Browse agent skills

Multi-modal RAG engine: HNSW + BM25 + knowledge graph, fused with RRF.

OpenCode TUI plugin: counts the turns cut off mid-thought, says whether the output cap or the context window did it, and shows what thinking costs per turn and after which tool.

Playwright driver for @harnessed-ts/core, plus API stubbing and BDD world helpers.

Draw on Mapbox, MapLibre, Google Maps, Leaflet, or ArcGIS in React: freehand, arrows, shapes, geodesic measure, and labeled markers — with your React state and your toolbar.

EchoMirror SDK social module — global feed, leaderboard, and real-time updates

Pi Coding Agent extension: granular per-model context-window hard cap + configurable compaction summariser model, all from settings.json.

MIDDAG Agent Skills — portable skills, modes and adapters for AI coding agents.

An extensible music player daemon written in Rust — Rockbox playback engine, FTS5 search, TUI, web UI and gRPC/GraphQL APIs

Write a bot for Aurival. Declare commands, call run(), we handle the rest.

Streaming CSV parser and stringifier. Bounded memory. RFC 4180 plus the real-world dialects. Strict-by-default malformed-input handling.

Scaffold AI applications quickly

Unified Weft SDK for the Weft API and x402 Facilitator

Headless browser audio primitives for Antiphony clients — a microphone recorder, a playback hook, and client-side waveform peaks. Renders nothing, styles nothing, uploads nothing.

Ethiopian payment transaction verification SDK for Telebirr, CBE and Bank of Abyssinia.

Cryptographic primitives, key handling, and wire-format compatibility for Pigeon Swarm.

Codex-style lossless-retrieval context manager for the DeepSeek Harness: token-budgeted sliding window, persistent working notes, and cold-history search over the append-only session log

Codex-style lossless-retrieval context manager for the DeepSeek Harness: token-budgeted sliding window, persistent working notes, and cold-history search over the append-only session log

A terminal coding assistant that reads your codebase, adopts an engineering role, and proposes changes you review before anything is written.

Minimal cron expression parser and next-run-time calculator — standard 5-field cron, no dependencies.

Simple input/output guardrail checks for LLM apps — PII patterns, length limits, banned phrases, and custom rules, composable into a single check pipeline.

Lightweight in-memory vector similarity cache — cosine-similarity nearest-neighbor lookup over a small in-memory vector set, for when a full vector DB is overkill.

Asynchronous Pi subagents with nested delegation, parallel multi-agent orchestration, and supervision in herdr.

Line-level diff viewer for prompt template versions — see exactly what changed between two prompt revisions.

Streaming-friendly CSV parser and writer — handles quoted fields, embedded commas/newlines/quotes, and custom delimiters. Zero dependencies.

Rolling conversation context trimmer for LLM chat apps — keeps the most recent messages within a token budget while preserving a pinned system message.

Standalone Vanilla TypeScript audio player library with encapsulated Shadow DOM and sound horeg aesthetics.

Token-bucket rate limiter — per-key limiting for APIs, jobs, or LLM calls, with a pluggable clock for deterministic tests.

Token budget estimation and tracking for chained LLM calls — approximate token counting, per-call budgets, and running totals across a multi-step pipeline.

Lightweight prompt templating with {{var}} interpolation and version tracking, so you know which prompt version produced which LLM output.

React voice agent UI components with audio-reactive adapters for Vapi, ElevenLabs, LiveKit, Pipecat, OpenAI Realtime, Gemini Live, and custom voice AI apps.