roozbeh-bahrami

Claude Code Advisor — AI skill for Claude Code

AI community

Read-only verification agent for Claude Code — reviews your AI sessions' work, separates PROVEN from CLAIMED, catches the traps before you greenlight.

How to install Claude Code Advisor

This entry records only its repository, not the path inside it, so there is no exact command to give. Open roozbeh-bahrami/claude-code-advisor and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Claude Code Advisor does

Read-only verification agent for Claude Code — reviews your AI sessions' work, separates PROVEN from CLAIMED, catches the traps before you greenlight.

Alternatives in AI

Agor — Orchestrate Claude Code, Codex, and Gemini sessions on a multiplayer canvas 1.2k ★
SoloDawn — Catches and repairs AI hallucinations before delivery: three quality gates, 31 built-in rules, a self-healing 312 ★
Ring — 89 skills and 38 specialized agents that enforce proven engineering practices for AI-assisted development 176 ★

README

claude-code-advisor 🔍

**A read-only verification agent for Claude Code — the AI that audits your other AIs.**

Turn a second Claude session into your **brutally honest, read-only advisor**: it reviews what your working sessions build, verifies claims with its own eyes, separates **PROVEN from CLAIMED**, and tells you plainly whether to greenlight — with the exact text to paste next.

Built from months of running real client automations with AI executor sessions. The pattern it packages:

 EXECUTOR session  →  builds, fixes, ships
 ADVISOR session   →  verifies independently (read-only), catches traps, gives the verdict
 YOU               →  make the call

Why a separate advisor?

A session that builds something can't audit itself — it shares its own blind spots. The advisor:

Never writes, never runs, never fixes — read-only by hard rule, so its verdict stays independent
Verifies by a different path than the executor used (agreement ≠ verification)
Separates built / verified / still-draft / live and working — so "done" is never fuzzy
Catches the classic traps: re-litigation, churn, scope drift, over-engineering, "impossible" that nobody actually checked
Talks like a sharp friend, not a corporate report: 4-line status, verdict color 🟢🟡🔴, then the next move

Install (Claude Code)

Copy the `advisor/` folder into your skills directory:

cp -r advisor ~/.claude/skills/advisor

New session → type `/advisor` → paste your working session's output → get the verdict.

Works in any Claude surface: the body of `advisor/SKILL.md` also works pasted as a standalone system/mode prompt.

The loop

Your working session reports "done" or proposes a plan
Paste that output to the advisor
Advisor opens the real state read-only, verifies, answers: status color · what's PROVEN vs CLAIMED · the exact text to paste back
You decide. Repeat.