Deliver messages an app queued on an MCP outbox, on a schedule, with no model in the loop.

MCP server: input/output safety gate for AI agents: match known prompt-injection/jailbreak patterns, vet URLs & IPs, check pwned passwords (HIBP k-anonymity), scan for leaked secrets & PII. Deterministic, free, no LLM.

Compaction quality / hallucination detector: when a conversation span is compacted, check on two axes that key information survived 鈥?FIDELITY (assertions in the summary must be found verbatim in the original leaf events, else flagged as fabricated) and C

MCP server giving AI agents (Claude Code, Claude Desktop, Cursor, ChatGPT, Codex, OpenClaw) persistent long-term memory backed by your local Obsidian markdown vault. Hybrid retrieval (BM25 + ML embeddings + BGE reranker, RRF-fused), HNSW + int8 quantizati

A local memory layer for coding agents.

Agent skills (SKILL.md) that find and fix the five security failures AI-generated code leaves behind most often: missing RLS, frontend-only authorization, IDOR, exposed secrets, unvalidated input. Works in Claude Code, Codex, OpenCode, Antigravity, Devin,

LMC's config for Textlint

Local-first MCP server for persistent, persona-aware memory across projects.

The Hiveloom workbench: a local inspector for building, running, and improving agent harnesses.

Local, free MCP server that compresses large file reads and noisy tool output for coding agents. Byte-exact code spans, lossless retrieval, no account, no network.

TypeScript-native semantic cache for LLM responses (OpenAI/Anthropic/Google). Sub-millisecond exact-match plus an optional zero-API-key semantic layer that serves paraphrased prompts. Memory or Redis.

Freeze your app's real contract, run your app's own cases against any OpenAI-compatible endpoint, and measure structural correctness plus honest warm-serving latency. Refuses to print numbers it cannot verify.

Model-driven context epochs for Pi: same-session fresh working contexts without LLM-generated summaries. Provides new_context, context_notes and context_history primitives.

Citation validation for RAG outputs — catches hallucinated citations by checking them against the retrieved source chunks that were actually provided.

SLC — spec-driven generation for AI builders. Uses your own LLM to turn one requirements file into a complete, validated SLC spec tree. Generates specs, never code.

On-device Chinese-English semantic embeddings for JavaScript via INT4 and ternary-weight WASM.

Install the /prompt2md agent skill: convert messy prompts and documents into token-optimized, layout-aware Markdown with an honest token report.

A Claude Code / Agent skill that rewrites AI-generated text to read naturally and survive AI detectors (GPTZero, Originality, Turnitin, Copyleaks, Pangram, DetectGPT, Binoculars). Includes a before/after detection-risk scorer.

Automatic long-term-memory extraction at compaction checkpoints: evidence projection, proposal/canonicalization pipeline, gated ctx.memory commit, and durable cursors/receipts/failure ledger.

Routes Anthropic API traffic from your eliza agent through a **Claude Max / Pro subscription** instead of paying per-token Extra Usage rates.