AI writes code faster than you can review it. See where you're needed and where AI can handle the rest.
go install github.com/zanetworker/highstakes/cmd/highstakes@latest
Or build from source:
git clone https://github.com/zanetworker/highstakes
cd highstakes
go install ./cmd/highstakes
The binary bundles parsers for every supported language (about 27 MB). Add -ldflags="-s -w" to either command to strip debug symbols and get about 19 MB.
Export your OpenRouter API key:
export OPENROUTER_API_KEY="sk-or-..."
Add it to your shell profile (~/.zshrc, ~/.bashrc) to persist across sessions.
cd /path/to/your/repo
highstakes init
highstakes analyze
highstakes init creates a .heatmap/ directory with a config.yaml file. highstakes analyze scans the repo, sends source files to the LLM for blast radius assessment, and writes results to .heatmap/heatmap.json.
On a 300-file repo this takes about 1-2 minutes and costs roughly $0.12 with the default model (GPT-6 Luna).
Three ways to see the heatmap:
# Interactive HTML dashboard (opens browser)
highstakes dashboard
# Terminal TUI
highstakes
# CLI queries
highstakes list --tier high
highstakes get src/auth/oidc.rs
highstakes report
Every file gets a posture from two questions, “Will it break?” and “How bad if it breaks?”, each shown with its repo cutoff. highstakes get <file> lists what to do before the next merge and what to put on the backlog; highstakes dashboard opens on the same actions as an inbox.
Harden now (tier critical): senior owner sign-off, regression and contract tests, auto-merge off on every PR that touches them. Gate them in CI with pr check --fail-on critical (see CI Integration).
Protect (tier high): owner review and no auto-merge; add tests first when none sit next to the file.
Stabilize (tier medium): regression tests when fixes recur, refactor what you touch. Never blocks.
Routine (tier low): normal review; agents may auto-merge here (autonomy L3) once the impact is AI-assessed. Don’t spend senior engineer time here.