Projects

Short by design. The public bets, the serious shipped work, and the proof objects that show what actually shipped — not just a graveyard of side-project names.

last updated: 2026-06-01

// Featured proof

Status, stack, proof, current focus, and a real artefact where there is one.

  • CLI suite

    live

    A growing suite of agent-first CLI tools for useful public data and real-world workflows — built for humans in the terminal and predictable for agents when piped.

    $ tfl status
    Bakerloo: Good Service
    Central: Good Service
    District: Minor Delays
     
    $ parliament search bills "energy" --take 3
    Great British Energy Bill               5172   Committee
    Energy Prices Act 2022 (amendment)      4891   2nd reading
    
    stack
    Node · TypeScript · Commander · public APIs · npm · Vercel
    proof
    Five published CLIs with matching landing pages, npm packages, GitHub repos, and reusable agent skills.
    current
    Tighten shared output conventions, JSON schemas, examples, and companion skill quality across the suite.

    London transport: status, journeys, arrivals, disruptions, bike docks.

    UK company search, filings, officers, ownership, insolvency.

    UK rail departures, arrivals, station search, and fast travel queries.

    UK fuel prices, ranked by price, distance, or freshness.

    UK Parliament bills, members, votes, and written questions from official APIs.

    Proof notes

    • Five published CLIs with matching landing pages, npm packages, GitHub repos, and reusable agent skills.
    • Consistent pattern across transport, companies, rail, fuel, and parliament data: human-readable defaults, JSON/projection paths for agents.
    • Parliament domain now uses the properly spelled parliament-cli.xyz; the old misspelling redirects there.
  • Hermes Console

    live

    Local-first web dashboard for Hermes Agent: runtime health, sessions, cron, skills, memory, files, and usage in one place.

    Hermes Console overview screenshot
    stack
    React · TypeScript · local files · Hermes Agent · read-only UX
    proof
    Shipped as OSS and dogfooded against the real Hermes setup rather than fake dashboard data.
    current
    Keep dogfooding the operator surfaces and tighten what is genuinely useful versus just interesting once.

    Proof notes

    • Shipped as OSS and dogfooded against the real Hermes setup rather than fake dashboard data.
    • Built around inspection, not mutation: useful when running agents locally without turning the browser into a dangerous control plane.
    • Supported by public writing that explains the problem, constraints, and design trade-offs.
  • RoastMaster

    live

    Consumer AI app with real product constraints: image uploads, auth, credits, share loops, and a lot of prompt + output iteration.

    RoastMaster landing page screenshot
    stack
    Next.js · AI image/text workflows · auth · credits · payments · sharing
    proof
    Complete product loop rather than a toy prompt: upload, generate, pay/credit, share, and iterate on output quality.
    current
    Keep tuning output quality and run focused distribution pushes to find repeat usage loops.

    Proof notes

    • Complete product loop rather than a toy prompt: upload, generate, pay/credit, share, and iterate on output quality.
    • Prompting and model-selection lessons turned into public writing instead of staying as vibes in a repo.
    • Useful contrast to the serious CLI/operator work: still ships, still has product plumbing, just more stupid-fun.
  • OpenClaw Blackbox

    live

    Local-first run inspector for OpenClaw agents: failed, stalled, expensive, or weird runs turned into deterministic terminal, Markdown, JSON, and HTML reports.

    $ openclaw-blackbox inspect ./runs/latest
    status: failed
    cost: £4.21 · turns: 93 · transcript: found
    reports: terminal · markdown · json · html
    next: inspect missing tool output before rerun
    
    stack
    Node · TypeScript · Markdown · JSON · HTML reports · local-first CLI
    proof
    Built from real OpenClaw debugging pain: cron failures, context overflow, missing transcripts, weak evidence, and delivery weirdness.
    current
    Dogfood against real OpenClaw failures, tighten report quality, and keep the public v0 small enough to trust.

    Proof notes

    • Built from real OpenClaw debugging pain: cron failures, context overflow, missing transcripts, weak evidence, and delivery weirdness.
    • Disk-first by default, with no hidden LLM calls, so the report is predictable enough for agents and readable enough for humans.
    • Shipped publicly with npm package, skill release, docs, and site updates.

// Shipped at scale

  • Agglayer UI

    live

    Bridge anything to Agglayer via a production-ready web experience.

    stack
    production web3 app / SDK surface
    current
    Keep tightening UX and reliability across supported routes.
  • Agglayer SDK

    live

    Unified client interface for native and aggregator bridging.

    stack
    production web3 app / SDK surface
    current
    Keep hardening ergonomics and integration reliability.
  • sPOL liquid staking

    Polygon's native liquid staking product for staking, unstaking, migration, and DeFi-ready sPOL positions.

  • Cielo

    Instrumental in evolving Cielo from a Discord bot into a fully-fledged wallet discovery platform.

  • Aragon UI Kit

    Major contributor to the open-source governance UI kit used to build onchain governance experiences.