Forgeo is a software factory for your coding agent.

Give it a backlog and an agent CLI — Forgeo picks the next runnable task, runs the agent, and commits the result. It tracks progress in plain files and a web dashboard, and only interrupts you when a human decision is needed.

• One task at a time — oldest OPEN task whose dependencies are COMPLETED (or a run_at schedule). REVIEW blocks dependants but not independent tasks.
• Agent-agnostic — any CLI that reads FORGEO_TASK (aider, Claude, custom script).
• Refactors when idle — runs a refactoring pass when the backlog is empty.
• Handles failure gracefully — BLOCKED for human input (BLOCKER.md), FAILED with retry policy, snapshots for file backlogs, Telegram/webhook notifications.
• Optional review — review_mode: branch commits to forgeo/review/TASK-001, marks REVIEW, pushes and waits for human merge → Complete.

Requires a terminal, a git repo, and an agent CLI.

Full walkthrough: Getting Started.

Pick one (no root; re-run to upgrade):

brew install lucaGazzola/forgeo/forgeo          # Homebrew, no Python needed
curl -fsSL https://forgeo.org/install.sh | bash  # binary or pip fallback
pipx install forgeo-cli                          # Python 3.11+

forgeo init

Wizard in your project root. Creates forgeo.yaml and .forgeo/ (backlog, logs, blockers). Prompts for backlog provider (file / github / gitlab / jira / http) and agent command. For issue providers it offers a PAT or OAuth login (forgeo auth login → ~/.config/forgeo/tokens/); GitHub repository detection is automatic when possible, while Jira asks for its URL and JQL.

# file provider: edit .forgeo/backlog.json (see Backlog format)
# github/gitlab/jira/http: configured in forgeo.yaml, then:
forgeo validate

# or add tasks from the dashboard:
forgeo web      # http://0.0.0.0:8790  (use -d to keep it running)

forgeo validate   # dry run: config, repo, backlog, agent, locks
forgeo start      # daemon in background, one cycle per interval_minutes

Each cycle: pick task → run agent → commit/push (or REVIEW branch when review_mode: branch). Empty backlog → refactoring pass.

forgeo status              # config, counts, next task, daemon state, last outcome
forgeo once                # one cycle in foreground, no daemon
forgeo run --task TASK-012 # run a specific OPEN task now
forgeo stop                # stop daemon
forgeo restart             # restart daemon
forgeo web                 # dashboard for all instances

forgeo web is open by default on 0.0.0.0:8790. On a shared host use forgeo web --token for bearer auth — see Web console.

Run the agent isolated:

agent_sandbox: docker
agent_sandbox_image: your-image   # must contain agent CLI + sh
agent_sandbox_network: none       # default, no network
agent_sandbox_mounts: [~/.claude] # read-only mounts

Repo is bind-mounted at the same path; task arrives as FORGEO_TASK. See Configuration.

One config per repo, fully independent (own backlog, logs, locks). Use the instance registry:

forgeo instance add site-a --config /path/to/site-a/forgeo.yaml
forgeo start --name site-a
forgeo list          # all instances
forgeo web           # aggregate dashboard

Backlog providers: file · HTTP · Jira · GitHub · GitLab — file/HTTP exchange the full document; Jira/GitHub/GitLab sync issues individually. File backlogs are snapshotted (backlog.json.bak) before each run and restored if corrupt.

Dashboard: forgeo web aggregates every instance. For file/http it is the primary editor; for jira/github/gitlab it is a read-mostly mirror (links to native issues, surfaces BLOCKED/FAILED reasons and retry state).

pip install -e ".[dev]"
pytest

See CONTRIBUTING.md (quality gates: pytest, ruff check, mypy src/forgeo).

MIT — see LICENSE.