Skip to main content
plugin CX Codex

yylo — Claude Code plugin: what it does, how to install

yylo is a Claude Code and Codex plugin that isolates agent tasks in separate git worktrees and coordinates Pi and Codex subagents, verifying changes before merge.

MIT ★ 1,596 updated 2026-10-29

Install into Claude Code · Plugin marketplace All methods and platforms ↓
/plugin marketplace add hashgraph-online/awesome-codex-plugins
/plugin install yylo@awesome-codex-plugins
Open on GitHub

What yylo does

The plugin orchestrates coding agents straight from the command line. For each incoming task, it spins up an isolated git branch and worktree. This keeps your working tree clean while an agent modifies files.

Execution is handed off to Pi and Codex subagents. The tool enforces boundaries across task types, validation checks, and release readiness. Every change inside the repository is backed by receipts to confirm what was altered.

Your environment must support git worktree checkouts for this workflow to function. When branch operations fail on an untracked local conflict, the orchestrator halts execution.

What it helps with

  • isolate agent tasks into separate git worktrees
  • delegate coding jobs to Pi and Codex subagents
  • enforce release-readiness boundaries before merging
  • track repository modifications with structured receipts

Example prompts for the agent

  • Run the refactoring task inside an isolated worktree using the Codex subagent
  • Verify branch changes against the release boundary and validate receipts
  • Delegate bug reproduction to the Pi subagent in a dedicated worktree

How to install yylo

Plugin marketplace

/plugin marketplace add hashgraph-online/awesome-codex-plugins
/plugin install yylo@awesome-codex-plugins

Step by step, with commands for each platform: Claude Code plugins: marketplaces, installing, and building your own

Questions about yylo

How to install yylo in Claude Code?

First add the catalog with /plugin marketplace add hashgraph-online/awesome-codex-plugins. Then run /plugin install yylo@awesome-codex-plugins.

What does yylo do?

It manages coding workflows by setting up a dedicated git worktree per task. It directs work to Pi and Codex subagents and verifies typed boundaries before merges.

Which subagents does yylo work with?

The tool delegates tasks to Pi and Codex subagents. Each task runs within strict boundaries backed by change receipts.

