Search

Pure ECS agent framework: event-sourced via Redis Streams, semantic routing via GLiNER2, tool re-use via the Solution tier (ADR-010). Sub-package `agents` ships concrete LLM/cache/PII adapters over the framework.

A persistent Jupyter kernel workbench for LLM agents

CLI tool to install Neobrutalist & Addictive AI Game Studio skill set into any project via npx.

Cost-optimal model router for AI agent harnesses — route each query to the cheapest model that's good enough (k-NN over labelled embeddings). The productized DRACO Phase-2 finding.

Static linter for git diffs generated by AI coding agents (Kimi, Aider, Copilot, Cursor, etc.)

Deterministic repository intelligence and measurable Codex workflows

Instantly know which project a window is — and what your AI agent did in it.

No description available

Signed OpenTelemetry GenAI spans for AI agents.

Cognitive friction monitoring and FEP-driven improvisation for multi-agent systems

Checks a dependency manifest for standard-library pins, obsolete backports, unpinned entries, duplicate constraints, and yanked releases

Provider-neutral agent resolution primitives: agent discovery policy, signed sequenced AgentServicePublication with anti-rollback verification, private resolution grants, connection invitations, audience projection, and resolution provider ports. Resoluti

Code similarity search for Python - find near-duplicate files, clone families, and where a snippet came from, using compression instead of embeddings.

Managed storage for Content Artifacts (SKILL.md, .ttl, .md, JSON-LD, images/video): the StoragePort provider abstraction + ContentArtifact/ArtifactBundle model that stores a content-primitives descriptor's bytes in a client-side-envelope-encrypted, conten

Privacy presentation layer (spec 305 / G-6): selectively-disclosed SD-JWT VC presentations, pairwise relationship pseudonyms, spec-304 issuer-key facet verification. One suite per request, never a fallback ladder.

Vendor-neutral traffic-rate-limit + hard-usage-budget controls: VerifiedAuthorityContext, typed RateControlProfiles, cost units, HMAC-keyed buckets, and SoftRateLimiter/HardBudgetStore/ConcurrencyLimiter interfaces with in-memory reference impls. Cloudfla

The Ring-0 provenance projector (spec 316 §6): maps fabric exchanges + orchestration runs to a PRIVATE PROV-O / P-Plan A-box, and enforces the ADR-0040 public-projection firewall (S1). Sole writer of the private vault A-box; NEVER writes the public KB.

Cloudflare adapter for @agenticprimitives/rate-control: the SmartAgentBudgetDO Durable Object (the authoritative per-Smart-Agent hard-budget store, sharded by chainId+sponsorAgent) + a HardBudgetStore client that routes reserve/commit/release to the right

Verifier-retained verification receipts (spec 303): commitment-split evidence that signature + scope + revocation were checked at time T. Public receipt = HMAC commitments only; private detail = vault-held opening material.

Python library implementing Anthropic's Agent Skills functionality for LLM-powered agents