The Anthropic Hackathon Winner Who Open-Sourced His Entire AI Coding Stack

The winner of an Anthropic hackathon released his complete AI coding harness for free under the MIT license. The video pitches it as 10 months of work packaged into 183 agent skills, 48 sub-agents, and 79 ready-made commands, all published in one public GitHub repo. The repo is affaan-m/ECC, and it has kept growing since the video went up.

What Is Actually Being Released

ECC bills itself as an "agent harness performance optimization system." The idea is that instead of re-explaining your planning, testing, and review process in every prompt, you install the harness once and your coding agent inherits a standard set of workflows. The README frames it as an operating system for AI agents rather than a single tool or plugin.

The pieces named in the video map to three layers:

Component	Count in video	What it does
Agent skills	183	Reusable capability modules (TDD, security, docs, frontend/backend patterns)
Sub-agents	48	Delegated specialists for planning, code review, security analysis, build repair
Commands	79	Named entry points that trigger a workflow

The counts are a snapshot. By the time of this writeup the public repo lists higher numbers (roughly 281 skills, 67 sub-agents, 94 commands), so the video captured an earlier state of an actively maintained project.

Why the MIT License Matters Here

The hook is not just that the code is free. It is that the license is permissive. MIT lets you copy, modify, and ship the stack inside commercial work with almost no restriction beyond keeping the copyright notice. For a setup someone spent 10 months building and then won API credits with, releasing under MIT means teams can lift the whole thing into their own tooling without a legal review cycle. That is the difference between a repo you admire and a repo you actually adopt.

The Prize Context

The creator says he won $15K in API credits at the hackathon. That detail does more than flex. API credits are the fuel an agent harness burns, so the prize and the product line up. A system that fans work out to dozens of sub-agents makes a lot of model calls. Winning credits and then open-sourcing the harness is a coherent story: prove it at a hackathon, then hand the framework to everyone else.

What to Watch For Before You Adopt It

A large skill and agent library is a starting point, not a guarantee. The value depends on how well the harness matches your stack, how much token overhead it adds, and whether the workflows fit how your team already ships. The README itself leans on the phrase "optimize the context window, persist everything else," which is the right instinct: the win is in what the harness keeps out of the active prompt, not in raw component count. Treat the 183 or 281 number as inventory, then test whether the planning, review, and verification loops hold up on your own code.

Key Takeaways

• The repo is affaan-m/ECC, released MIT-licensed after an Anthropic hackathon win.
• The video cites 183 agent skills, 48 sub-agents, and 79 commands. The live repo now lists more, since it is actively maintained.
• ECC is a coding-agent harness: install once, and your agent inherits standard planning, testing, review, and verification workflows.
• It advertises cross-harness adapters for Claude Code, Codex, Cursor, and OpenCode, so it is not tied to a single tool.
• Judge it by how well its workflows fit your stack and context budget, not by the headline component count.

Resources

• affaan-m/ECC -- The open-sourced agent harness, MIT-licensed, with skills, sub-agents, and commands.