{"id":51228931,"url":"https://github.com/ystepanoff/awo","last_synced_at":"2026-06-28T14:01:48.529Z","repository":{"id":359238671,"uuid":"1245138659","full_name":"ystepanoff/awo","owner":"ystepanoff","description":"AWO — Agent Worktree Orchestrator: a local Go CLI that coordinates Claude Code and Codex across isolated git worktrees with deterministic verification.","archived":false,"fork":false,"pushed_at":"2026-05-21T01:08:24.000Z","size":70,"stargazers_count":0,"open_issues_count":0,"forks_count":0,"subscribers_count":0,"default_branch":"main","last_synced_at":"2026-05-21T07:17:01.631Z","etag":null,"topics":["agent-orchestration","ai-agents","claude","claude-code","cli","cobra","codex","developer-tools","git-worktree","go","golang","llm","orchestration","worktree"],"latest_commit_sha":null,"homepage":null,"language":"Go","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"mit","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/ystepanoff.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":null,"dco":null,"cla":null}},"created_at":"2026-05-21T00:26:34.000Z","updated_at":"2026-05-21T01:08:27.000Z","dependencies_parsed_at":null,"dependency_job_id":null,"html_url":"https://github.com/ystepanoff/awo","commit_stats":null,"previous_names":["ystepanoff/awo"],"tags_count":null,"template":false,"template_full_name":null,"purl":"pkg:github/ystepanoff/awo","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/ystepanoff%2Fawo","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/ystepanoff%2Fawo/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/ystepanoff%2Fawo/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/ystepanoff%2Fawo/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/ystepanoff","download_url":"https://codeload.github.com/ystepanoff/awo/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/ystepanoff%2Fawo/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":34890795,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-05-26T15:22:16.424Z","status":"online","status_checked_at":"2026-06-28T02:00:05.809Z","response_time":54,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["agent-orchestration","ai-agents","claude","claude-code","cli","cobra","codex","developer-tools","git-worktree","go","golang","llm","orchestration","worktree"],"created_at":"2026-06-28T14:01:46.867Z","updated_at":"2026-06-28T14:01:48.517Z","avatar_url":"https://github.com/ystepanoff.png","language":"Go","funding_links":[],"categories":[],"sub_categories":[],"readme":"\u003cp align=\"center\"\u003e\n  \u003cimg src=\"docs/images/awo-logo.png\" alt=\"AWO logo\" width=\"240\"\u003e\n\u003c/p\u003e\n\n# AWO — Agent Worktree Orchestrator\n\nAWO is a local Go CLI that coordinates Claude Code and Codex across isolated\ngit worktrees, runs deterministic verification commands against the result,\nand produces a structured artifact bundle (`run.json`, `proof-pack.md`,\n`diff.patch`, agent stdout/stderr) that a human reviews before merging.\n\nAWO does not commit, push, merge, or open pull requests on your behalf.\n\n## What AWO is\n\nA small, opinionated wrapper around two existing CLI tools — `claude` and\n`codex` — that:\n\n- Carves out an isolated git worktree per agent so agent edits never\n  touch your working tree directly.\n- Runs the agents you choose against the same task, in one of three modes\n  (`single`, `writer-reviewer`, `competitive`).\n- Runs the verification commands *you* configure (e.g. `go test ./...`)\n  inside the worktree, and treats the **exit code as the only trusted\n  signal** of success.\n- Writes a deterministic artifact bundle under `.awo/runs/\u003crun-id\u003e/` so\n  the run is auditable after the fact.\n- Hands the result back to you as a candidate change to review,\n  commit, push, and PR — manually.\n\n## Why isolated worktrees matter\n\nLetting an agent edit your live working tree is high-blast-radius:\nmid-run failures leave you with a half-applied change, parallel\nprocesses (your editor, your dev server) see partial state, and there\nis no clean \"undo\" if the agent goes off the rails.\n\n`git worktree` lets each agent run in its own checkout of your repo on\nits own branch, sharing the same `.git` directory. AWO uses that to:\n\n- Run multiple agents in parallel (competitive mode) without them\n  stepping on each other.\n- Bound the blast radius: AWO will only ever delete paths under\n  `.awo/worktrees/`, never your real source tree.\n- Capture an exact diff per agent for review.\n\n## Why Claude + Codex as separate backends\n\nClaude Code and Codex have meaningfully different strengths, prompt\nergonomics, and failure modes. Forcing one to ape the other loses\ninformation — you want the *real* output of each.\n\n- **Single mode** lets you pick the better tool for the task at hand.\n- **Writer-reviewer mode** uses one as the writer and the other as the\n  reviewer, which surfaces blind spots that a same-model review would\n  not catch.