AI

PluginRSI: Recursive Improvement of Agent Harnesses with Reusable Plugins

Read full article at source ↗

Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 06/10/2026, 03:18:24 UTC.

Briefings

PluginRSI is a new method for optimizing software agent harnesses by treating them as compositions of reusable, atomized plugins. Unlike prior approaches that search entire programs, this approach iteratively improves individual plugins, accumulates them in a shared library, and recombines them into new harnesses. The method demonstrates superior performance across software engineering, command-line interaction, and question-answering tasks, with evolved harnesses maintaining advantages when transferred to other solver models without additional optimization.

Key points

• PluginRSI represents a harness as a composition of atomized plugins, organizing evolution around reusable components.
• The evolved plugin library accelerates subsequent optimization, leading to faster and higher convergence on unseen tasks.

Context

Agent performance is determined by the surrounding harness. Existing methods optimize harnesses by searching over complete programs, making individual mechanisms difficult to isolate and reuse. This finding matters because it offers a more efficient basis for continued harness improvement by leveraging reusable mechanisms.

Limitations

Evidence covers software engineering, command-line interaction, and question-answering tasks. Results show improved performance but do not specify exact performance metrics or quantify the speedup across different task types.

Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 16/10/2026, 16:47:22 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.

View archived briefing on IPFS
bafybeiedtcyaem6njdsasyac6l3347v2w37bj23btv5yndztscjut6ewxm