Projects

The hard part was the platform: no staging, 350+ form fields generated at runtime, label names that change whenever the vendor pushes an update. I built a Playwright harness where field mappings live in config files. When labels shift, I update config instead of rewriting tests. When that's still not enough, a Claude API layer does fuzzy selector matching.

The 14 scripts came separately. Each one started with a manual process I got tired of doing: video capture for HIPAA audit evidence, Jira release summaries from ticket metadata, audio transcription, interactive HTML checklists, daily QA digests, 63-section form autofill. Two and a half years at Chorus, accumulated one annoyance at a time.

The public repo has 5 of the 14 scripts. The rest touch PHI-adjacent workflows or internal Jira structure that can't be extracted cleanly.

Holiday break, backlog of piled-up tickets, most of them queued for release within a day or two of getting back. JIRA's default view wasn't cutting it. The ticket names didn't have the context I needed to sort them at a glance, and I couldn't reorder things without changing the actual board, which would mess with everyone else's setup. Built a local wrapper instead. Changes stay local.

I kept re-explaining the same context to Claude. Not what I decided, but why, and what was happening at the time. New session or different machine, same result: gone. Chronicle is an MCP server with a SQLite backend that stores the timeline and surfaces it on request, mid-conversation, as a live tool. Not a static file you paste in once. So the context follows me.

Three files, no build step, no npm. Tasks have three time states (due-by, open, recurring), subtasks, dependencies, and an activation date that controls when something surfaces. The ADHD angle is deliberate. Blocked status comes from unfinished dependencies, not something you toggle manually. The app figures out what you can do right now. Optional Supabase sync.