prd-taskmaster, my MIT-licensed Claude Code tool, turns a one-line goal into a graded PRD and an ordered task list. I also have open pull requests on the AI agent runtime I build on.
- Role
- Author and maintainer
- Licence
- MIT
- Status
- Pre-alpha, as the README says
- GitHub
- 607 stars, 59 forks (11 Oct 2026)
- npm
- prd-taskmaster 5.2.2
- Outside contributions
- 1 merged pull request from another developer
- Upstream
- 11 open pull requests, 7 to NousResearch/hermes-agent; none merged yet
- Obsidian
- Unread Changes plugin, listed in the community plugin directory
prd-taskmaster on GitHub · npm · Obsidian Unread Changes
A secrets manager I built so AI agents can use passwords and API keys without the plaintext reaching a prompt, log or commit. It ships as a CLI, an MCP server and a browser extension.
- Role
- Design, build, run
- Stage
- Early access; in daily use on my own systems
- Latest release
- Core v0.7.2 (11 Oct 2026), browser extension v0.4.0
- Tests
- 2,328 passing on Python 3.12 and 3.13 (CI, 11 Oct 2026)
- Cryptography
- Argon2id key derivation, XChaCha20-Poly1305 encryption
- Surfaces
- CLI, MCP server (65 tools), browser extension; iOS and Android in beta
- Source
- Private repository. I can walk you through it running on a call.
- What you see here
- Recordings of the real CLI against a throwaway demo vault
A governed context layer for a business run with AI agents: one markdown vault that every agent loads at boot, changed only by merged pull request, with receipts of what each agent read.
- Stage
- Early access, pre-v1; runs my own business every day
- Engine versions
- 9 tags, v0.1.0 to v0.9.0
- Boot set
- 48 notes, about 40,000 tokens, loaded by every agent (11 Oct 2026)
- Scheduled jobs
- Drift detector every 15 minutes, health check, nightly digest email
- Source
- Private repository. I can walk you through it running on a call.
The MCP servers my AI agent fleet runs on: isolated browser slots for parallel agents, a launcher that starts and watches Claude Code and Codex sessions, and a ledger-backed worker controller.
- Status
- In daily internal use since April 2026
- Launcher
- 86 MCP tools
- Browser MCP
- 20 lock-protected browser profile slots
- Worker controller and task board
- 12 and 23 MCP tools
- Source
- Private repository. I can walk you through it running on a call.
- What you see here
- One real run on 11 Oct 2026: three agents, three isolated browsers
Intro to AI: Build Your Personal AI Playbook. Six 60-minute workshops for people who are not developers, with slides, a 24-page workbook and a facilitator guide.
- Status
- Written October 2026; first workshops being booked
- Format
- Six 60-minute workshops, any chatbot, no coding
- Materials
- 106 slides in 6 decks, 24-page workbook, 44-page facilitator guide
- You leave with
- A personal AI playbook: context card, saved recipes, a check-it list
Course details
A website and iPhone app, built in memory of Vic, that turn picking up litter into photo-checked acts with points. An AI model checks each photo before it counts.
- Status
- Web app live; iPhone app in public TestFlight beta, not yet on the App Store
- Built
- September to October 2026
- Database
- 51 migrations; row-level security covered by 263 database tests (recorded in the build log)
- Stack
- Next.js on Vercel, Supabase, SwiftUI
Visit the site