swarajbachu/yolorun

swarajbachu のプロフィール

記録された最新のソース活動: 2026年6月28日 19:37

SkillsMP カタログ更新: 2026年10月17日 00:09

収集済み skills: 1

GitHub スター: 0
GitHub フォーク: 0

リポジトリの README

README の元ファイル：README.md

README の元ファイルを読む

README 全文を表示

# Yolorun Send the agent. Keep the gate. Yolorun is a local-first workflow OS for agentic coding work. It helps agents create bounded feature workflows and recurring maintenance loops that run in isolated git worktrees, persist state under `.yolorun`, execute objective gates, review their own diffs, and stop for human review before risky actions. The point is not to let an agent run forever. The point is to turn vague prompting into a controlled loop: ```txt task source -> worktree -> agent role -> gates -> review -> fix -> summary -> draft PR or escalation ``` Agents can propose loops and feature workflows, but the Yolorun kernel decides whether they are runnable. ## Why Modern coding agents are good enough to do useful work, but long-running agent work gets messy fast: - prompts disappear into chat history - context rots - parallel agents fight over the same checkout - "done" becomes subjective - failed attempts are forgotten - unsafe changes creep toward merge or deploy Yolorun makes agent work explicit. Every workflow has a manifest, state files, logs, gates, budgets, permissions, and review rules. ## Install Yolorun is published on npm: ```sh npm install -g yolorun ``` Yolorun runs on Bun, so install Bun first if you do not have it: ```sh curl -fsSL https://bun.sh/install | bash ``` Check the CLI: ```sh yolorun --help ``` ## Quickstart Initialize Yolorun in a repo: ```sh yolorun init ``` Create and run a feature workflow in a worktree: ```sh yolorun feature "Build onboarding settings, add e2e tests, and make all gates pass" --worktree ``` Create a dependency-aware loop graph: ```sh yolorun graph create "Build a SwiftUI habit tracker with onboarding, reminders, stats, settings, and e2e tests" yolorun graph status yolorun graph run build-a-swiftui-habit-tracker-with-onboarding --max-parallel 3 yolorun graph report build-a-swiftui-habit-tracker-with-onboarding ``` Inspect workflow state: ```sh yolorun feature status yolorun feature review build-onboarding-settings-add-e2e-tests-and-make-all-gates-pass ``` Create a draft PR body or draft PR: ```sh yolorun feature pr build-onboarding-settings-add-e2e-tests-and-make-all-gates-pass --draft ``` Design recurring loop candidates for a repo: ```sh yolorun design yolorun validate ``` Enable a recurring loop: ```sh yolorun enable lint-fix --every 30m yolorun daemon start ``` ## Core Concepts **Feature workflows** are for bounded feature work: ```txt plan -> write-e2e -> implement -> run-gates -> review -> fix -> summarize ``` The supervisor repeats `implement -> run-gates -> review -> fix` until all required gates pass, review has zero blocking findings, or the workflow exhausts its budget. **Recurring loops** are for small scheduled tasks such as CI triage, lint fixes, flaky-test reproduction, and dependency review drafts. **Loop graphs** are dependency-aware groups of feature workflows. They let an agent design multiple loops, queue some behind others, and run ready nodes through the existing feature workflow executor. Use `yolorun graph reconcile <graph>` when child features were rerun or edited outside the graph supervisor. It syncs graph node status from the child feature manifests and workflow checkpoints. Use `yolorun graph report <graph>` when you need a durable completion record with every node, child feature state, handoff path, attempt count, and summary. **Worktrees** isolate agent runs so multiple workflows can happen in parallel without sharing one dirty checkout. **Gates** are objective commands such as `bun run typecheck`, `bun test`, `bun run e2e`, CI reproduction commands, or project-specific smoke tests. **State files** are the spine of the loop. They preserve what happened, what passed, what failed, and why the workflow stopped. ## What Works Today - `yolorun init` creates the local OS filesystem. - `yolorun design` proposes safe and draft recurring loops. - `yolorun