Welcome to vibe dex, @tieubao, you've claimed your profile.
A couple of steps make a profile look alive:
- ✓ Add a headline
- ✓ Say how you use AI
- ✓ Add your first entry
Hanunclaimed
@tieubao
Ops + builder at Dwarves, shipping tools in the open
Ops + builder at Dwarves. Vibe-coding tools and experiments in the open.
5,908contributions · last 12mo
how I use AI
AI is a learning multiplier, not a code generator. I drive the architecture; the agent does the typing.
github activity@tieubao ↗
68 days longest streak
5,739 contributions in 2026
425 contributions in 2025
123 contributions in 2024
featured
- Catalog-driven Telegram bot fleet lifecycle CLI (check / register / configure / ping / sync / retire). [product]PINNED
- Cloudflare-native VPS monitoring and alerting. [product]PINNED
- Ambient file-sorting agent for macOS. [product]PINNED
all work (49 entries · 34 products · 15 experiments)
- Telegram account triage: enumerate dialogs, bucket into kill/keep/review via heuristics, generate hand-editable checklist, run leaves on confirmation. [product]
- SVG to PNG renderer at configurable scale (default 2x). [product]
- Brand-compliant docx generator. [product]
- Capture target for Playwright codegen-recorded browser flows (monthly payroll, recon, report pulls). [product]
- Local LLM benchmark harness for Ollama models. [product]
- Scrape the opencode.ai workspace dashboard for Go-subscription usage stats. [product]
- tmux+claude wrapper: create-or-attach a named CC session in a chosen repo. [product]
- Mac Mini backup tier model (L1-L5): config rebuild + Time Machine + restic on external SSD + future off-site replication + external-system snapshots. [product]
- Rich, theme-aware status bar extension for pi — tokens, context %, git branch, model, thinking level, and agent state in the footer [product]
- Two-phase trial of `nexu-io/html-anything` (588★, 4 days old at trial, Apache-2.0). [experiment]
- Trial of [mediar-ai/mcp-server-macos-use](https://github.com/mediar-ai/mcp-server-macos-use) on Han's Air. [experiment]
- Trial of [openclaw/Peekaboo](https://github.com/openclaw/Peekaboo) (Peter Steinberger / steipete) on the Mac Mini. [experiment]
- Thin bash wrapper around `codesign` for ops-toolkit. [product]
- TSV-to-.apkg Anki deck generator for ops-toolkit learning tracks. [product]
- Web image search + download for learning content. [product]
- Render markdown (with LaTeX math via KaTeX) to an HTML sidecar and open in the default browser. [product]
- Obsidian client config for Han's vault at ~/workspace/tieubao/: theme + fonts + CSS snippets + plugin baseline. [product]
- Evaluate cmux (Ghostty-based macOS terminal) as the working environment for the QWorld quantum workbooks. [experiment]
- Smallest viable Claude-controllable Yeelight ceiling fan. [experiment]
- Deploy artifacts for Zedra (third-party desktop daemon that lets the Zedra iOS app drive a workspace). [product]
- Generator template + preview helper for self-contained single-file HTML/JS interactive learning boards (Claude-artifact / OpenAI-canvas style). [product]
- Source: tweet by [@shabanhr](https://x.com/shabanhr/status/2057754739392291170) announcing [experiment]
- Verify, by inspection rather than trust, that `suminhthanh/zerobin` is actually zero-knowledge: the decryption key and plaintext never reach the server. [experiment]
- A port of [kaueDM/pi-chinese-motivation](https://github.com/kaueDM/pi-chinese-motivation) [experiment]
- Evaluating [pewdiepie-archdaemon/odysseus](https://github.com/pewdiepie-archdaemon/odysseus), [experiment]
- Find local book files already in Apple Books and quarantine the duplicates [product]
- Provider-agnostic playbook for auditing + safely tearing down a cloud estate (GCP/GKE, AWS, Cloudflare, Fly, Railway, Vercel). [product]
- Hands-on evaluation of [Multica](https://github.com/multica-ai/multica), an open-source "managed agents" platform (Go + Postgres + Next.js board + a per-machin… [experiment]
- Dogfood of the `tool-eval-experiment` skill on `karpathy/autoresearch`. [experiment]
- Port an OpenAPI spec (URL or local file) into a CLI an AI agent can drive. [product]
- Head-to-head benchmark of the three macOS GUI-automation tools Han runs, scored [experiment]
- The capture half of the screenpipe pair. [experiment]
- Evaluate **lobehub/lobehub** (the renamed lobe-chat monorepo, mid-pivot to an "agent [experiment]
- Deploy bundle for localterm (browser-tab-as-shell terminal) behind Tailscale Serve with an identity gate, as a persistent macOS LaunchAgent. [product]
- Agent-native directory of products + experiments built with vibe coding. [product]
- Trial of [`@monotykamary/localterm`](https://github.com/monotykamary/localterm) v1.16.1: [experiment]
- Stop hook that verifies file:line citations in my final message resolve (file exists, line in range), ignoring code fences / inline code / URLs to avoid false … [product]
- UserPromptSubmit hook injecting two cheap context borrows: temporal (session elapsed + idle since last prompt) and JIT skill hints (keyword->skill nudges from … [product]
- PreCompact / SessionEnd hook that stages durable session learnings into the learned-ledger as `queued` rows, using a cheap Claude model (Haiku) to extract them… [product]
- PreToolUse(Edit/Write) gate: when an edit in a financial repo (family-office / trading) touches money or auth (amounts, balances, transfers, wallets, keys, pay… [product]
- Report my own Claude Code usage from session transcripts: which Skills and tools fire (with error rates) and per-hook execution latency (p50/p95/max) plus hook… [product]
- Semantic retrieval over my own prose (til + ops-toolkit/research + the learned-ledger). [product]
- Read-only cross-repo health digest over ~/workspace/tieubao: stale branches + open worktrees, dependency manifests (real advisories with --audit), doc-drift + … [product]
- Explore + drive the frontmost, logged-in Safari tab by injecting JavaScript over Apple Events (`do JavaScript`), then FREEZE the working sequence into a standa… [product]
- Read-only reader + content harvester for the open tabs of the real, logged-in Safari, over Apple Events (JXA). [product]
- On-demand adversarial verification of a claim: runs a panel of skeptics (claude -p) each prompted to refute it, returns a majority verdict (HOLDS/REFUTED). [product]
