{
"$type": "site.standard.document",
"bskyPostRef": {
"cid": "bafyreib7bwi6qxw6adziqiv2clr625gqqzbelohd3kadpfyoyjw6bsqiqm",
"uri": "at://did:plc:pgryn3ephfd2xgft23qokfzt/app.bsky.feed.post/3miwvseyowom2"
},
"path": "/t/orca-a-cognitive-runtime-layer-for-agent-systems-paper-open-source/175055#post_1",
"publishedAt": "2026-04-07T16:47:37.000Z",
"site": "https://discuss.huggingface.co",
"tags": [
"github.com",
"GitHub - gfernandf/agent-skills: Agents should execute whenever possible — runtime...",
"Zenodo",
"Beyond Prompting: Decoupling Cognition from Execution in LLM-based Agents..."
],
"textContent": "I’ve been exploring some of the structural limitations of prompt-based agent systems and built a framework to experiment with an alternative approach.\n\nThe core idea is to introduce a “cognitive runtime layer” (ORCA) between the agent and the underlying tools.\n\nIn this model:\n\n * capabilities represent atomic cognitive operations (e.g., retrieve, transform, evaluate)\n * skills define composable workflows over capabilities\n * execution is explicit and structured, rather than embedded in prompts\n\n\n\nThis aims to separate concerns that are often tightly coupled in current agent designs:\n\n * cognition (what needs to be done)\n * execution (how it is carried out)\n * orchestration (decision-making at the agent level)\n\n\n\nThe hypothesis is that making this separation explicit can improve:\n\n * composability\n * observability\n * controllability of execution\n\n\n\nOpen-source implementation:\n\ngithub.com\n\n### GitHub - gfernandf/agent-skills: Agents should execute whenever possible — runtime...\n\nAgents should execute whenever possible — runtime for composable AI agent skills\n\nPaper (DOI):\n\nZenodo\n\n### Beyond Prompting: Decoupling Cognition from Execution in LLM-based Agents...\n\nRecent advances in large language model (LLM) agents have largely relied on prompt-centricdesigns, where complex tasks are executed through monolithic, single-shot or loosely structuredprompting strategies. While effective in some settings, this...\n\nI’d be particularly interested in feedback on a few points:\n\n * how far capability granularity should go before overhead dominates\n * whether declarative execution models can realistically replace prompt pipelines\n * how this kind of abstraction would behave in more complex, real-world agent systems\n\n\n\nHappy to expand on the execution model, design principles, or concrete examples if useful.",
"title": "ORCA: A Cognitive Runtime Layer for Agent Systems (paper + open source)"
}