Skip to content
View yashkhou's full-sized avatar

Achievements

Achievement: Pull Sharkx2Achievement: QuickdrawAchievement: YOLO

Achievements

Achievement: Pull Sharkx2Achievement: QuickdrawAchievement: YOLO

Highlights

Block or report yashkhou

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
yashkhou/README.md
Josh Khoury's live lab card: public contributions over the last eight weeks, where the commits went, merged pull requests, npm downloads and the latest push. Refreshed daily from the GitHub and npm APIs.

I'm Josh, an applied AI builder in Dubai. My work sits where an AI agent meets real people: what it can safely handle by itself, what it should pass to a person, and proof afterwards of what it actually did.

How my tools fit together

Animated illustration of one agent run: Commander Plus runs the tool calls, ActionMesh blocks a write to ~/.ssh/config that is outside its contract, RunLedger chains every step with a hash, BrowserProof confirms the browser landed on the claimed page and Verify checks the saved PDF. When block 3 of the ledger is edited after the run, the chain turns red from that block on and the run is flagged for a person.

Each tool call is checked against its contract and written to a ledger that can't be quietly rewritten. The page and the file are checked instead of taking the agent's word for it. If anything doesn't add up, a person takes over.

What I'm building
Commander Plus My daily driver. It extends DesktopCommanderMCP so ChatGPT and other MCP clients can do real work on my Mac. 129,811 tool calls in its first three weeks.
RunLedger A tamper-evident record of what an agent actually did. Edit, delete or reorder one step and verification fails.
Verify Checks an agent's output without trusting the agent: documents, code and reversible actions. Three packages on npm.
BrowserProof Proves a browser agent really ended up where it says it did.
ActionMesh One typed contract for each thing an agent is allowed to call.
Agent Reliability Lab Twelve small tools that break agent infrastructure on purpose, before production does.
Glyph An experiment: a design language agents can reason about, instead of pixels.

How I work

I design the system and plan each feature. AI coding agents type most of the code. I review what they write, test it hard, and fix what they get wrong myself.

Lately, upstream

yashkhou.com · X · LinkedIn

Pinned Loading

  1. actionmesh actionmesh Public

    Typed TypeScript action contracts for AI tools, MCP-style calls, HTTP and CLI with shared Zod validation, errors and retry semantics.

    TypeScript

  2. agent-reliability-lab agent-reliability-lab Public

    Twelve deterministic tools for testing and hardening AI agent infrastructure: MCP chaos, contract fuzzing, context firewalls, replay, evals.

    Python

  3. browserproof browserproof Public

    Verify AI browser agents and Playwright workflows with explicit assertions, evidence hashes, HTML reports and CI-ready pass/fail results.

    TypeScript

  4. runledger runledger Public

    Tamper-evident local execution ledger for AI agents: hash-chained tool calls, decisions, retries, results and verifiable HTML reports.

    TypeScript

  5. verify verify Public

    Verification infrastructure for AI work: artifacts, agent-written code, and reversible actions.

    JavaScript

  6. commander-plus-mcp commander-plus-mcp Public

    Forked from wonderwhy-er/DesktopCommanderMCP

    Extends DesktopCommanderMCP into a remote, OAuth-secured MCP control plane for macOS. 129,811 tool calls since 15 Sep 2026.

    TypeScript