groupwork
A second agent on the work, and a result you can cite
A second opinion that has already been told your conclusion is not a second opinion, it is agreement with extra steps. groupwork runs five named patterns - attack an idea, judge material independently, rule on finished work, think out loud with a peer, or run a blind debate - and what each pattern is refused is enforced rather than promised.
Install groupwork
Add the DBHQ marketplace, once
/plugin marketplace add dbhq-uk/marketplaceInstall the skill
/plugin install groupwork@dbhq
Both commands work in Claude Code and in Codex. The marketplace is a public repository you can read first: dbhq-uk/marketplace (external site)
What it is good at
Withholding your view in code, not in prose
The adversarial pattern runs in an empty directory with no repository access at all, and three of the five refuse to carry your conclusion. A flag fails the run outright if your own wording reached the brief, so the independence claim is checked rather than asserted.
Writing the record at the moment of the run
Every run appends what the counterpart was told and what it was withheld, copied from the pattern definition rather than typed afterwards from memory. Nobody writes the independence claim, so nobody can write one that is not true.
Running a debate blind before it runs adversarial
Proposals are made without sight of each other, then argued in rounds with a fresh session each round, so neither side is anchored on the other's framing and no round inherits the last one's fatigue.
Treating the counterpart as a provider, not the product
Three command-line agents sit behind one three-function contract, with nothing vendor-specific outside a providers directory. Swapping which agent gives the second opinion does not change the patterns.
Saying which counterparts are actually usable
It reports which are installed and which are authenticated, which are not the same thing, and one adapter is marked experimental in the output and in every citation produced through it because it has never been run against the real tool.
What it does not do
It does not delegate work
There is no go-and-build-this mode. Another tool already does that well, and it is the mode where a sandbox mistake costs real money.
It does not write findings into your repository
It records the run and hands back a citation. Where a finding belongs, and whether it belongs anywhere, stays your call.
It does not score or benchmark the counterpart
It reports what the other agent said. It does not rank models, grade answers or claim one counterpart is better than another.
One adapter has never been run against the real tool
The Copilot adapter was written from documentation and is marked experimental at version 0.1.0 - in the provider listing, in the README and in any citation produced through it.
Read it before you install it
Every line is public and MIT licensed. The README is the long version of this page, including the parts that are still rough.