Skip to content
View edycutjong's full-sized avatar
9 followers · 1 following

Achievements

Achievement: Pull Sharkx3Achievement: Pair Extraordinairex3Achievement: Arctic Code Vault Contributor

Achievements

Achievement: Pull Sharkx3Achievement: Pair Extraordinairex3Achievement: Arctic Code Vault Contributor

Block or report edycutjong

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.

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
edycutjong/README.md

Edy Cu

Agent infrastructure & MCP. I ran AI coding agents as the primary builder across 72 hackathons and 139 projects — zero confirmed wins — then published the failure taxonomy my agents now read before they build.

→ The essay: 139 projects, zero confirmed wins

Based in Jakarta. Async and remote.

Pinned work

  • agent-postmortem-loop — 34 postmortems from 139 projects, distilled into a failure taxonomy and rules my agents load before building. The dataset behind the essay.
  • BagOS — MCP server for Solana (Bags.fm): simulate → approve → sign → send → confirm on every write, spending caps, devnet by default. v2.0.0 on npm with provenance. v1 shipped a fake-success bug — the fix and postmortem are public.
  • armsmith — ARM/Graviton performance benchmark toolkit. Signed releases on PyPI via trusted publishing.
  • aegis — Multi-agent support engine that stops and waits for human approval before any destructive action.
  • revoker — Approval hygiene for agent, keeper and relayer wallets: threat rules fire and the revoke lands before the drainer does.
  • antigen — Prompt-injection immune system for DataHub metadata: sweeps hidden payloads, defuses them in-graph, proves the cure.

The through-line: agents that act on real systems need brakes — approvals, caps, simulation, revocation, postmortems.

Stack

TypeScript · Python · Go · Rust — MCP servers · Solana (Web3.js / Anchor) · FastAPI · Next.js · PostgreSQL · Docker · GitHub Actions

Hackathons

93 events tracked, 72 entered, 139 projects shipped — DoraHacks · Devpost. Zero confirmed wins; the postmortems were the payout.

Contact

edy.cu@live.com · X @edycutjong · LinkedIn · edycu.dev

Pinned Loading

  1. agent-postmortem-loop agent-postmortem-loop Public

    139 projects. 92 hackathons. 1 award. The failure corpus that came out of running AI agents as the primary builder — and the loop that feeds it back into the agents.

    JavaScript

  2. BagOS BagOS Public

    MCP server for Bags on Solana. Every AI-initiated spend is token-gated, SOL-capped, and requires a single-use confirmation before anything is signed.

    TypeScript 1

  3. aegis aegis Public

    Autonomous Enterprise Action Engine — Multi-agent AI with Human-in-the-Loop approval, dynamic model routing, semantic caching, and real-time observability. FastAPI + LangGraph + Next.js + Supabase.

    Python

  4. revoker revoker Public

    Φ Autonomous on-chain security agent — revokes dangerous token approvals through KeeperHub in 13.47s median, so there is never a standing allowance worth attacking

    TypeScript 1

  5. antigen antigen Public

    🧬 A prompt-injection immune system for the DataHub metadata graph — detects, quarantines, and cures OWASP LLM01 payloads hidden in catalog metadata.

    Python

  6. armsmith armsmith Public

    🛠️ The agent that forges your repo for Arm — profiles any AI repo on aarch64, diagnoses why it's slow, and opens a fix PR where every fix must clear a reproduce gate or get dropped.

    Python