Open Source Radar — October 16: the agent supply chain
Today's daily trending board is almost entirely plumbing — the four AI repos that surfaced are a harness, a skills library, a plugin directory and a retrieval engine. Not one new model among them. That is the signal: the layer around the model is getting productised.
ruflo (TypeScript, ~75,000 stars) — The agent harness that keeps returning to the daily board: a swarm orchestrator with a shared memory layer, a large fleet of specialised agents, and a marketplace of plugins and packages that plug into Claude Code, Codex and other hosts. It sat around 68,000 stars when we covered it in August and has added roughly 7,000 since, which is quiet evidence that teams keep hitting the ceiling of single-agent setups. Reach for it when you want several agents sharing state and handing work off, instead of one agent carrying a longer prompt. The catch is that it is opinionated and large — you are adopting a framework, not a library.
scientific-agent-skills (Python, ~51,000 stars) — A curated library of 177 validated skills plus access to more than a hundred scientific databases across biology, chemistry, medicine and drug discovery, packaged for any agent that speaks the open Agent Skills standard. It recently renamed from Claude Scientific Skills and dropped its single-vendor framing; the maintainers now cite use by more than 250,000 scientists, up from the 175,000 they claimed in August. The value is correctness — each skill encodes how to query a database or run an analysis properly, so the model stops improvising at the bench. You would pull it in when research output needs to be reproducible rather than merely plausible.
claude-plugins-official (Python, ~40,000 stars) — Anthropic's own directory of Claude Code plugins, split between the ones the company maintains and third-party entries from partners. It matters less as software than as a governance move: the first-party vendor is now curating the surface that extends its agent, rather than leaving discovery to scattered community lists. If you are choosing what to install into an agent, this is the shortest path to something reviewed. If you are building on Claude Code, this is the directory you want to be listed in.
xerj (Rust, ~6,200 stars) — A local retrieval engine, Elasticsearch-compatible, that indexes a folder of code, docs, logs and PDFs so an agent asks for the passage it needs instead of grepping and dragging whole files into its context. The project's own case study reports a large reduction in output tokens on code the model had not memorised, and the honest pitch underneath is simply that retrieval beats reading once a corpus is big. It exposes keyword, vector and hybrid search plus durable agent memory to agents over MCP, and existing Elasticsearch clients connect without changes. One oddity worth knowing before you adopt it: the project asks new users for a field report — a short write-up of what you pointed it at — instead of a signup or a licence fee.
Worth watching this week: whether curated skills libraries and vendor plugin directories become the default route by which new capability reaches agents, or just one more layer someone has to audit.
If an agent's toolbox is now curated by vendors, who is checking what ends up in it? Tell us in the comments.