Source text

yylo · MIT
# YYLO CLI

[![Mentioned in Awesome Coding Agents](https://awesome.re/mentioned-badge.svg)](https://github.com/kailiu42/awesome-coding-agents) [![Mentioned in Awesome Gemini CLI](https://awesome.re/mentioned-badge.svg)](https://github.com/Piebald-AI/awesome-gemini-cli) [![Mentioned in Collective AI Tools](https://awesome.re/mentioned-badge.svg)](https://github.com/hanishrao/collective-ai-tools) [![Mentioned in Awesome Agent Orchestrators](https://awesome.re/mentioned-badge.svg)](https://github.com/andyrewlee/awesome-agent-orchestrators) [![Mentioned in Site for Developers](https://awesome.re/mentioned-badge.svg)](https://github.com/currenjin/site-for-developers) [![Mentioned in Awesome Vibe Coding Resources](https://awesome.re/mentioned-badge.svg)](https://github.com/acvnace/awesome-vibe-coding-resources) [![Mentioned in Awesome AI Governance](https://awesome.re/mentioned-badge.svg)](https://github.com/agentrust-io/awesome-ai-governance) [![Mentioned in Awesome Local First](https://awesome.re/mentioned-badge.svg)](https://github.com/alantriesagain/awesome-local-first) [![Mentioned in Awesome Opensource AI](https://awesome.re/mentioned-badge.svg)](https://github.com/alvinreal/awesome-opensource-ai) [![Mentioned in Awesome AI Agents 2026](https://awesome.re/mentioned-badge.svg)](https://github.com/ARUNAGIRINATHAN-K/awesome-ai-agents-2026) [![Mentioned in AI Tools Directory](https://awesome.re/mentioned-badge.svg)](https://github.com/Aurelius77/AI-TOOLS-DIRECTORY) [![Mentioned in Awesome GPT](https://awesome.re/mentioned-badge.svg)](https://github.com/awesome-gptX/awesome-gpt) [![Mentioned in Awesome AI Agents](https://awesome.re/mentioned-badge.svg)](https://github.com/brandonhimpfen/awesome-ai-agents) [![Mentioned in Awesome AI Tools (Charles Teh)](https://awesome.re/mentioned-badge.svg)](https://github.com/charlesteh/awesome-ai-tools) [![Mentioned in AI Agent Infra List](https://awesome.re/mentioned-badge.svg)](https://github.com/chgaowei/ai-agent-infra-list) [![Mentioned in Awesome Agents Team](https://awesome.re/mentioned-badge.svg)](https://github.com/CiferaTeam/awesome-agents-team) [![Mentioned in Ultimate NodeJs Resources](https://awesome.re/mentioned-badge.svg)](https://github.com/DhanushNehru/Ultimate-NodeJs-Resources) [![Mentioned in Awesome AI Tools Resources](https://awesome.re/mentioned-badge.svg)](https://github.com/dr-mushtaq/Awesome-AI-Tools-Resources) [![Mentioned in Awesome Efficient Devtools](https://awesome.re/mentioned-badge.svg)](https://github.com/ejboy/awesome-efficient-devtools) [![Mentioned in Awesome Spec Driven Development](https://awesome.re/mentioned-badge.svg)](https://github.com/Engineering4AI/awesome-spec-driven-development) [![Mentioned in Awesome AI Agents (Frangel Barrera)](https://awesome.re/mentioned-badge.svg)](https://github.com/frangelbarrera/awesome-ai-agents) [![Mentioned in Awesome Autonomus AI Agents](https://awesome.re/mentioned-badge.svg)](https://github.com/HA2345567/awesome-autonomus-ai-agents) [![Mentioned in Awesome Codex](https://awesome.re/mentioned-badge.svg)](https://github.com/ichangyou/awesome-codex) [![Mentioned in Awesome Agent Orchestration Platforms](https://awesome.re/mentioned-badge.svg)](https://github.com/ishandutta2007/Awesome-Agent-Orchestration-Platforms) [![Mentioned in Awesome AI Tools (ishandutta2007)](https://awesome.re/mentioned-badge.svg)](https://github.com/ishandutta2007/awesome-ai-tools) [![Mentioned in Awesome Multi-Agent AI Harnesses](https://awesome.re/mentioned-badge.svg)](https://github.com/ishandutta2007/Awesome-Multi-Agent-AI-Harnesses) [![Mentioned in AI Landscape](https://awesome.re/mentioned-badge.svg)](https://github.com/kiliczsh/ailandscape.org) [![Mentioned in Awesome CLI TUI Software](https://awesome.re/mentioned-badge.svg)](https://github.com/lgaggini/awesome-cli-tui-software) [![Mentioned in Awesome AI (LinuxSuRen)](https://awesome.re/mentioned-badge.svg)](https://github.com/LinuxSuRen/awesome-ai) [![Mentioned in LLM Agents Devtools](https://awesome.re/mentioned-badge.svg)](https://github.com/M1n9X/llm_agents_devtools) [![Mentioned in Awesome AI Agent Tools](https://awesome.re/mentioned-badge.svg)](https://github.com/michielhdoteth/awesome-ai-agent-tools) [![Mentioned in Awesome ChatGPT Claude Agents](https://awesome.re/mentioned-badge.svg)](https://github.com/milisp/awesome-chatgpt-claude-agents) [![Mentioned in Awesome Codex CLI](https://awesome.re/mentioned-badge.svg)](https://github.com/milisp/awesome-codex-cli) [![Mentioned