Root Cause for Claude

You fixed the bug. You did not fix what let it happen.A test fails. You patch what you saw. The ticket closes. Six months later a different part of the code hits the same root cause and you start from zero.I spent 12 years in threat intelligence. Same pattern at every company. Teams found the failure, fixed the symptom, wrote nothing down. The learning never compounded.A flaky fix is the same problem. The surface cause is what fired. The structural cause is why it was allowed and went uncaught. Most "root cause" writeups name one neat cause and stop. That is the trap.What Root Cause for Claude doesIt turns a failure into a real RCA, inside the same Claude Code session where the work happens. It separates the surface cause from the structural cause. It forces multi-factor analysis and tags each cause by type. It requires verification you actually ran, a code fence or "ran X, got Y". It tracks action items as checkboxes with owners. It stays blameless by contract.The gates are Python, not prompts. rca-lint.py reads every RCA on save and blocks it if a required section, a cause-type tag, evidence, or a checkbox is missing. The discipline holds even when the model would rather hand you a tidy single cause.What is in the kitThe rca skill: the canonical template plus FAANG-lineage methodology. /rca-start scaffolds an RCA, pre-filled with what the session already knows. /rca-check lints any RCA and names exactly what is missing. The rca-lint hook validates every RCA on save. The rca-notify hook taps you when a command or test fails. A premortem variant: assume it already failed in front of the customer, and enumerate how.Self-contained. Python standard library only. No dependencies, no API keys, runs offline. MIT for the shipped plugin.Install: no terminal, no gitYou need a paid Claude plan (Pro or higher). Buy the kit, open the Claude desktop app, pick a folder in the Code tab, and paste the install prompt that ships in the zip. Claude Code wires the commands, skill, and hooks itself. Then type /rca-start the next time something breaks.I build RCA and review discipline for teams running agent coders in production. Want it wired into your repo, or a heavier spec-and-review pipeline? assaf@askconsulting.io

Found on Devadex — the discovery index for independent software the big search engines bury. More from gumroad.

Report this listing