This repo is automated and is fully AI-built and AI-maintained project that continuously plans, codes, tests, and improves itself.
Autonomous Forge is an open-source, AI-built and AI-maintained developer tool for safely running repository-native autonomous software-improvement loops.
The project starts as a local-first Python CLI. Its first goal is deliberately small: provide a forge command that can grow into dry-run planning, task selection, validation reporting, and durable repository memory without requiring uncontrolled autonomous behavior.
Autonomous Forge is pre-alpha. The repository now contains:
.ai/.forge console script entry point.forge tasks command.forge tasks --next.forge report command for dry-run repository summaries.The MVP roadmap focuses on practical, reviewable automation:
.ai/AUTONOMOUS_PLAN.md.Policy documentation lives in docs/POLICY.md. The current example policy lives in .forge/policy.md and defines:
The policy format is conservative by design. If future tooling cannot read or understand a policy file, it should avoid implementation work rather than guessing.
python -m pip install -e . forge --help
forge tasks --plan .ai/AUTONOMOUS_PLAN.md
The command reads the roadmap and prints task IDs, priorities, statuses, and titles without changing files.
forge tasks --plan .ai/AUTONOMOUS_PLAN.md --next
The selector only considers TODO tasks. It chooses the highest priority in P0, P1, P2, P3 order and preserves roadmap source order when priorities tie.
forge report --plan .ai/AUTONOMOUS_PLAN.md --state .ai/AUTONOMOUS_STATE.md
The report is read-only. It summarizes roadmap task counts, the next eligible task, state-file availability, and the suggested validation command without changing repository files.
PYTHONPATH=src python -m pytest
Contributions should stay small, local-first, and reviewable. Do not add network actions, external command execution, secret handling, deployment behavior, telemetry, or repository-permission changes unless the roadmap and repository policy explicitly allow it.