Projects
Each case study follows the same structure: the problem, the existing state, the constraints, the options considered, the decision, the safeguards, the outcome, and what I’d change next time. Employer work is generalized; code shown is public or recreated from scratch.
Case studies
-
gh-baseline: the required checks that accepted any source — A gh extension that reports where repositories drift from a declared security baseline. I built it in Go, a language I don’t work in day to day, with an LLM writing most of the code, and the judgment mattered more than the language.
-
Keel: the retry that could send one alert four times — Why “retry on server errors” isn’t safe for a request that can’t be repeated, and the redesign that makes a PowerShell mail library resend only when Microsoft Graph provably refused the message.
-
Email alignment checker — A scheduled GitHub Actions workflow that monitors this domain’s mail authentication records — and the three iterations it took to make it quiet, honest, and self-sustaining.
-
Making this site faster on mobile — A PageSpeed mobile score of 86, three things the page was loading that it didn’t need, and the measurements that showed removing them changed nothing visible.
Open source
- gh-baseline — read-only
ghextension in Go that audits GitHub repositories against a tiered security baseline: rulesets and where required checks come from, secret scanning, and workflow hygiene. Anything it can’t read is reported as unknown, never pass. - Keel — PowerShell modules for unattended automation, CI-tested on Windows PowerShell 5.1 and PowerShell 7: bounded HTTP retries and Graph-first mail that won’t resend after an ambiguous failure.
- Email alignment checker — scheduled GitHub Actions workflow that monitors SPF, DKIM, and DMARC for this domain and commits only when something actually changes.
- LinuxConfig — modular Bash pipeline that bootstraps RHEL hosts into Active Directory and Red Hat Satellite, with dry-run mode on every stage and bats + ShellCheck in CI.
- adaptive-comm-tool — Python CLI that uses Claude to score and rewrite a message for each teammate persona, with output validated against a schema so no persona is invented or dropped.