OpenSpec Plus Tasks: Vertical Slices for Reliable Agent Plans

Where this fits

You're mid-change with an OpenSpec spec, and your agent generates a tasks.md that lists work layer-by-layer: first all models, then all services, then all UI. Halfway through, you realize the first vertical feature isn't testable until everything else lands. openspec-plus-tasks fixes this by forcing each task group to be a vertical slice—end-to-end through the layers—and by adding a subagent review pass before the file is written. It activates automatically during /opsx-new, /opsx-continue, or any OpenSpec tasks phase, so you don't have to remember to invoke it.

Why agents benefit

• Enforces vertical-slice discovery before writing — the skill refuses to write tasks.md until Phase 1 analysis is complete, preventing premature, horizontal task lists.
• Adds a subagent review gate — after drafting, a separate review pass checks that each task group ends in a testable outcome, catching gaps before you commit to a plan.
• Anchors every task group to a user-verifiable result — the agent must define what "done" looks like per group, making progress measurable and demos possible.
• Works automatically with OpenSpec workflows — triggers on openspec-new-change, openspec-continue-change, or openspec-explore, so the discipline applies without extra prompts.
• Keeps format ownership with the schema template — the skill focuses on thinking and review, not on inventing a new format, so it stays compatible with your existing OpenSpec setup.

Practical scenarios

Refactoring a payment service

You're splitting a monolithic payment module. The agent's first draft lists "update all database schemas" then "rewrite all API endpoints"—classic horizontal layering. With this skill, it instead groups tasks by feature: "refund flow end-to-end" (DB + API + UI) and "recurring billing end-to-end"—each independently testable after its last task.

Adding a new user role

Your spec adds an "admin" role. The agent must plan tasks that touch auth middleware, permission checks, and UI navigation. The skill forces a single vertical slice for "admin can view audit logs" and another for "admin can assign roles," each with a clear verification step (e.g., "log in as admin, navigate to audit tab, confirm entries load").

Reviewing a stale tasks.md

You have an existing tasks.md from a previous session that's vague. Invoking openspec instructions tasks triggers the skill's review mode. It identifies task groups that lack testable outcomes and suggests rewrites—without you having to manually audit the whole file.

Add it to your agent workflow

Install the skill into your agent's skill directory:

npx skills add sudokar/openspec-plus --skill openspec-plus-tasks

Then simply run your normal OpenSpec flow—for example, /opsx-new or openspec-new-change. The skill activates automatically and guides the tasks.md generation. You can also manually trigger it with openspec instructions tasks if you need to refine an existing file. For a broader view of related skills, see this skill's listing.

Compare before adopting

Before you add it, check three things: quality signals—does the skill's README show concrete examples of vertical-slice grouping? maintenance freshness—is the repo actively updated (version 1.3.0 suggests yes)? alternatives—other OpenSpec task generators exist, but most don't enforce the subagent review step. Also consider workflow fit—if you rarely use OpenSpec's tasks phase, this skill adds little value.

Why it is worth tracking

With 154 stars and a clear, opinionated design, openspec-plus-tasks signals a maturing ecosystem around OpenSpec agent workflows. Its rigid enforcement of vertical slices and review gates directly addresses a common failure mode in agent-generated plans: untestable, layer-ordered tasks. Track it if you rely on OpenSpec for change management—it could become a default part of your agent's planning discipline, and its approach to subagent review is worth borrowing even if you don't adopt it wholesale.