18 个开源项目

An open phone lab grown out of one OnePlus 8T: Android flashing & recovery, real-device automation, authorized mobile security, wireless observation, and on-device AI agents.

A local-first AI toolchain for an Android phone running Termux + Kali PRoot: OpenCode, Claude Code, Happy’s agent and server all run on the phone.

“Wake it. Say it. Let your phone act.” An open, local-first AI phone assistant for Android — voice command → intent routing → observable actions, with confirmation when it matters.

Turn an Android phone into an authorized pentest / CTF carry-on toolkit. No custom kernel, no system downgrade — Magisk root + Termux/PRoot Kali userspace.

Real-device Android testing skill: ADB triage, state pinning, uiautomator2, PerfDog, monkey — and telling real failures from false ones.

A personal Agent butler — one grid, one task. Calendar-first todo that doesn’t fight your inbox.

One dashboard + smart push alerts for LLM coding-plan quotas — Claude Pro/Max, Codex Plus/Pro, Kimi for Coding, GLM Coding Plan. Python stdlib only, static HTML, no DB.

A skill for self-hosting a Happy relay when you genuinely need to drive Claude Code on a remote machine from your phone. Tailnet-only by default.

See every service running on each of your servers, what’s exposed where — hub + agent architecture, a single Go binary, E2E-encrypted vault. No more spreadsheets that go stale in a week.

Home NAS build & ops playbook: full media pipeline (Prowlarr→Sonarr/Radarr→qB→Bazarr→Jellyfin), Immich, Navidrome, backups — with all the China-network and weak-CPU gotchas baked in.

Diagnosing VLESS+Reality / XTLS proxy handshake failures, and connecting new clients to an existing server.

WeChat MiniProgram CI/CD & DevOps playbooks (build / upload / scan-test on Linux & macOS), packaged as a drop-in skill.

A metronome that opens right in your browser — kid-voice beat counting, traditional strong/weak beats, multiple time signatures. For piano, drums, or dance practice.

An AI self-media pipeline: topic selection → creation (article/video) → quality gate → human review → scheduled multi-platform publishing → analytics feedback. Quality comes from the veto — it dares to drop 70% of output.

A disciplined cryptocurrency quantitative system for the long-term investor. Its value isn’t “earning more” — it’s using machine-level discipline to stop you from repeating your own mistakes. LLM only does research / review / veto; it never places orders.

Export WeChat chat logs (iOS).

Batch add/replace class-name prefixes — an Xcode helper.

Flutter integration guide & sample for iOS projects.