v0.2.2

Reviewing existing work

Asked to "audit this" or "critique this", Form & Force reviews what exists and redesigns nothing uninvited. The review uses the same instruments and the same jury as a build, so its findings are evidence, not opinion. The agent follows the review path in SKILL.md and project-modes.md §13.

What it does

  1. Reads the repository, if there is one: inspect_repo.mjs for the stack and git state, slop_scan.mjs for failures and default patterns, with its coverage stated.
  2. Gathers rendered evidence with audit_page.mjs on the routes that matter, at narrow and wide widths, and filmstrips of the entrance and the main interactions with motion_strip.mjs.
  3. Runs the floor on what exists: each of the ten gates passes or fails, with evidence.
  4. Sits the jury: cold read, twin test, ten dimensions with evidence, findings with a location and a change, and a verdict.
  5. Reports the findings in order of level (concept, system, execution) and severity. It does not change code unless you ask.

Asking for a review

Audit https://example.com/pricing and /checkout. Report only; do not change anything.
Critique the landing page in this repo against its brief in docs/brief.md.
Tell me whether the idea is strong enough before we polish anything.

Say whether you want a report only or fixes for the failures (C findings and failed gates). Redesign is a separate decision: if the finding is that the idea itself is weak (a redirect verdict), the review says so and stops there.

What a review can reach

A review can conclude that a clean, well-built page has no idea of its own. That is the purpose of the twin test: a page whose only differences from the twin are of finish fails the concept dimension however well it is executed. The eval case jury-rejects-the-twin exists to test exactly this behaviour.

What it cannot see

  • Field performance: LCP and CLS are measured once, in a lab.
  • Behaviour that needs a person: how motion feels at speed, under a thumb, on the fiftieth use; screen-reader experience beyond what automated checks and a focus walk show.
  • Heavy WebGL in a headless browser, which can stall at the loader. The report says so and asks for a check in a headed browser or on a device.

For the agent

floor.md · jury.md · project-modes.md §13 · report-template.md