01 / INVERSION TESTING
Does this step have to exist?
Work backward from the finished result. Test the dependencies that stand in its way.
Emergency recovery · AI Coding Factory
Code Red.
Burning resources.
Nothing shipping.
An emergency recovery playbook. It uses ablation testing to strip away the steps blocking delivery and get your factory moving again.
Four hours, one permission checkpoint. Code Red plans a four-hour window. At the one-hour mark it shows what changed, what it spent, and the strongest counterevidence — then asks your permission to continue the remaining three hours. Silence is not consent.
Draws on Munger’s inversion and Popper’s falsification. Also informed by research on subtractive thinking.
Change the routeKeep the result
— direct route━ alternate route
THE TOOLKIT
Three tests for the blocked step.
02 / PROCESS ABLATION
What can we remove?
Give a fresh agent the goal, not the favored plan. Remove one dependency or handoff. Compare the same inputs and test.
03 / NEGATIVE CONTROL
Can the test catch a bad answer?
Try a known-bad case. It must fail before you trust the new route.
GET CODE RED
Install the playbook.
Clone the public repository, then copy the skill into the personal directory for your coding tool. Review an existing installation before replacing it.
Browse Code Red on GitHub Download the source ZIP
git clone https://github.com/bradheitmann/code-red.git
cd code-redCODEX
mkdir -p "$HOME/.codex/skills"
cp -R skills/code-red "$HOME/.codex/skills/code-red"Start a new session or reload skill discovery, then invoke $code-red.
CLAUDE CODE
mkdir -p "$HOME/.claude/skills"
cp -R skills/code-red "$HOME/.claude/skills/code-red"Start a new session or reload skill discovery, then invoke /code-red.
Open two agent sessions: one to lead execution, one to challenge the plan. Use the role prompts and setup guide to connect them.