lishogi.org shogi ui

Compact Pi footer statusline with Git/worktree context, token usage, and neon celebrations for exceptional prompt-cache hits.

Your AI's shortcut to mastering Claude Code's architecture — 310K lines distilled into 32 queryable modules via MCP.

Show OpenAI Codex, GitHub Copilot, Z.ai GLM Coding Plan (global and China), DeepSeek, and Xiaomi MiMo balance usage in the pi coding agent.

MCP server for CreatorDB V3 API — search and analyze creators across YouTube, Instagram, and TikTok

OpenCode-style proactive tool output pruning for pi — reduce token usage by pruning stale tool outputs before each LLM call

Calculate inference costs and effective cost-per-resolved-issue across AI models for autonomous SWE agents.

Homebridge plugin for Hejhome IR devices.

A standalone terminal UI bundle for the DeepSeek Harness: an interactive in-process host surface over the official dsh base with session list, streaming conversation, tool cards, approvals, questions, jobs, subagents, goals, plan mode, model selection, se

Pi coding agent extension that streams OpenTelemetry GenAI traces to Arize Phoenix. Zero dependencies.

Observational-memory compaction strategy for pi with observer/reflector token thresholds

One-command setup for Flatland — financial modeling engine for AI agents. Configures Claude Code and Cursor to use the durable local Flatland client (your models are files you own).

Typed decision layer for the pi coding agent. Jev answers bounded questions with calibrated probabilities, so pi filters, verifies and gates cheaply instead of spending frontier tokens and turns on it.

> A simple package to add the VSCode marketplace to VSCodium.

<!-- Core implementation of [`good-fences-rs`](https://github.com/Adjective-Object/good-fences.rs) -->

MCP proxy that aggregates multiple servers with tool filtering - stop carrying 70k tokens of tools you'll never use

Convert BSData BattleScribe XML files (.cat/.gst) to JSON — raw or curated

Dependency-free geodesic helpers: haversine distance, bearing, destination, point-to-path distance, bounding boxes and Douglas-Peucker simplification.

Autonomous context-engineered development system with SQLite brain. 5 agents, 20 commands, per-task fresh context, 70-90% fewer tokens.

Reduce LLM token usage by up to 50% through intelligent context pruning. Auto-supersede duplicates, manually discard/distill content, and preserve critical state.