Developers are currently witnessing a fundamental shift in how software is constructed. The rise of autonomous coding agents has moved the workflow beyond simple autocomplete suggestions toward complex, multi-step trajectories. These trajectories—the sequence of tool calls, the analysis of test failures, and the subsequent iterative corrections—represent the actual reasoning process of an AI. However, this high-value data remains trapped in private session logs. For the Korean developer community, this creates a critical bottleneck in post-training data, as most high-quality reasoning traces available for model refinement are dominated by English-centric datasets.
The Architecture of Open Contribution
Show GN is addressing this data scarcity with the release of `/take-my-data`, a specialized tool designed to harvest session logs and transform them into a CC0 public dataset. By utilizing the CC0 license, the project ensures that the data is placed in the public domain, stripping away copyright restrictions to allow for unrestricted use in AI research and model development. The tool specifically targets trajectories from prominent coding agents, including Claude Code, Codex, Pi, and OpenCode. By collecting these sequences of tool calls and test results, Show GN aims to build a robust foundation for Korean-language post-training, allowing future models to better understand the nuances of coding workflows and technical problem-solving within a Korean linguistic context.
To maintain a consistent pipeline of information, the donated data is released on Hugging Face once a month. This cadence allows the community to integrate new trajectories into their training loops incrementally. To incentivize early adoption and data volume, Show GN has introduced a participation reward. Five individuals who contribute to the dataset construction by September 18, 2026, will be randomly selected to receive coffee gifticons.
Solving the Privacy Paradox
The primary barrier to sharing session logs is the inherent risk of leaking sensitive information. Coding logs are notorious for containing environment variables, proprietary logic, or accidentally hardcoded credentials. The tension lies between the desire for open-source progress and the necessity of corporate and personal security. Show GN resolves this by implementing a local-first sanitization pipeline that shifts the trust model from the server to the user's own machine.
Before any data is transmitted, the tool employs `gitleaks` to scan the logs for secrets and API keys. This is followed by a rigorous cleaning process using regular expressions and agent-specific masking techniques to erase personally identifiable information. The critical distinction here is that the user remains the final arbiter of what is shared; only content explicitly approved by the user is uploaded. Once the sanitized data reaches the server, it undergoes a secondary verification check to ensure no leaks occurred during the local process. Because the raw, original logs never leave the local environment, the tool eliminates the most significant risk associated with data donation.
This approach transforms the act of data sharing from a security liability into a controlled contribution. By automating the scrubbing process through `gitleaks` and masking, Show GN lowers the friction for developers who want to contribute to the ecosystem but cannot risk a security breach.
This initiative marks a significant step toward community-driven data sovereignty for non-English AI development.
