Claude Agent

architecture-drift-reviewer

Imported from arbiterforge/codearbiter/site/src/curated/agents/architecture-drift-reviewer.md.

LLM Mart · 0 points · 11 0 0

What vetted this — trust report

Download arbiterforge-codearbiter-site_src_curated_agents_architecture-drift-reviewer.md-44989c5.zip · 0 KB

arbiterforge/codearbiter 145 7 forks AGPL-3.0 Updated 9d ago

Part of arbiterforge/codearbiter — 238 skills

Install

skills CLI

npx skills add https://github.com/arbiterForge/codeArbiter/tree/main/site/src/curated/agents/architecture-drift-reviewer.md

Git git clone https://github.com/arbiterForge/codeArbiter.git

The skills CLI installs just this skill, for any of its supported agents. Git is the plain clone.

Files (codearbiter)

architecture-drift-reviewer.md 1.1 KB

---
entity: agents/architecture-drift-reviewer
related: [commands/checkpoint, commands/review, finding-triage]
---

## Role

Read-only checkpoint reviewer: for every ADR marked `accepted` in `.codearbiter/decisions/`, it scans
the codebase for evidence the decision is being followed or contradicted. It is dispatched as part of
the reviewer fleet by `/ca:checkpoint` and `/ca:review`, and its output feeds `finding-triage` on the
way to `checkpoint-aggregator`. It never edits code — this review is informational, pairing with the
`decision-variance` skill's append-only decision record rather than gating a change.

## Why this model tier

Ships `model: haiku`. Matching a known ADR against grep/glob evidence in the codebase is a high-volume,
pattern-matching check rather than open-ended judgment, which fits a fast, low-cost tier.

## What it emits

A CONFIRMED / PARTIAL DRIFT / DRIFT / INSUFFICIENT EVIDENCE classification per ADR, plus a
CRITICAL–LOW severity finding with file:line for every drift or partial-drift case. All output is
informational — it never blocks a change.