\n- **Competitive mode** runs both on the same task and ranks them with\n  a deterministic, explainable scoring function — never with an LLM\n  judge.\n\nThe two backends are wired through small adapter layers so adding a\nthird (or swapping CLI versions) is a config change, not a code change.\n\n## Installation from source\n\nRequires Go 1.22+.\n\n```sh\n# install into $GOBIN\ngo install github.com/awo-dev/awo/cmd/awo@latest\n\n# or build a local binary\ngit clone https://github.com/ystepanoff/awo.git\ncd awo\ngo build ./cmd/awo\n```\n\nThe `awo` binary is the only thing AWO ships — no daemon, no service,\nno remote dependencies.\n\n## Prerequisites\n\n- `git` on `$PATH`\n- Go 1.22+ (only for building from source)\n- The Claude CLI (`claude`) installed and authenticated:\n  \u003chttps://docs.anthropic.com/en/docs/claude-code\u003e\n- The Codex CLI (`codex`) installed and authenticated:\n  \u003chttps://github.com/openai/codex\u003e\n\n`awo doctor` checks all four and prints what's missing or\nunauthenticated.\n\n## Quick start\n\n```sh\n# Inside your project's git repo:\nawo init                                  # scaffold .awo/, awo.config.json, CLAUDE.md, AGENTS.md\nawo doctor                                # confirm git/go/claude/codex are reachable\nawo run \"add tests for calculator\" \\\n    --mode single \\\n    --agent claude \\\n    --verify \"go test ./...\"\n```\n\nWant a safe sandbox before pointing AWO at your real code?\n`awo examples create-fixture` materializes a tiny self-contained Go\nmodule under `.awo/fixtures/sample-go-app/` (its own git repo) so you\ncan dogfood every mode without risk.\n\n## Single mode\n\nOne agent does the work end-to-end inside a writer worktree.\nVerification runs in the same worktree.\n\n```sh\nawo run \"fix the off-by-one in pagination\" \\\n    --mode single \\\n    --agent claude \\\n    --verify \"go test ./...\"\n\nawo run \"fix the off-by-one in pagination\" \\\n    --mode single \\\n    --agent codex \\\n    --verify \"go test ./...\"\n```\n\nUse this when you've already decided which agent is best suited to the\ntask.\n\n## Writer-reviewer mode\n\nA primary agent writes the change in a writer worktree; a different\nagent reviews the writer's diff in a separate read-only worktree\ncarved from the same base. The reviewer's findings are surfaced in the\nproof pack, but **the reviewer cannot modify the writer's worktree** —\nany files it touches in its own worktree become a warning, not a\npatch.\n\n```sh\nawo run \"fix checkout validation\" \\\n    --mode writer-reviewer \\\n    --primary claude \\\n    --reviewer codex \\\n    --verify \"go test ./...\"\n```\n\nUse this when you want a different model's eyes on the change before\nyou spend your own attention on it.\n\n## Competitive mode\n\nTwo agents attempt the same task in parallel in independent worktrees.\nAWO runs verification against each, scores them with a deterministic\nfunction (verification status, diff size, test files added, protected\npaths touched), and surfaces the comparison.\n\n```sh\nawo run \"migrate date utility usage\" \\\n    --mode competitive \\\n    --competitors claude,codex \\\n    --verify \"go test ./...\"\n```\n\nThere is **no LLM judge**. The scoring is intentionally\nexplainable — the proof pack lists the inputs to every score so you can\ndisagree with the ranking on the spot.\n\n## Safety model\n\nAWO's safety stance is captured in three rules:\n\n1. **Verification command exit codes are the only trusted signal of\n   success.** Agent self-reports (\"I ran the tests and they passed\")\n   are persisted as advisory metadata only.\n2. **AWO never mutates state outside its sandbox.** Worktree deletions\n   are constrained to paths under `.awo/worktrees/`. Branches outside\n   `config.branchPrefix` (default `awo`) are never touched. The outer\n   repo's `HEAD` and working tree are never modified by AWO.\n3. **Human review is always required.** AWO has no merge button. The\n   final step of every run is a recommendation to a human, who is the\n   only thing that turns AWO output into a real PR.\n\nAdditional hard rules implemented in code:\n\n- Protected paths (default: `auth/**`, `payments/**`, `migrations/**`,\n  `infra/**`, `.github/workflows/**`, `**/.env*`, `**/*secret*`,\n  `**/*credential*`, `**/*permission*`) escalate the recommendation to\n  `needs_human_attention` whenever they are touched.\n- Patches that exceed `safety.maxChangedFiles` (default 50) escalate\n  to `too_large_for_auto_review`.\n- Reviewer-side worktree edits in writer-reviewer mode are detected\n  and surfaced as warnings, never applied.\n- Agents are not allowed to inspect each other's worktrees.\n\nSee [`docs/safety.md`](docs/safety.md) for the full list.\n\n## What AWO does not do\n\n- It does **not** auto-merge.\n- It does **not** auto-commit.\n- It does **not** push to remotes.\n- It does **not** open pull requests. (`awo pr prepare` writes a\n  `pr-description.md` you can paste; it does not call `gh`.)