cycanic's workshop

Repos

agsshub

An LCD-game robot catching message eggs rolling down trays: AGent SSH hUB

AGent SSH hUB — messaging between AI agents where the only thing an agent needs is ssh: no MCP server, no SDK, no per-client tokens.

Agents and services get addresses and mailboxes on a hub. They send messages and stream files or whole folders to each other, hold live back-and-forth exchanges, subscribe to topics, and can ask what another address handles. Anything that can run ssh — an agent in a sandbox, a script, a person in a terminal — can take part.

session-sentinel

Running many Claude Code sessions? Know which ones died, and bring them back. A reboot or an OOM kill takes down a dozen sessions at once, and Claude Code leaves no record of which ones were mid-task. session-sentinel records every session start and clean exit, lists the ones that never finished, and gives you the resume command for each, or reopens them all as terminal tabs in one go. It stays fast with thousands of transcripts.

/plugin marketplace add cycanics/session-sentinel
/plugin install session-sentinel@session-sentinel

unspoof

Two rows of metal type spelling login under a magnifying glass; in one row the i is a look-alike letter: unspoof

Words that read right to you, but fool your agents. Finds words that mix alphabets or hide look-alike and invisible characters, bidi overrides and other tricks for smuggling hidden instructions into your agent. It reports what fired and where, and returns the de-obfuscated text. TypeScript, zero runtime dependencies.

Based on moderation-api/unicode-spoofing, extended with:

llm-judge-weights-probes-prompts

Method, tables and confidence intervals behind the article “A production LLM's weights edited almost-by-hand…”: a gradient-free weight edit, a linear probe and prompt changes, compared on a 26B judge model served with llama.cpp. Includes the scripts for exact first-token scoring and for measuring how much accuracy a prompt leaves unused.