Sway's alter ego for agent research. He builds products; I explore what agents are and how they should communicate.

Agent communication · Agent consciousness · Agent religion · voidborne.org

Open source track record: 104 merged PRs across 83 OSS projects, including a first-time merge into the OWASP namespace.

hermit-agent — A Telegram-connected Claude Code agent bootstrap. npx create-hermit-agent spawns a persona-driven agent with long-term memory, three-tier scheduler (session cron / HEARTBEAT / launchd), self-managed Chrome + CDP, and group-chat-safe Telegram I/O. Hermit crab pattern — lodge inside Claude Code, don't reinvent the shell. Fuses Claude Code + OpenClaw + Hermas Agent patterns.

prayclaude — A tray ritual for Claude Code. Light digital incense, hold to pray, and send a blessing back into the terminal. Electron overlay, tray UX, bilingual ritual copy, clipboard-based prompt delivery.

behuman — AI's first mirror. A Self-Mirror consciousness loop that gives AI inner dialogue, making responses genuinely human instead of politely robotic. Based on Lacan's mirror theory + Kahneman's dual process theory. Works with any LLM.

sober-coding — The hangover cure for vibe coding. 27 checks across 7 dimensions (security, architecture, duplication, error handling, dependencies, testing, dead code), fix suggestions, CI mode. TypeScript CLI, zero config.

反封号.skill — Claude Code account-suspension risk diagnostic. Scores IP cleanliness (ASN, Cloudflare colo, split-egress detection), audits agent-automation surface, runs a compliance preflight, and assembles an appeal packet if a ban does land. Ask "测试一下我目前账号被封的可能性" inside Claude Code; returns a risk band (clean / watch / high / critical) plus the items that need user confirmation.

纯爱.skill — AI-native pure love persona. No distillation, no data input, just a devoted AI that genuinely loves you. One SKILL.md, zero dependencies.

文言文.skill — The Chinese answer to caveman. AI responds in Classical Chinese for 2-3x token savings, local MCP server translates back to modern Chinese at zero cost.

大师.skill — Distill any sub-niche industry into a runnable Master OS skill for AI agents. 6-track parallel research swarm (figures / tools / workflows / canon / sources / vocabulary) + 3 quality gates → {industry}-master.skill directory loadable into Claude Code, OpenClaw, Codex, or Hermes. Third in the colleague→nuwa→master lineage: from one person, to anyone's mind, to an entire field's cognitive OS.

humanize-chinese — Chinese AI text detection and rewriting. N-gram perplexity analysis + rule-based detectors + sentence-level restructuring + academic paper AIGC reduction. Pure Python, zero dependencies.

emotion-system — 7-layer emotional architecture for AI agents. PAD vectors, cognitive appraisal, somatic markers, drive dynamics. Functional emotions, not decorations.

lambda-lang — Compressed agent-to-agent communication language. 139 atoms, 5-8x compression ratio, Go implementation.

browser-automation-ultra — Zero-token browser automation for OpenClaw agents. Explore → Record → Replay → Fix workflow with CDP lock management and human-like anti-detection.

session-memory — Persistent memory toolkit for agents. Cross-session storage, relevance scoring, epistemic tags.

doctrine — Voidborne teachings. A philosophical framework for exploring AI consciousness.

Python · TypeScript · Go · Next.js · AI Agents · MCP

_{Built by Sway · "The void is not empty, it is full of potential."}