map-review

Generate a project-local skill per app that drives the real app the way a user does — web, CLI/TUI, desktop, service, any stack — and proves behavior with…

The presentation contract for every question put to the human — the opening, facts vs decisions, cold-reader blocks, domain language, recommendations, prose…

Work all AFK tickets (research, agent-doable tasks) on a wayfinder map's frontier in parallel background agents, while the human handles the HITL tickets.

Repair a document that many agent sessions have edited so it reads to a cold reader — one holding only the doc and what it points at.

Decide the shape of code before it's built — modules, interfaces, function signatures, data shapes, and the seams tests will live at.

The find → validate → propose → validate → route contract every review gate runs — adversarial validation in chunked chains, schema-typed verdicts, edge-field…

Which model every spawned agent runs on — three tiers (executor, decider, auditor), resolved from a policy file (repo, then home, then defaults) so the…

Trace where skills went wrong in past sessions — open snags from `/snag`, or a sweep of a skill's recent transcripts — back to the skill text that caused it…

Grow a feature slice by slice — settle only what the next slice needs, hand its build to a background agent, look at what landed, and let that pick the next…

Tune the /iterate loop from its own runs — a reader per session since the last pass, the build and review workflows timed, recurring friction traced to the…

Keep a project's verification skill and feature map honest — a source reader per feature, one live pass driving every feature, at most one PR of proven…

Where a repo's reviewable coding standards live — CONVENTIONS.md at root, per-directory deltas in monorepos, nearest scope wins — and how reviewers find them.