Portfolio / Open source / Epiphany Dynamics
I'm Patrick Gibbs, founder of Epiphany Dynamics. I build software for businesses, from the interface people use to the integrations that keep the work moving. I also repair systems that break when real customers use them.
Most of my public code lives in @epiphany-dynamics. Client implementations stay private; the projects here are public tools, products, and reference implementations.
|
A native Mac app for reviewing stale listening ports. Local evidence and typed AI recommendations inform the decision; you confirm every stop.
|
Connect Hermes to the official Cursor Agent SDK through a provider plugin. Hermes remains responsible for tool approvals and execution.
Subscription-oriented, not a promise of included inference. Cursor controls entitlement and billing. Offline tests don't establish live permission equivalence. |
|
AI lessons with quizzes, progress tracking, and rewards. Browser-local progress works without an account; optional sign-in supports cross-device sync.
|
Reference patterns for row-locked balances, access-control checks, and concurrent writes. Database-backed verification needs a configured test environment.
|
| I want to... | Start here |
|---|---|
| Connect Hermes to Cursor | Cursor SDK provider |
| Inspect ports on my Mac | Port Cleanup |
| See client work | Patrick's portfolio |
| Learn the AI basics | Epiphany Learn |
Open the systems lab: where does the human stay in control?
These diagrams describe the intended boundaries, not a certification of safety.
flowchart TD
subgraph PORT[Port Cleanup]
A[Local process evidence] --> B[Typed AI recommendation]
B --> C[Human confirmation]
C --> D[Recheck before graceful stop]
end
subgraph CURSOR[Cursor SDK for Hermes]
E[Hermes request] --> F[Cursor SDK adapter]
F --> G[Tool request returned to Hermes]
G --> H[Hermes approval and execution]
end
Port Cleanup never stops a process automatically. The Cursor adapter is experimental; its offline tests don't prove live billing or permission equivalence. Read each repository's limits before use.
More open-source work
| Project | Focus |
|---|---|
| Prime Desktop | Early public macOS client for Prime Agent with split-pane chats and live attachment. Known limits. |
| Resilient Data Source Kit | Backoff, circuit breaking, and confidence-scored entity matching. |
| Email Delivery Fallback Kit | Webhook validation, transactional email, and delivery-event tracking. |
| Call Speaker Diarization Kit | Known-speaker matching with human-review thresholds. Embedding and transcription calls are mocked. |
| Claude Skills Portfolio | Business-workflow skills and a configurable presentation generator. |
| Claude Discord Relay | Per-channel agent routing and persistent sessions. |
| Field Service Stack | An Astro publishing system for software reviews and guides. |
Upstream contributions and reports
Selected public submissions to Hermes Agent. Follow each link for its current status.
- WhatsApp capability documentation: a Baileys-versus-Cloud feature matrix.
- Technical-writing skill: a checklist and tests for repository documentation.
- SQLite connection-lifecycle bug report: a reproduction and proposed fix for accumulating read connections.
The failure path matters as much as the happy path. I look for bounded retries, explicit access rules, useful tests, and a way to verify the final result. A delivery event, stored record, or correct balance tells you more than a success message.
Working stack: TypeScript, JavaScript, Python, SQL, Swift · React, Next.js, Astro, Electron · PostgreSQL, Supabase, Firebase · APIs and webhooks.
Have a workflow or build in mind? See my work or book a conversation.