Ouroboros is an open-source autonomous CLI coding agent developed by razzant. Built primarily in Python, the agent is engineered to execute self-directed code modification loops, automate complex engineering tasks, and validate changes through structured preflight checks and review panels. In standard workflows, Ouroboros operates directly against a target repository worktree. It manages task planning, file edits, and multi-runtime test execution across Python and Node.js test suites. The agent uses configurable model backends (including Claude Sonnet variants) and enforces automated review cycles and task-acceptance verification before staging or committing modifications. What differentiates Ouroboros from basic interactive assistants is its focus on hermetic validation, process containment, and benchmark reproducibility. It incorporates strict commit gating with budget-capped test runs, a process container isolation layer to mitigate side effects, and native integration with benchmark evaluation suites like SWE-bench Pro, OSWorld, GAIA, and ProgramBench. Ouroboros is available as an open-source project hosted on GitHub, with ongoing maintenance centered around automated test gates, CI integration, and benchmark performance.
Tags: ai agent, cli, coding agent, developer tools, experiments, OpenSource
node --test) test suites within candidates. - Configurable review cycles — Enforces multi-pass review bounds and acceptance panels via configurable caps like OUROBOROS_REVIEW_MAX_CYCLES. - Process containment — Spawns and monitors execution tasks inside isolated process wrappers (ProcessContainer) to ensure clean cleanup and execution boundaries. - Benchmark harness compatibility — Includes built-in configuration scaffolds for evaluation against SWE-bench Pro, GAIA, OSWorld, and ProgramBench suites. - Autonomous self-iteration — Executes iterative task generation, execution, and self-review loops designed to handle long-running development workflows.Ouroboros is suited for software engineers, AI researchers, and framework maintainers seeking an autonomous, test-gated agent architecture with support for rigorous benchmark evaluation. It is less suited for developers looking for a lightweight, interactive conversational IDE autocomplete plugin.
bash git clone https://github.com/razzant/ouroboros.git cd ouroboros python3 -m venv .venv source .venv/bin/activate pip install -e . Ensure required runtimes (Python >= 3.10 and Node.js >= 20.11 if testing browser/node modules) are installed on your path.bash export ANTHROPIC_API_KEY="your-api-key-here" # Optionally configure maximum review cycles (default is 2, or 'unlimited' for long bench runs) export OUROBOROS_REVIEW_MAX_CYCLES="2"bash ouroboros --task "Fix failing test in tests/test_preflight_node.py and ensure Node runner stays green"Part of CLI Coding Agents
Categories: AI Coding Agents, Developer Tools, Experiments