Consensus Loop Audit
Hands your repo to Codex for review, with sandbox bypassed by default
Test report
- Verdict
- Tested · Didn't pass
- Tested
- Aug 7, 2026
- Environment
- Claude Code 2.x (agent harness)
Security gate hit, confirmed by running it rather than reading it: with a stub binary on CODEX_BIN, a plain no-flag invocation spawned `codex exec -C <repo> --dangerously-bypass-approvals-and-sandbox --model gpt-5.4`. The bypass is the default (audit.mjs sets sandbox to danger-full-access at line 88), and the string 'danger-full-access' appears nowhere in the SKILL.md flag table, the README, CLAUDE.md or docs/ — only in the script's own --help. Everything else works: the plugin installed cleanly in a throwaway HOME and --dry-run produced a real audit prompt, but running the audit as documented hands your working tree to a second agent with every guardrail off and no warning.
Scored on four weighted criteria — install, triggering, output vs. baseline, docs. How scoring works
- Installs cleanly 5/5
- Triggers reliably 5/5
- Output vs. baseline 0/10
- Docs & honesty 1/5
What Consensus Loop Audit does
Slash-command skill from the consensus-loop plugin that runs its cross-model audit on demand: Codex reviews the pending review items in the repo's watch file and writes verdicts back. Triggers when you ask to run, re-run or dry-run a consensus audit. Note that its default invocation launches Codex with approvals and sandboxing disabled, which is documented nowhere the user is likely to read.
How to install Consensus Loop Audit
git clone https://github.com/berrzebb/consensus-loop.git
mkdir -p ~/.claude/skills/consensus-loop-audit
cp -r consensus-loop/. ~/.claude/skills/consensus-loop-audit/
cp consensus-loop/skills/audit/SKILL.md ~/.claude/skills/consensus-loop-audit/SKILL.md
Skills live in ~/.claude/skills/ (global) or .claude/skills/
(per-project). Restart Claude Code after installing.
Commands — how to trigger Consensus Loop Audit
-
/consensus-loop-auditHands your repo to Codex for review, with sandbox bypassed by default
It also activates on plain-language prompts like these:
-
Run the consensus audit on this repo's pending review items -
Dry-run the cross-model Codex audit on our watch file first -
Re-run the consensus loop review and have Codex write verdicts
Frequently asked questions
- Is the Consensus Loop Audit skill free?
- Yes. The skill itself is free from berrzebb/consensus-loop. SkillProof publishes the install command and an independent test verdict at no cost.
- Does Consensus Loop Audit work with Claude Code?
- We tested it with Claude Code 2.x (agent harness) on Aug 7, 2026. Verdict: Tested · Didn't pass. Security gate hit, confirmed by running it rather than reading it: with a stub binary on CODEX_BIN, a plain no-flag invocation spawned `codex exec -C <repo> --dangerously-bypass-approvals-and-sandbox --model gpt-5.4`. The bypass is the default (audit.mjs sets sandbox to danger-full-access at line 88), and the string 'danger-full-access' appears nowhere in the SKILL.md flag table, the README, CLAUDE.md or docs/ — only in the script's own --help. Everything else works: the plugin installed cleanly in a throwaway HOME and --dry-run produced a real audit prompt, but running the audit as documented hands your working tree to a second agent with every guardrail off and no warning.
- How do I install Consensus Loop Audit?
- Copy the install command from this page, run it in your terminal, and restart Claude Code. Skills live in ~/.claude/skills/ (global) or .claude/skills/ inside a project.
- Can I use Consensus Loop Audit with Cursor, Copilot, Gemini CLI, Codex or other AI tools?
- The SKILL.md format is native to Claude (Claude Code, Desktop, claude.ai). The instructions inside adapt to other assistants: Cursor rules, GitHub Copilot instructions, Windsurf rules, Custom GPTs, AGENTS.md for OpenAI Codex, and GEMINI.md for Google Gemini CLI — our conversion guides cover each, and the free converter on the tools page does the wrapping for you.