Skip to content

docs(README): first-run expectations, trim a repeat, qualify priority claim #6

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
windaddict merged 1 commit into from
Aug 9, 2026
Merged

docs(README): first-run expectations, trim a repeat, qualify priority claim #6

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
23 changes: 19 additions & 4 deletions README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -38,10 +38,9 @@ flowchart TB
```

The reviewer (indigo) proposes; the host (blue) verifies and applies the fixes they agree on; the
judgment calls come to you. The independent reviewer never edits — the critic and the editor stay
separate. A refutation only *drops* a finding when the host has contradicting evidence — a host
disagreement with no evidence isn't a refutation, so it goes back to the reviewer for one round,
then to you if neither side can win.
judgment calls come to you. A refutation only *drops* a finding when the host has contradicting
evidence — a host disagreement with no evidence isn't a refutation, so it goes back to the reviewer
for one round, then to you if neither side can win.

**Status: pre-release.** The open implementation of the pattern — named in
[the CLAR essay](https://www.movingavg.com/essays/cross-lab-adversarial-review.html) and told as a
Expand Down Expand Up @@ -289,6 +288,22 @@ Impasse performs on *your* work.
- Python 3 (standard library only — the shipped helpers have no pip dependencies).
- macOS or Linux. Windows via WSL; native Windows is on the [roadmap](docs/windows.md).

### First run — what to expect

- **You need an account with the reviewer's provider.** The cross-provider backend logs in with its
own credentials — a ChatGPT account or an OpenAI API key for Codex, a Claude account or an Anthropic
key for the Claude CLI (exact plan/entitlement is the provider's to set). That is separate from your
host, and reviews spend that provider's tokens.
- **It costs real tokens.** A small artifact at default effort is typically cents; large inputs or
high `--effort` cost more and take longer (a review can run from well under a minute to several,
depending on effort and size). It is not free to run.
- **The first send pauses for consent.** Reviewing sends your artifact to a third-party provider, so
the run **blocks the first time** until you approve the destination (see Data boundary & consent).
Nothing leaves your machine before you approve it.
- **What it does to your files.** The reviewer is **read-only** — it never touches your artifact. The
*host* applies the fixes it verifies to your working copy for you to review (like any edit your
agent makes); the review step itself changes nothing on its own.

### How independent is it?

Independence is a ladder, not a switch — and Impasse always tells you which rung you're on. A
Expand Down
Loading