An agent built on Claude and Agno implements capabilities in its own source code in Docker, with changes validated, committed, or rolled back; it reached 21+ capabilities.
An agent built on Claude and Agno implements capabilities in its own source code in Docker, with changes validated, committed, or rolled back; it reached 21+ capabilities.
Miguel, built on Claude and the Agno framework, runs inside Docker with the project mounted read-only and only its own code directory writable. A host-side runner snapshots git, gives the agent its source plus the next capability, then validates changes with AST, JSON schema and import checks. It commits on success and rolls back on failure.
It finished all 10 seed capabilities, wrote its own checklist, and reached 21+ capabilities. A consolidation batch cut code by 10% and the system prompt by 63%. It still sometimes makes questionable decisions.
Each improvement batch costs roughly a few euros in Claude API calls, and the agent can exhaust its context by reading six or more of its own files before writing any code.
Help me set up a Docker sandbox where an agent can add new capabilities to its own source code in [repo path]. Each change should run the tests, be committed if they pass, and be rolled back if they fail, with a log of what happened. Finish by having it add [first capability, e.g. a web search tool] and show me the commit and test results; pause before any change that touches files outside the sandbox.
Score every demo request against your ICP, update HubSpot and ping the right rep in Slack within a minute of the form submit.
Every new Zendesk ticket gets a category, priority and sentiment tag plus a draft reply in an internal note before an agent opens it.
Copilot and Claude review each PR for bugs, missing tests and risky changes before a human looks, so reviewers can focus on design.