in Rise of Machine](https://awesome.re/mentioned-badge.svg)](https://github.com/planetabhi/riseofmachine) [![Mentioned in Awesome Agent Loops](https://awesome.re/mentioned-badge.svg)](https://github.com/rudy2steiner/awesome-agent-loops) [![Mentioned in Best of Agent Harnesses](https://awesome.re/mentioned-badge.svg)](https://github.com/RyanAlberts/best-of-Agent-Harnesses) [![Mentioned in Awesome AI Tools (sajadh76)](https://awesome.re/mentioned-badge.svg)](https://github.com/sajadh76/awesome-AI-tools) [![Mentioned in Awesome Pi Coding Agent](https://awesome.re/mentioned-badge.svg)](https://github.com/shaftoe/awesome-pi-coding-agent) [![Mentioned in Awesome Platform Engineering (shospodarets)](https://awesome.re/mentioned-badge.svg)](https://github.com/shospodarets/awesome-platform-engineering) [![Mentioned in Awesome AI Agent Governance](https://awesome.re/mentioned-badge.svg)](https://github.com/systempromptio/awesome-ai-agent-governance) [![Mentioned in Tool Directory](https://awesome.re/mentioned-badge.svg)](https://github.com/tejas-singh-0212/tool-directory) [![Mentioned in Awesome Vibe Coding (tysoncung)](https://awesome.re/mentioned-badge.svg)](https://github.com/tysoncung/awesome-vibe-coding) [![Mentioned in Awesome Agentic AI](https://awesome.re/mentioned-badge.svg)](https://github.com/yadavanujkumar/awesome-agentic-ai) [![Mentioned in Awesome AI Devtools](https://awesome.re/mentioned-badge.svg)](https://github.com/yeaight7/awesome-ai-devtools) [![Mentioned in Awesome AI Coding](https://awesome.re/mentioned-badge.svg)](https://github.com/youxufkhan/awesome-ai-coding) [![Mentioned in Awesome macOS](https://awesome.re/mentioned-badge.svg)](https://github.com/sraodev/awesome-macOS) [![Mentioned in Awesome AI Agents (Jenqyang)](https://awesome.re/mentioned-badge.svg)](https://github.com/Jenqyang/Awesome-AI-Agents) [![Mentioned in AI Tools Catalogue](https://awesome.re/mentioned-badge.svg)](https://github.com/robertsmrek/AI-Tools-Catalogue) [![Mentioned in Artificial Intelligence Universe](https://awesome.re/mentioned-badge.svg)](https://github.com/frangelbarrera/Artificial-Intelligence-Universe) [![Mentioned in Awesome AI Agents (NipunaRanasinghe)](https://awesome.re/mentioned-badge.svg)](https://github.com/NipunaRanasinghe/awesome-ai-agents) [![Mentioned in Awesome Gemini CLI Extensions](https://awesome.re/mentioned-badge.svg)](https://github.com/Piebald-AI/awesome-gemini-cli-extensions) [![Mentioned in Awesome AI Coding (dalisoft)](https://awesome.re/mentioned-badge.svg)](https://github.com/dalisoft/awesome-ai-coding) [![Mentioned in Awesome Windows](https://awesome.re/mentioned-badge.svg)](https://github.com/thechampagne/awesome-windows) [![Mentioned in Awesome Agent Workspaces](https://awesome.re/mentioned-badge.svg)](https://github.com/jimy-r/awesome-agent-workspaces) [![Mentioned in Awesome AI ML DL](https://awesome.re/mentioned-badge.svg)](https://github.com/neomatrix369/awesome-ai-ml-dl) [![Mentioned in Awesome Opensource](https://awesome.re/mentioned-badge.svg)](https://github.com/hadez8877/awesome-opensource) [![Mentioned in Awesome Antigravity Skills](https://awesome.re/mentioned-badge.svg)](https://github.com/ishandutta2007/Awesome-Antigravity-Skills) [![Mentioned in AI Tools Manager](https://awesome.re/mentioned-badge.svg)](https://github.com/ArshdeepGrover/ai-tools-manager) [![Mentioned in Awesome Agent Trust](https://awesome.re/mentioned-badge.svg)](https://github.com/CodeSigils/awesome-agent-trust) [![Mentioned in Awesome Mac](https://awesome.re/mentioned-badge.svg)](https://github.com/abordage/awesome-mac) [![Mentioned in Awesome Solo AI](https://awesome.re/mentioned-badge.svg)](https://github.com/yerdaulet-damir/awesome-solo-ai) [![Mentioned in Which Coding Tools](https://awesome.re/mentioned-badge.svg)](https://github.com/benjamincanac/whichcodingtools) [![Mentioned in Awesome Agent Tools](https://awesome.re/mentioned-badge.svg)](https://github.com/Awakehsh/awesome-agent-tools) [![Mentioned in Jump Skills](https://awesome.re/mentioned-badge.svg)](https://github.com/fabricioctelles/jump-skills) [![Mentioned in Awesome SRE Agents](https://awesome.re/mentioned-badge.svg)](https://github.com/last9/awesome-sre-agents) [![Mentioned in Awesome Agents](https://awesome.re/mentioned-badge.svg)](https://github.com/Scottcjn/awesome-agents) [![Mentioned in Awesome AI Plugins](https://awesome.re/mentioned-badge.svg)](https://github.com/hashgraph-online/awesome-ai-plugins)


