rising open-weight option in cost-routing setups
Kimi / Kimi K2 — used as a swap-in inside CC harnesses
GLM 4.7 — Zhipu's open model, also routing-friendly
Qwen 3.6 — Alibaba's local-runnable open model
Ollama — the default local-LLM runtime in hybrid setups
MCP — the connector standard underlying every advanced workflow above
Skills — Claude's primitive that almost every keeper depends on
Notion — destination for agent-generated docs
Polymarket — the trading platform bait for almost every Claude Code trading bot
TradingView — the chart layer agents are now reading via Chrome DevTools Protocol
Figma — design-handoff target for Claude Design and Claude Code workflows
n8n — workflow automation glue that pairs with CC for cheap production agents
Remotion — programmatic video pipeline used in agent content factories
ByteRover — context-preservation layer across multiple harnesses (3+ mentions)
ZenMux — a flat-rate provider often paired with Hermes / OpenClaw
Devin / Devin CLI — Cognition's offering, cited as the closest production-grade alternative for shut-laptop-and-walk-away workflows
Comments