Skip to content
Development
Skill

/cad-plan-review

On-demand adversarial plan review of a phase PLAN.md before code - for a hand-written, imported or just-edited plan (/cad-plan fires this itself)

From plugin
cadence
534 skills30 agents3 hooks
Install
$ npx -y skills add crenshawdev/cadence --skill cad-plan-review --agent claude-code

How it fires

How this skill gets triggered: by you, by Claude, or both.

  • Fires itselfAuto-invocation. Claude auto-loads it when your prompt matches the work.Auto-invocation is when the right skill fires by itself at the right moment, driven by a FLOW.md router and a hook, instead of you invoking it by name. It is the difference between a skill being installed and a skill actually getting used.Read the full definition →
  • You can call itInvoke it directly when you want it.
  • Slash command/cad-plan-review

Context preview

The summary Claude sees to decide when to auto-load this skill.

On-demand adversarial plan review of a phase PLAN.md before code - for a hand-written, imported or just-edited plan (/cad-plan fires this itself)

SKILL.md

cad-plan-review.SKILL.md
name: cad-plan-review
description: "On-demand adversarial plan review of a phase PLAN.md before code - for a hand-written, imported or just-edited plan (/cad-plan fires this itself)"
argument-hint: "[phase number | path/to/PLAN.md]"
allowed-tools:
  - Read
  - Bash
  - Glob
  - Grep
  - Task
  - AskUserQuestion

<objective> Manually run the review subsystem's `plan` trigger against a chosen plan. The review capability lives entirely in the subsystem (references/review-triggers.md); this skill is just the on-demand entry point for cases the automatic review does not cover - a plan written by hand, imported from elsewhere, or edited after its first review. When /cad-plan writes a plan it already fires this trigger, so you do not need this skill in the normal flow.

There is no separate reviewer here and no convergence loop (cut in DESIGN §6): this delegates to `fire(plan)`, which grounds and adjudicates once. </objective>

<process> 1. **Resolve the target plan** from `$ARGUMENTS`:

  • a number `N` -> `.planning/phases/<N>/PLAN*.md` (all slices of that phase).
  • a path -> that file.
  • empty -> the current phase from the STATE.md cursor; if that is ambiguous,

ask (ask-user seam) which phase. If no plan file is found, report it and stop - there is nothing to review. (Resolve the plan path and, for the empty-args case, the cursor read as one batched step - independent; conventions.md Parallel work.)

2. **Fire the `plan` trigger** with the resolved PLAN file(s) as the artifact - Read `${CLAUDE_PLUGIN_ROOT}/cadence-core/references/review-triggers.md` here, not preloaded (one site). Honor `review.triggers.plan.gate` (default adjudicated) always; its `tier` and `effort` reach cross-model reviewers only. This resolves the reviewer set (claude-subagent and/or a configured cross-model reviewer), runs them, and - for an adjudicated gate - grounds each finding against the real repo, kills false positives, and merges convergent findings.

3. **Report** the outcome: for advisory, the findings; for blocking, the PASS/FAIL and surviving blocker/high findings; for adjudicated, the grounded survivor list. Do NOT auto-apply changes to the plan - present the survivors and let the user decide what to fix (the plan-creation flow triages the same way). </process>

Read more
Ships withcadence

Appearance is cheap. Verification is the work. Cadence is for developers using Claude Code on software they will still own after the session ends. Claude can write a convincing plan, produce working code, and tell you the job is finished.

Get the whole plugin

Other skills on cadence.