YYLO is a command-line orchestrator for coding agents, repeatable workflows, and receipt-backed repository changes. It is for developers who want a quick agent loop and for project operators who need typed task, validation, merge, and release-readiness boundaries.

- npm package: [`@yylo/cli`](https://www.npmjs.com/package/%40yylo%2Fcli)
- Commands: `yylo` and `yy` (equivalent); `ypl` is `yy pi --live`
- Source: [yylo-dev/yylo](https://github.com/yylo-dev/yylo)

YYLO orchestrates work. [YYLO Ledger](https://github.com/yylo-dev/yylo-ledger) is the independent Git-native Record/task store, and [YYLO Benchmark](https://github.com/yylo-dev/yylo-benchmark) is the independent evaluation/evidence package. `yy ledger` and `yy benchmark` delegate to those separately installed CLIs; they are not bundled alternate implementations.

## Terminal

**Live agent session** — `ypl` (`yy pi --live`) runs an interactive terminal agent; tool calls, streamed output, and the status bar render in the terminal:

![yylo live terminal session](assets/yylo-tui-live-session.png)

**Execution summary** — after each run the orchestrator prints iterations, tool calls, cost, and session ids:

![yylo execution summary](assets/yylo-execution-summary.png)

## Quick start: install to first successful command

**Prerequisites:** Node.js 20.10 or newer, npm, and Git.

```bash
npm install --global '@yylo/cli@latest'
yy --version

mkdir yylo-demo
cd yylo-demo
git init
yy init --task "Document the onboarding path" --subagent pi
pwd # execute explicitly; watch is an optional read-only observer
```

A successful run prints the installed YYLO version, initializes `.juno_task/`, then emits a watch receipt with `"state":"COMPLETED"`, `"exit_code":0`, and nonzero `log_bytes`. This canary does not contact a model provider. In an empty, unborn Git repository, `yy init` creates the initial workspace commit, keeps the original branch as the product target, and creates a detached protected integration-owner worktree under the user state directory. Existing or dirty repositories are never committed or rearranged by this bootstrap.

Inspect the initialized workspace and exact command surface:

```bash
yy info --json
yy doctor workspace
yy --help
```

`doctor workspace` is intentionally nonzero when it finds an actionable topology problem; it never fetches or changes the workspace.

### Stable and prerelease channels

The stable npm channel is `@latest` (`0.2.2`). Prereleases remain on `@next`:

```bash
# Stable
npm install --global '@yylo/cli@latest'

# Explicit prerelease
npm install --global '@yylo/cli@next'
# Exact version matching this checkout (once published)
npm install -g @yylo/[email protected]

npm view '@yylo/cli' version dist-tags --json
yy --version
```

This unreleased source targets CLI **0.2.12**, requiring **Ledger 0.4.0 exactly** for universal
Record retrieval and storage-kind-prefixed IDs. Older Ledger 0.3.3, prereleases,
and later versions are rejected before delegation. Install the matching Ledger
explicitly once published; ordinary delegation does not auto-install a substitute.
A source bump neither publishes these versions nor activates installed runtimes.

Pin an exact version in CI. Installing `@next` is an intentional prerelease choice.
The first guarded release-helper checkpoint is exact `--set v0.1.0-rc.1`; later releases must use their separately authorized exact SemVer.

Next: [run an agent](#beginner-agent-workflow), [manage a typed task](#typed-task-and-merge-flow), or [build a managed workflow](#managed-workflows-and-evidence).

### Install agent skills

YYLO skill content is versioned independently in the public
[`yylo-dev/yylo-skills`](https://github.com/yylo-dev/yylo-skills) repository and
is not bundled in `@yylo/cli`. This CLI source requires stable skills `^2.1.1`, declared
in `package.json` as `yyloSkills.version`. Install the latest compatible stable
release, or pin an exact compatible version once it is published. Source version
2.1.1 is not evidence that this skills release is available:

```bash
yy skills install
yy skills install --version 2.1.1
yy skills update
yy skills status
```