\n- It does **not** delete files outside `.awo/worktrees/`.\n- It does **not** guarantee correctness. Agents make mistakes; tests\n  miss things; the recommendation is a heuristic.\n- It is **not** a replacement for human code review.\n\n## Artifact layout\n\nEvery run writes a directory under `.awo/runs/\u003crun-id\u003e/`:\n\n```\n.awo/runs/20260525-094200-abc123/\n├── run.json            # canonical machine-readable record (RunReport)\n├── proof-pack.md       # long-form human report\n├── summary.md          # short-form human summary\n├── comparison.md       # competitive mode only\n├── pr-description.md   # written by `awo pr prepare` (not by run)\n├── diff.patch          # the diff produced by the selected candidate\n├── agents/\n│   └── \u003cagent\u003e-\u003crole\u003e/ # per-agent stdout, stderr, prompt, command\n└── verify/\n    └── 000/            # per-verification-command stdout, stderr, exit\n```\n\nInspecting after a run:\n\n```sh\nls .awo/runs/$(ls -t .awo/runs | head -n 1)/\ncat .awo/runs/\u003crun-id\u003e/proof-pack.md\ngit -C \u003cworktree-path\u003e diff\n```\n\n`awo worktrees list` shows the worktrees AWO is tracking; `awo\nworktrees cleanup --run-id \u003cid\u003e` removes them when you no longer need\nthem.\n\n## Configuration reference\n\n`awo init` writes `awo.config.json` with sensible defaults. The full\nschema and recommended overrides are in\n[`docs/configuration.md`](docs/configuration.md). Highlights:\n\n| Field                                  | Default            | What it controls                                              |\n| -------------------------------------- | ------------------ | ------------------------------------------------------------- |\n| `branchPrefix`                         | `awo`              | All AWO branches start with this prefix; nothing else is touched. |\n| `worktreeBaseDir`                      | `.awo/worktrees`   | Where worktrees live; deletions are bounded to this path.     |\n| `artifactDir`                          | `.awo/runs`        | Where run artifacts are written.                              |\n| `defaultVerifyCommands`                | `[]`               | Commands run when `--verify` is not passed.                   |\n| `agents.claude.command` / `writerArgs` / `reviewerArgs` | `claude` / `-p --permission-mode acceptEdits` / `-p --permission-mode plan` | Per-role argv. AWO runs every agent non-interactively; if the CLI hits an approval prompt the run fails closed with `permission_required`. |\n| `agents.codex.command` / `writerArgs` / `reviewerArgs` | `codex` / `exec --sandbox workspace-write` / `exec --sandbox read-only` | Same. `codex exec` is non-interactive by definition; the sandbox flag is the trust boundary. AWO refuses dangerous bypasses (`bypassPermissions`, `danger-full-access`, `dangerously-bypass-approvals-and-sandbox`). |\n| `safety.maxChangedFiles`               | `50`               | Patches above this escalate to `too_large_for_auto_review`.   |\n| `safety.protectedPaths`                | (9 globs)          | Hits escalate to `needs_human_attention`.                     |\n| `safety.requireConfirmationForProtectedPaths` | `true`      | Reserved for future interactive prompts.                      |\n\nRun `awo config print` to see the effective config (file values layered\non top of defaults).\n\n## Roadmap\n\nShort term:\n\n- More verification adapters beyond shell commands (lint, typecheck).\n- Pluggable scoring weights for competitive mode.\n- A \"rerun\" subcommand that resumes a failed run from artifacts.\n- Iteration support (`safety.maxIterations` is currently fixed at 1).\n\nLonger term:\n\n- Adapters for additional agent backends (Gemini CLI, etc.).\n- A small TUI for inspecting runs in place.\n- A signing/attestation pass over `run.json` so artifacts can be\n  trusted across machines.\n\nExplicitly out of scope:\n\n- LLM-as-judge scoring.\n- Auto-commit, auto-push, auto-merge — ever.\n- Anything that mutates state outside the configured AWO sandbox.\n\n## Docs\n\n- [`docs/architecture.md`](docs/architecture.md) — package layout and\n  how a run flows through AWO.\n- [`docs/safety.md`](docs/safety.md) — the full safety model and the\n  invariants tests enforce.\n- [`docs/prompts.md`](docs/prompts.md) — the prompt contract AWO uses\n  with Claude and Codex.\n- [`docs/examples.md`](docs/examples.md) — worked examples for every\n  mode using the bundled fixture.\n- [`docs/configuration.md`](docs/configuration.md) — full\n  `awo.config.json` reference.\n- [`docs/development.md`](docs/development.md) — how to build, test,\n  and contribute.\n\n## License\n\nSee [`LICENSE`](LICENSE).\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fystepanoff%2Fawo","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fystepanoff%2Fawo","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fystepanoff%2Fawo/lists"}