Agent tools
Make agent work visible, local, and recoverable.
The invariant: long-running agent work needs observable state, narrow tools, and a way back to the evidence. Most of my tooling sits around that loop.
Raycast PiCommand-palette access to Pi prompts, sessions, project launch, selected-text transforms, and local resource browsing.
Session ReviewSQLite/FTS backed review of Pi, Claude Code, Codex, and Cursor session logs with grouping and transcript views.
Pi OpenTelemetryTrace context, metrics, diagnostics commands, and privacy-first payload capture for coding-agent runs.
Sticky Prompt HeaderA small Pi extension that pins the active prompt in the terminal so the task stays visible.
SVGL SkillAn intentionally tiny exact-match skill: find the brand, choose one route, write one SVG.
Private workbench
Some tools stay private because they depend on local auth, browser profiles, home machines, or personal datasets: browser workers, endpoint discovery, Twitter/X digests, Slack huddle sync, job-search dashboards, and file/resource browsers.
I keep those private until they are safe enough to explain without leaking credentials, home topology, or account state.