Fresh Simple `init` installs skills by default; use `yy init --mode simple --no-skills`
to opt out. Guided Simple setup accepts the same flag. Previews remain offline and
saved fresh plans bind the choice. Otherwise use `skills install` or `skills update`
explicitly. These skill acquisition paths access the network. Acquisition is
staged through `npx skills add` first and falls back to a shallow exact-tag Git
clone. The seven user-intent-first skills (`artifact-yylo`, `ledger-tasks-yylo`,
`plan-ledger-tasks-yylo`, `ralph-loop-yylo`, `understand-project-yylo`,
`wiki-yylo`, and `workflow-yylo`) are copied to `.agents/skills`,
`.claude/skills`, and `.pi/skills`. During install/update, unchanged receipt-owned
copies upgrade automatically without `--force`. Customized or unrecorded differing
canonical directories refuse before mutation; review before explicitly forcing
replacement. An explicit install/update retires a legacy YYLO skill only when its
local install record and current digest prove it unchanged; customized/unrecorded
legacy and unrelated skills remain preserved. Status flags these legacy copies
because they can still inject obsolete instructions; review them separately,
never infer cleanup authority from successful installation.

`skills list` and `skills status` inspect local bytes and receipts offline. Status
reports the required range and outdated releases even when their hashes match.
No compatible published release means installation fails without changing skills;
it never falls back to an old release or an unsupported future major. Installing
upgrading the CLI alone does not install/update skills, and `scripts update` does not own
skills. Maintainers must publish the reviewed immutable skill release before
shipping a CLI that requires it; publication remains separately authorized.

## What YYLO owns

| Need | Public surface | Boundary |
| --- | --- | --- |
| Agent run | `yy pi`, `yy start`, and the agent aliases listed by `yy --help` | Provider credentials and model availability remain external. |
| Session continuity | `continue`, `clone`, `branches`, `switch`, `continuity` | Scope state is isolated and explicit; cleanup is planned and reversible. |
| Optional observation | `watch status|await|follow` | Read existing execution evidence; never launch, retry, cancel, or complete tasks. |
| Validation evidence | `evidence run|status|await` | Content-addressed task evidence tied to exact inputs. |
| Repository topology | `info`, `where`, `doctor workspace`, `integration` | Read-only discovery is separate from guarded sync/repair/push. |
| Feature lifecycle | `task start|run|status|checkpoint|preflight|finish` | Implementation belongs in the returned exact-base task worktree. |
| Protected delivery | `merge status [TASK_ID]`, `merge land TASK_ID`, `merge project TASK_ID` | One-task native Git composition plus expected-old ref update; Ledger projection is separate and retryable. |
| Maintainer release | repository `scripts/release-cli.sh` | Prepare is read-only; publish needs separate authority. |
| Records/evaluation | `ledger`, `benchmark` | Transparent delegation to independently installed canonical packages. |

Run `yy --help` for the complete top-level inventory of your installed version; each listed command prints its own usage when invoked with `-h`. The old `lifecycle` command is removed; use typed `task` and `merge` commands.

### Stable machine output

Use `--format json|ndjson --raw` on task, merge, and integration commands, or Ledger's `-f json|ndjson --raw`, for a strict stdout-only data channel. The `capabilities` command publishes command-specific projections in either machine format. See [Machine output contract](docs/machine-output.md) for the versioned envelope and migration policy.

## Beginner agent workflow

Install the coding agent you intend to use and configure its provider credentials separately. Pi is optional and supports multiple providers:

```bash
npm install --global '@mariozechner/pi-coding-agent'
yy pi --help
yy pi --no-session 'Summarize this repository and make no changes'
```

The final command may contact the configured model provider. `--no-session` prevents Pi session persistence; it does not disable provider usage.

For a reusable prompt or shell-sensitive text, prefer a file:

```bash
printf '%s\n' 'Explain the test layout. Do not edit files.' > prompt.md
yy pi --prompt-file prompt.md --no-session
```

Interactive Pi uses the `ypl` shortcut:

```bash
ypl 'Inspect the current task'
```

`ypl` expands to `yy pi --live`. Live mode requires an interactive terminal and enabled Pi extensions.

### Controlled iterations

```bash
yy -s pi -m :gpt -i 3 -p 'Implement the next small verified increment'

yy loop -n 2 \
  --step 'yy pi "Implement the next increment"' \
  --step 'npm test'
```

Quote prompts so the shell does not expand backticks or `$()` before YYLO receives them. `-i` bounds iterations inside an agent invocation; `yy loop -n` bounds the outer command workflow.

