Generates a project-local verification skill at .cursor/skills/verify-<app>/ that drives the real app the way a user does, for any language, framework, or platform. The premise: every serious project needs a scripted way to launch the app, exercise a feature the way a user would, and capture evidence — and most repos have none.
This is pattern-tooling-leverage pointed at the hardest thing for an agent to fake. gstack solves the same problem with shipped infrastructure (a persistent browser daemon behind gstack-browse); pstack instead generates a bespoke harness per repository, which is why it works for a CLI, a TUI, a desktop app, an API, a mobile app, or a library rather than only for the browser.
It interviews the repo, not the user, answering five questions from the codebase and asking only what cannot be observed: the surface a user actually touches, how the app runs locally (preferring the repo’s own documented dev command), how an agent can drive it programmatically (existing Playwright or Cypress specs, expect scripts, PTY helpers, curl-able endpoints, a debug port — only then a generic recipe), what evidence can be captured, and whether two instances can be isolated side by side. If the checkout does not build or start, that is fixed or reported precisely first, because a skill written against a broken base teaches wrong steps.
The generated skill has a fixed shape, and its author is told who reads it: you write the generator’s output for the next agent, not for a human — it will be read cold, mid-task, by an agent that has never seen the app. Sections are Launch (with a readiness signal and teardown), Doctor (one read-only “is this instance worth driving?” check), Drive (real selectors and commands from this repo, preferring stable handles like ARIA labels, data attributes, prompt strings, and route paths over coordinates and tab order), Evidence, Cleanup, and Helpers.
The proof standards written into the Evidence section are the framework’s operational definition of verification: exercise the real user path, not internal setters or test-only endpoints; capture the action and the resulting state, not just the final screen; verify side effects (files written, rows inserted, messages sent) alongside what is visible; and mock only where a production boundary already isolates the external system. A pointed one closes it — when the safe path is a dry-run or test mode, verify what it actually skips by observing files, network, and git refs rather than trusting its name, because some dry-runs still touch the network or open a browser.
Cleanup never kills by process name, only what the run started, and proof artifacts survive the teardown in a location the skill names.
See Also
- pstack-maintain-verification-skill — the upkeep loop for what this generates.
- pstack-tdd — the cheap-local-test path; this is the answer when there is none.
- gstack-browse — the shipped-infrastructure alternative to a generated harness.
- stage-validate — the canonical stage this implements.