Full-lifecycle feature delivery with Agent Skills slash commands and Superpowers

What they did

The user wanted to add a small but real feature to an existing TypeScript web app: a saved-search feature with a REST endpoint, a UI toggle, and persistence. Instead of prompting the agent ad hoc, they installed the Agent Skills plugin in Claude Code via the marketplace and also added Superpowers for its brainstorming and verification workflows. They started with /spec, answering the interview-me style questions one at a time until the agent produced a short PRD with explicit non-goals and acceptance criteria. They then ran /plan, which broke the work into atomic tasks (schema migration, endpoint, client hook, UI toggle, tests, docs). Before building, they ran /constraints to lock in the quality bar: no new dependencies without approval, 90% coverage on new modules, and typed API contracts. They used /build to implement one slice at a time, with the agent writing failing tests first and committing each task individually. When a migration failed on a nullable column, the agent paused, surfaced the error, and proposed a fix rather than barreling ahead. They ran /test to confirm the suite was green, then /review for a five-axis code review that flagged an unhandled empty-state in the UI and a missing index on the new table. After fixing those, they ran /code-simplify to collapse a redundant helper, and /ship to generate the changelog and release checklist. Superpowers' verification step was used to double-check the acceptance criteria against the actual diff before merging.

How it went

The feature landed in a single afternoon with a clean commit history, one commit per task, and a test suite that covered the new endpoint and hook. The review caught two real issues before merge, and the constraints file meant no surprise dependencies crept in. The agent stayed on rails because each command scoped its own context. Polish level was high for a small feature; the main friction was approving the plan and re-running a couple of tasks after the migration fix.

Worth knowing

Per-skill npx installs copy only the skill directory, so shared references/ checklists are missing unless you do a whole-repo install or clone. The /review command name collides with Claude Code's built-in, so you need the namespaced /agent-skills:review form. /build auto runs autonomously after one plan approval, which is convenient but means you should keep the test gate on and watch for risky steps. The workflow assumes a repo with a working test runner and CI; without that, the verification steps degrade to manual checks.

Wiring

1) In Claude Code, run /plugin marketplace add addyosmani/agent-skills (use the HTTPS URL if SSH keys aren't set up), then /plugin install agent-skills@addy-agent-skills. 2) Add Superpowers through its own install path so its brainstorming and verification skills are available alongside. 3) Confirm the slash commands appear; remember /review needs the /agent-skills:review namespace. 4) Create a constraints file via /constraints so the quality bar is enforced across commands. 5) Run /spec, then /plan, then /build (or /build auto after approving the plan once). 6) Use /test and /review as gates before /ship. Skills also auto-activate by context, so API design and frontend work pull in the relevant skills without explicit invocation.

Try it yourself

Install the Agent Skills plugin for Claude Code and add Superpowers. Then, for the feature I describe below, run the full lifecycle: /spec to produce a short PRD with acceptance criteria and non-goals, /plan to break it into atomic tasks, /constraints to lock in our quality bar (no new deps without approval, tests required for new modules, typed API contracts), /build to implement one task at a time test-first with a commit per task, /test to confirm green, /review for a five-axis review, /code-simplify, and /ship to draft the changelog and release checklist. Pause and ask me before any risky step or on any failing test. Feature: add saved searches with a REST endpoint, persistence, and a UI toggle.