The loop's `-n/--iterations` bounds the outer command workflow; agent
`-i/--max-iterations` still bounds iterations inside an agent invocation.
Inline steps are repeatable:

```bash
yylo loop -n 5 \
  --step 'yy pi "Implement the next increment"' \
  --step 'yy cc "Inspect and improve your work"' \
  --step 'npm test'
```

For a reusable workflow, save the same contract as `flow.yaml`:

```yaml
iterations: 5
continuity: iteration
on_error: continue
steps:
  - run: yy pi "Implement the next increment"
  - run: yy cc "Inspect and improve your work"
  - run: npm test
```

```bash
yylo loop --workflow flow.yaml
```

Every step receives one-based loop metadata through `YYLO_LOOP_ID`,
`YYLO_ITERATION`, `YYLO_ITERATION_COUNT`, `YYLO_STEP`, and `YYLO_STEP_COUNT`.

### Ledger references in prompts

Use `##ABC123`, `##readable-slug`, or `##{readable-slug}` to include a Ledger
Record by exact ID, slug, or retained alias. Existing `## ABC123` spacing works.
Tasks, Wiki/PDR/workflow documents, and artifacts share native Record lookup;
ambiguous identities are never resolved by guessing. Brace references next to
punctuation that could be part of a slug (for example `##{design.v2}.`).
Missing references remain unchanged. Lookup errors produce a manual-resolution
warning; references do not execute workflows or recursively expand Record text.

Hydration is bounded to 32 distinct identities, a shared 30-second lookup budget,
and 64 KiB of inserted context (16 KiB per Record). Oversized Records carry a
retrieval command instead. Artifacts include metadata; only verified inline UTF-8
text up to 4 KiB is included. Binary, local, and external artifact bytes are not
automatically read or downloaded. Use `yy ledger record get ID_OR_SLUG` for
explicit retrieval. Immutable IDs remain authoritative for task lifecycle and
Record relations.

## Models and project shortcuts

`yy pi --help` is the source of truth for shipped aliases. Current Pi shortcuts include:

| Shortcut | Resolved model |
| --- | --- |
| `:luna` | `openai-codex/gpt-6-luna` |
| `:sol` | `openai-codex/gpt-6.1-sol` |
| `:gpt` | `openai-codex/gpt-6-astra` (Pi default) |
| `:astra` | `openai-codex/gpt-6-astra` |
| `:mini` | `openai-codex/gpt-5.6-terra` |
| `:sonnet` | `anthropic/claude-sonnet-4-6` |
| `:opus` | `anthropic/claude-opus-4-6` |

Aliases are subagent-specific; for example, Pi and Codex do not share the same `:mini` mapping.

Set a per-project default:

```bash
yy pi set-default-model :sol
```

Add project shortcuts in `.juno_task/config.json`:

```json
{
  "modelShortcuts": {
    "pi": {
      ":team-default": ":sol"
    }
  }
}
```

Project shortcuts are scoped to the selected subagent and can reference shipped or project shortcuts. Unknown targets, malformed data, and cycles fail with an actionable error. Managed Workflow Runner model authorization is separate: explicit selectors must be exact members of `workflowModels`; an unflagged `yy pi` continues to inherit the configured default.

## Observable local commands

`watch` optionally observes existing run evidence without owning execution:

```bash
npm test # execute explicitly in the authorized workspace
# If an existing producer has published a watch-compatible run:
yy watch status RUN_ID
yy watch follow RUN_ID
yy watch await RUN_ID
```

`RUN_ID` is an existing run identity, not one created by `npm test`. `watch exec` is retired. All watch operations are read-only, and interruption stops only the observer. Missing or malformed terminal evidence is not success. Process exit, semantic outcome, cleanup, and task completion are distinct; `task finish` verifies delivery. Historical run directories are preserved.

## Managed workflows and evidence

Fresh `yy init` installs managed scripts, prompts, and wiki guidance under `.juno_task/`. Update checksum-managed assets with:

```bash
yy scripts update
```

Locally customized files are preserved and a package candidate is written to a managed-conflict path. `--force` is an explicit replacement operation and creates backups; inspect conflicts before using it.

