@windyroad/planning

Keep the outcome clear while the plan evolves. Confirm a SMART objective, deliver one reviewed increment, then verify, learn and replan Now, Next and Later. Maturity: Experimental.

Part of Windy Road Agent Plugins. Read the outcome-led planning guide.

Install

For Claude Code:

npx @windyroad/planning

For Codex:

npx @windyroad/planning --runtime codex --scope user

Use --runtime both for both runtimes. Restart the runtime after installation and check its plugin and skill lists.

This package installs planning, contribute and GitHub tracking setup, with no hooks or reviewer agents. ITIL is not required. Initial and materially revised executable plans require five independent critics: architecture, JTBD, risk, cognitive accessibility and planning. Use available independent reviewers, or install the architect, JTBD and risk-scorer plugins for your runtime. If a substantive review cannot be obtained, the skill reports the missing capability and holds dependent implementation.

Arrange the cognitive accessibility and planning critics through available independent reviewers too. Routine task claims and status updates within an assessed plan do not restart all five reviews.

Before an increment is declared complete, its owner obtains an independent acceptance check through the planning critic. The critic inspects actual results and affected saved records; discrepancies are repaired and rechecked. An existing independent review covering these criteria is reused. Individual task completion does not trigger this increment-level check.

Use

In Claude Code:

/wr-planning:outcome-led-planning

In Codex:

$wr-planning:outcome-led-planning

Describe the result you want and any constraints. The agent asks one question at a time when information is missing, then plays back the complete objective and waits for your confirmation before planning. It makes one Now increment INVEST-shaped without requiring story documents unless your project needs them. After delivery it verifies the result, learns and reassesses the revised plan.

Remember the project's tracking choice

Planning and contribute read the same project configuration. Use the location declared by the project, or .agents/planning.json by default. When no choice exists, the skill asks GitHub Issues or Local? and waits. It saves and reads back your answer, then reuses it across skills and fresh sessions. An existing explicit choice needs no repeat question.

{"version": 1, "backend": "local"}

Use "github" for GitHub Issues. Ask to change the project tracking choice when needed; the agent preserves unrelated fields and contributor mappings. Invalid or unsupported settings are preserved while their repair is clarified. Configuration is agent-read data and grants no authentication or permissions. Installers do not create an adopter project's configuration.

Local mode uses the agreed plan and channel without GitHub setup. A runtime may protect the default configuration directory from writes. Use a permitted project-declared location, such as planning.json, or obtain the required write access; a denied write does not establish persistence.

With GitHub selected, the agent loads its bundled protocol and uses outcome → increment → task issues. Increments have Now, Next or Later horizons; execution status is separate. Contributors check existing claims, identify their account and agent/session, and coordinate dependencies and evidence in the increment thread. The outcome owner verifies delivery and replans. A GitHub Project can mirror those issues when available; it is optional. Conversation and local planning remain available. For missing selected setup, planning and contribute call the bundled setup skill. Complete setup and equivalent custom mappings are reused; Project fields and views apply only to a selected or required Project.

Each contributor uses their own access. Tracker selection does not authorize new messaging, broader permissions or persistent automation. Issue comments do not guarantee an idle agent will wake up.

Set up GitHub tracking

Invoke /wr-planning:setup-github-tracking in Claude Code or $wr-planning:setup-github-tracking in Codex when you need to configure or repair selected GitHub tracking. Planning and contribute call it when setup is missing or incomplete. It reuses the same project configuration and returns to the current work without interviewing you about the outcome again.

Issues-only tracking needs no Project. For a selected Project, the skill reuses equivalent setup and configures missing fields and Outcomes, Increments and Tasks views. Increments uses a table grouped by Horizon with Show hierarchy On. Changed settings are saved and reopened; related task children remain accessible beneath their increment. Existing custom views, ownership and history are preserved. Local tracking bypasses this setup.

Join an outcome with contribute

Invoke /wr-planning:contribute in Claude Code or $wr-planning:contribute in Codex. It reuses the saved backend and finds useful direct or indirect help within the current assessed Now. Contributors inspect current work and claims, then create or reuse one bounded task just before starting it. The owner keeps acceptance, sequence and integration clear without allocating every task.

A starting battle plan exposes useful opportunities and dependencies; it does not precreate or reserve the whole backlog. Finish, hand over or release one claim before taking another. Planning MUST NOT add speculative risk-mitigation tasks: the risk scorer derives necessary treatments from the applicable policy and register. Existing required controls remain binding.

Saved project tracking and contribute require planning 0.4.0 or later.

| Skill | Purpose | Maturity | | --- | --- | --- | | /wr-planning:outcome-led-planning | Confirm a SMART outcome and iteratively deliver through reviewed Now, Next and Later increments | Experimental | | /wr-planning:contribute | Find and carry out useful bounded help within the current assessed increment | Experimental | | /wr-planning:setup-github-tracking | Configure missing selected GitHub issue tracking and Project views | Experimental |

Existing installations

The existing /wr-itil:outcome-led-planning and $wr-itil:outcome-led-planning commands remain usable through the full compatibility skill bundled in ITIL. They do not require this package. New installations can choose planning independently.

An existing personal $outcome-led-planning skill remains separate. Installing, updating or uninstalling this plugin does not replace that personal skill.

Update or remove

npx @windyroad/planning --update
npx @windyroad/planning --uninstall

Include the same runtime and scope flags used for installation.

Maintain the source

Canonical planning and GitHub setup live under skills/. In the monorepo, run npm run sync:planning-skill after changing either; the checked sync keeps both ITIL compatibility skills and their supporting resources equivalent while preserving their names. The personal skill is user-managed and is not a synchronization target.