feature "<request>" --worktree` creates and runs a bounded feature workflow. - `yolorun graph create "<goal>"` creates an editable feature-loop dependency graph. - `yolorun graph run <graph>` runs ready graph nodes as child feature workflows. - `yolorun graph reconcile <graph>` repairs stale graph status from child feature states. - `yolorun graph report <graph>` writes and prints `.yolorun/graphs/<graph>/REPORT.md`. - Feature workflows run planner, e2e writer, maker, reviewer, fixer, and summarizer roles. - Required gates are detected from project scripts. - Feature workflows persist `PLAN.md`, `ACCEPTANCE.md`, `STATE.md`, `REVIEW.md`, `state.json`, and JSONL run logs. - Workflows can resume from checkpoints. - Draft PR bodies can be generated, and GitHub CLI draft PR creation is supported. - Recurring loop daemon supports interval schedules and a small cron subset. - Codex and Claude adapters exist, plus fake adapters for tests. - The Yolorun agent skill is packaged and installable through the Vercel Skills CLI. ## Current Limits Yolorun is early. It intentionally does not pretend to have more automation than it does. - Feature workflows run foreground per process in v1. - The daemon currently schedules recurring `yolorun.loop` manifests, not feature workflow queues. - Graph execution respects dependencies, but v1 may run ready nodes conservatively instead of true parallel fibers. - True daemon-backed parallel feature workers are planned next. - External task inboxes such as GitHub Issues, Linear, Sentry, Slack, and databases are scaffolded as driver packages but not complete product integrations yet. - No auto-merge, no auto-deploy, and no unattended dependency changes. - Long-running work should be split into queue-shaped tasks, not one infinite while loop. ## CLI ```sh yolorun init yolorun design yolorun validate [loop] yolorun run <loop> yolorun feature "<request>" --worktree yolorun feature run <feature> yolorun feature resume <feature> yolorun feature status [feature] yolorun feature review <feature> yolorun feature pr <feature> --draft yolorun graph create "<goal>" yolorun graph run <graph> --max-parallel 3 yolorun graph resume <graph> --max-parallel 3 yolorun graph reconcile <graph> yolorun graph report <graph> yolorun graph status [graph] yolorun graph plan <graph> yolorun enable <loop> --every 30m yolorun enable <loop> --cron "0 9 * * 1" yolorun daemon start yolorun daemon stop yolorun daemon status yolorun status yolorun logs <loop> ``` When working inside this repository, use: ```sh bun run yolorun --help ``` ## Filesystem Yolorun keeps local state inside the repo: ```txt .yolorun/ yolorun.json OS.md skills/yolorun/SKILL.md daemon.json loops/<loop-id>/yolorun.json loops/<loop-id>/SKILL.md loops/<loop-id>/STATE.md loops/<loop-id>/runs/*.jsonl loops/<loop-id>/logs/*.log features/<feature-id>/yolorun.json features/<feature-id>/PLAN.md features/<feature-id>/ACCEPTANCE.md features/<feature-id>/STATE.md features/<feature-id>/REVIEW.md features/<feature-id>/state.json features/<feature-id>/runs/*.jsonl features/<feature-id>/logs/*.log graphs/<graph-id>/yolorun.json graphs/<graph-id>/PLAN.md graphs/<graph-id>/GRAPH.md graphs/<graph-id>/STATE.md graphs/<graph-id>/state.json graphs/<graph-id>/runs/*.jsonl graphs/<graph-id>/nodes/<node-id>/HANDOFF.md worktrees/ ``` ## Architecture ```mermaid flowchart TB CLI["apps/cli"] Daemon["apps/daemon"] Kernel["packages/core Yolorun Kernel"] Schema["packages/schema yolorun.json contracts"] Skill["packages/skill bundled Yolorun skill"] Agents["packages/agents architect / critic"] Runtime["packages/runtime supervisors"] Scheduler["packages/scheduler interval / cron"] Verifier["packages/verifier gates + permissions"] Worktree["packages/worktree git isolation"] Store["packages/fs-store .yolorun filesystem"] Graph["yolorun.graph dependency graphs"] Adapters["packages/adapters Codex / Claude"] Drivers["packages/drivers GitHub / Linear / Slack / Sentry / CI"] Telemetry["packages/telemetry logs / metrics"] CLI --> Kernel Daemon --> Kernel Kernel --> Schema Kernel --> Skill Kernel --> Agents Kernel --> Runtime Kernel --> Graph Runtime --> Scheduler Runtime --> Worktree Runtime --> Adapters Runtime --> Verifier Runtime --> Store Runtime --> Drivers Runtime --> Telemetry ``` More detail lives in [docs-architecture.md](./docs-architecture.md). ## Agent Skill Yolorun ships a skill at [skills/yolorun/SKILL.md](./skills/yolorun/SKILL.md). Install it so Codex, Claude Code, and other Skills-compatible agents know how to create worktree-backed feature workflows, choose gates, review diffs, and avoid pretending unfinished daemon features exist. Install from GitHub: ```sh npx skills add swarajbachu/yolorun --skill yolorun ``` Install to a specific agent: ```sh npx skills add swarajbachu/yolorun --skill yolorun -a codex npx skills add swarajbachu/yolorun --skill yolorun -a claude-code ``` Validate the skill while developing: ```sh bun run skills:list bun run skills:validate ``` Publishing to the Vercel/skills.sh ecosystem does not require a registry submission. Keep `skills/yolorun/SKILL.md` in this GitHub repo, make the repo accessible, validate it with `bun run skills:validate`, and users can install it with `npx skills add`. The skill can also surface on skills.sh after people install it. ## Monorepo ```txt apps/ cli/ daemon/ packages/ core/ schema/ skill/ agents/ runtime/ scheduler/ verifier/ worktree/ fs-store/ telemetry/ config/ testkit/ adapters/ drivers/ e2e/ benchmarks/ examples/ skills/ ``` Yolorun uses: - Bun workspaces - TypeScript strict mode - Effect as the runtime model - `effect/Schema` for manifests and logs - `@effect/platform-bun` for Bun-native services - `@effect/cli` as the CLI foundation - TanStack Intent for skill validation ## Development Install dependencies: ```sh bun install ``` Run checks: ```sh bun run skills:validate bun run typecheck bun test bun run build ``` Run the CLI from source: ```sh bun run yolorun init ``` Dry-run npm packaging: ```sh bun run npm:dry-run ``` ## Release Automation The `Release npm` GitHub Action runs on every push to `main`. It reads `apps/cli/package.json`, checks whether that version already exists on npm, and publishes only when the version is new. For tokenless publishing, configure npm trusted publishing for: - Package: `yolorun` - Repository: `swarajbachu/yolorun` - Workflow: `.github/workflows/release-npm.yml` As a fallback, add an `NPM_TOKEN` repository secret with publish access. Either path publishes with npm provenance. ## Roadmap Near-term: - Daemon-backed feature workflow queue. - Parallel feature workers with per-task leases. - External task inboxes for GitHub Issues, Linear, Sentry, Slack, CI, and bug-report databases. - True parallel graph workers with per-node leases. - Better merge and rebase orchestration across parallel worktrees. - Richer telemetry for attempts, gate failures, review findings, and accepted-change rate. Principles: - Prefer queue-shaped `for each` work over infinite `while` loops. - Prefer objective gates over subjective claims. - Preserve failed worktrees for inspection. - Escalate when permissions, budgets, gates, or review rules fail. - Keep humans in charge of merge, deploy, billing, auth, secrets, and dependency changes. ## Contributing Contributions are welcome. Good first areas: - improve docs and examples - add fixture repos under `e2e/fixtures` - harden daemon tests - improve driver packages - add more manifest validation rules - improve Codex and Claude adapter parsing Before opening a PR, run: ```sh bun run skills:validate bun run typecheck bun test bun run build ``` Please keep changes scoped and include tests for behavior changes. ## License MIT. See [LICENSE](./LICENSE).

リポジトリの README を読み込んでいます…

このリポジトリの skills

ソフトウェア開発者

1 件の職業カテゴリ · 100% 分類済み

skill

職業分類

説明

更新

yolorun

ソフトウェア開発者

Design and operate Yolorun worktree workflows for long-running agentic feature work, recurring loops, objective gates, safe reviews, and draft PR handoff.

原文の言語: 英語

2026-06-28

収集済み skill 1 件中 1 件を表示しています。

AI アシスタントでインストール

このプロンプトをコピーして、使用中の AI アシスタントに送ってください。

Follow https://skillsmp.com/skill-install/prompt.md to install Agent Skills from https://github.com/swarajbachu/yolorun.