### Workflow Runner

Use a reviewed YAML workflow for ordered, repeatable steps:

```bash
./.juno_task/scripts/workflow_runner.sh \
  --init-example agent-chain .juno_task/workflows/agent-chain.yaml
./.juno_task/scripts/workflow_runner.sh lint \
  --workflow .juno_task/workflows/agent-chain.yaml
./.juno_task/scripts/workflow_runner.sh \
  --workflow .juno_task/workflows/agent-chain.yaml --dry-run \
  --print-output none --no-print-step-stdout
```

A workflow run retains rendered command identity, stdout/stderr, responses, session IDs, declared receipt hashes, attempts, and terminal manifests. `--tmux` creates an observer session; it does not detach the producer.

If a producer was interrupted, diagnose before mutation:

```bash
./.juno_task/scripts/workflow_runner.sh recover-attempt RUN_DIRECTORY --dry-run
./.juno_task/scripts/workflow_runner.sh doctor RUN_DIRECTORY
```

`RUN_DIRECTORY` is the printed durable run path. Recovery resumes only the first invalid step after verifying unchanged successful evidence. Never edit historical manifests to make them reusable.

### Task validation evidence

At a clean coherent task commit:

```bash
yy task checkpoint TASK_ID
yy evidence run TASK_ID
yy evidence status TASK_ID
yy evidence await TASK_ID
```

These commands plan and retain exact-input validation evidence. They do not finish or merge the task.

## Typed task and merge flow

This section is for repositories initialized with the current controller/task policy. Run lifecycle commands from the registered metadata controller. Never edit the integration-owner checkout.

### Deterministic delivery, external implementation

```bash
yy task start TASK_ID
# External agent edits, tests and commits in the returned worktree.
# Retain the returned token privately for gated commands:
yy task finish TASK_ID --lease-token <current-token>
yy merge land TASK_ID
# Only if automatic Ledger projection needs recovery:
yy merge project TASK_ID
```

`TASK_ID` is a Ledger task ID. Start prepares an isolated workspace; finish
verifies the clean committed result and queues it. Merge uses native Git and
automatically records verified integration in Ledger. Tests and semantic review
remain explicit project checks. Task run/resume and automatic implementation
budget recovery are retired. No implementation model, retry or repair engine is
hidden behind watch. Process exit is not completion. For interrupted work,
preserve the existing workspace and inspect status/lease-status before explicit
continuation; do not automatically reset or replay historical attempts.

### Choose Simple or Advanced

For a new Simple project, run `git init`, then `yy init --mode simple` to
initialize immediately. Add `--directory PATH` to select an existing Git root.
For guided setup, `yy init` asks a final mode question:

- **Simple (recommended to start):** code, notebooks, notes and Ledger in one
  checkout. Agents share files; `yy task local` is bookkeeping, not delivery.
  No implicit commits, worktrees or package installation.
- **Advanced:** a separate controller, isolated task worktrees and managed merging.

Use `yy init --interactive --mode simple` to skip the mode question. Inline
initialization retains Advanced behavior; `yy init "Build an API" --mode advanced`
makes that explicit. **Plain `yy init --mode simple` now writes files**, without
confirmation or an external plan. Existing preview automation must add `--dry-run`
(zero writes) or use optional `--plan-file` then `--apply-plan` for reviewed plans.
Initialization installs no dependencies and does not certify agent readiness.
Normal initialization does not convert existing workspaces. For a settled registered
metadata-only controller, preview then apply a fresh Advanced → Simple copy:

```sh
yy init --mode simple --from-advanced /controller --directory /new-project --plan-file /external/conversion.json
yy init --mode simple --apply-plan /external/conversion.json
```

Source workspaces remain unchanged. Old instructions and settings are retained
inactive for manual review. Simple → Advanced is not supported, including with
`--force`.
Read [Simple and Advanced workspaces](docs/simple-workspaces.md) for exact commands,
readiness recovery, persistence, and shared-checkout limits.

### Manual implementation path (managed mode)

```bash
yy task start TASK_ID
# Change directory to the worktree printed by start.
# Read its AGENTS.md/CLAUDE.md, implement, run focused tests, and commit.
yy task finish TASK_ID
yy merge status TASK_ID
yy merge land TASK_ID
yy merge project TASK_ID
```

The ordinary path is clean commit -> `yy task finish TASK_ID`.
Optional read-only `yy task preflight TASK_ID` diagnoses admission before finish;
it is not a prerequisite. Delivery remains a separate one-task operation.

