{"repo":"canvas-org/meta-agent","free":true,"listed":false,"github":"https://github.com/canvas-org/meta-agent","clone":"git clone https://github.com/canvas-org/meta-agent.git","description":"Continual harness optimization","language":"Python","stars":69,"topics":["agent","anthropic","claude","harness","llm","training","tuning"],"license":"MIT","category":"ai-agents","readme_excerpt":"Meta-Agent Continual learning for agents, without weight updates. Meta-Agent turns execution traces into persistent improvements to the system around a language model. It learns at two timescales: - a fast loop improves the task-facing harness, including prompts, tools, hooks, stopping criteria, subagents, and control flow; - an experimental slow loop improves the proposer skill that diagnoses failures and generates future harness changes. The base model remains frozen. Meta-Agent keeps what it learns in harness code and versioned proposer instructions, backed by an on-disk experience store. Research question: Can an agent improve both its harness and the process used to improve that harness? Read the technical report → Method · Result · Quickstart Overview An agent is more than its model. The harness decides what context the model sees, which tools it can call, how it tracks state, when it checks its output, and when it stops. Meta-Agent treats that harness as an editable program. It runs the current harness, records task-level traces and outcomes, proposes a targeted change, and evaluates the resulting candidate. It keeps both wins and failures so the next proposer can see what worked and what already failed. After several iterations, an optional second agent compares proposer traces with candidate outcomes, then updates the instructions that guide future proposals. One loop improves the task harness; the other improves the process that writes the next harness. Our technica","default_branch":null,"files":null,"tree":[],"storefront":"/r/canvas-org","claimed":false,"request_supported":{"post":"https://gitbuyer.com/r/canvas-org/meta-agent/request-supported","requests":0},"note":"indexed from public GitHub; nothing is for sale on this page. Clone it from GitHub. Paid listings live at /search."}