plan-reviewlisted
Install: claude install-skill aliasunder/agent-plugins
# Plan Review
Adversarial critique of an implementation plan before implementation starts. This
skill examines **premises, not mechanisms**: whether the plan solves a real problem,
whether its assumptions hold, whether the chosen approach beats the alternatives —
including doing nothing — and whether its own verification steps are safe to run.
How this differs from ship-check: ship-check reviews the mechanism that was built.
Plan-review examines whether it is the right mechanism to build. The founding case:
a row cap on client registrations passed all four ship-check phases (9 code-level
findings, all mechanism-correct) — and only manual review noticed that one address
operating under the existing rate limit fills the cap in ~75 minutes and locks the
owner out. The design was a denial vector; no amount of code review would have said
so, because the code faithfully implemented the flawed premise.
## What counts as a plan
Any of: a plan document (`plans/*.md`), a task note (`## Problem` + What + Done
when), a plan-mode plan, or an approach stated in conversation. The reviewer
produces **findings only — never edits the plan.** Revision belongs to the plan's
author, who holds context the reviewer doesn't.
## Before starting
1. Read the plan artifact IN FULL, plus the task note / card it implements and any
documents it links.
2. Load the target repo's AGENTS.md (conventions the plan must fit).
3. Load vault context where available (ToolSearch the `vault_*` schemas if
d