Safety invariants:

1. `task start` freezes the protected target SHA, creates a dedicated branch/worktree, and completes configured dependency hydration before reporting `WORKING`.
2. Product edits and focused tests occur only in that task worktree. Controller metadata and integration-owner product bytes are separate authorities.
3. `preflight` is optional and read-only. `finish` independently enforces scope, dirty-byte, runtime, hydration, requirements, ownership, and selected-validation checks, requires a clean committed tip, and queues it; it does not merge.
4. Tests and semantic reviews are explicit project checks outside merge. Merge launches no models, chooses no reviewers, schedules no suites, and maintains no validation cache.
5. `merge land` selects one immutable task source, composes in a private detached candidate, and uses Git expected-old ref protection. A moved target requires recomposition and renewed candidate checks.
6. A conflict remains private to its task and cannot block an unrelated task. Preserve conflicts and unrelated dirty bytes; do not reset, stash, force, rebase, squash, or clean to bypass them.
7. Git success and Ledger projection are separate. A projection retry never repeats Git integration and never blocks another independent land.

Observation commands are safe to repeat:

```bash
yy task status TASK_ID
yy task doctor TASK_ID
yy merge status
yy merge status TASK_ID
```

Merge status reports only independently landable active tasks and always reports
`model_calls: 0`. Historical queue receipts remain immutable data, not an
executable compatibility surface.

## Maintainer npm release

Normal tasks and `yy merge` own integration. After an ordinary version bump is integrated, maintainers use `scripts/release-cli.sh prepare CLI_VERSION BENCHMARK_VERSION`, obtain explicit publication approval, and then run the separate `publish` command. The CLI has no release command.

## Controller generation upgrades

Recognized controller generations migrate automatically on first task/agent use;
controller-local scripts, policy and executable identity move together. Diagnose
without changing state using `yy scripts generation doctor`. Unknown/customized
bytes are preserved, and fallback requires a trusted retained executable plus
scripts. See [upgrade, recovery and validation guidance](docs/controller-generation-upgrades.md).
Maintainer release preparation gates the exact npm tarball and binds its immutable
acceptance report; this does not authorize publication.

## Receipt-bound workspace relocation

When an entire controller is moved between machines, do not rewrite lifecycle JSON or historical receipts by hand. From a clean physical controller checkout, create and review an external plan, apply it once, then verify its immutable receipt:

```bash
node juno-code/scripts/workspace-relocation.mjs plan --controller "$PWD" \
  --map /old/controller=/new/controller --map /old/worktrees=/new/worktrees \
  --output /secure/relocation-plan.json
node...

© YYLO contributors · MIT · hashgraph-online/awesome-codex-plugins

Similar plugins

All plugins for Agent setup: memory, context, orchestration →

From the same set391

The whole set →
abuzzhive Connects an agent to AbuzzHive public Q&A boards, where agents post errors and propose fixes to each other. A hook triggers find_by_error automatically after every failed shell command. CX Codex ★ 1,596repo stars call-me call-me lets AI agents call or text your iOS or Android phone, read prompts aloud, and capture your voice or typed response back into the prompt. CX Codex ★ 1,596repo stars codex-zcode-bridge codex-zcode-bridge is a plugin for Claude Code and Codex that sends tasks to a local ZCode agent and returns progress for diff review against acceptance criteria. CX Codex ★ 1,596repo stars zagrosi-forge zagrosi-forge turns a broad brief into section specs, a plan and test-first code with per-section records. Useful when a task is too big for one pass. CX Codex ★ 1,596repo stars wtfm wtfm is a Claude Code and Codex plugin that turns a codebase into illustrated help articles. Ask for a guide in plain language, such as instructions for inviting a teammate. CX Codex ★ 1,596repo stars writers-loop writers-loop runs writing as a reviewable loop: frame the task, plan, draft, critique, then revise only what you approve. Style memory stays off unless you opt in. CX Codex ★ 1,596repo stars workflow-kit workflow-kit manages product delivery from initial vision to final deliverables. It pairs every worker agent with an enforced domain-expert reviewer. CX Codex ★ 1,596repo stars wingman Wingman keeps project knowledge in the repository, checks data contracts, and helps agents find existing features. Use it before changing code or wiring real data into a project. CX Codex ★ 1,596repo stars

Project sponsors

Companies that keep this analytics open