Hacker

Naimish Singh

@naimish

Joined Oct 2026Student at AKTUIndia

Projects

  • Safe Tool Execution Visibility & Defense-in-Depth Secret Redaction (PR #1684)

    Flip the Switch HackathonNov 2026

    In collaborative multi-agent environments, engineering teams reviewing an agent's work need to know what commands and tools the agent actually ran (e.g. pytest, git diff, database scripts) without switching context away from team channels. However, raw tool execution outputs frequently contain critical operational secrets (API tokens, database credentials, PEM keys) or flood channels with noisy output. This contribution (PR #1684, resolving Issue #803) brings safe, bounded tool detail visibility to Switch. It implements a robust Server-Side Defense-in-Depth architecture in Switch Core that deterministically strips sensitive secrets (OpenAI, Anthropic, AWS, Google, GitHub, Slack tokens, DB URIs, and private keys) regardless of whether an agent originates from a desktop app, CLI, or headless runner. It bounds payloads to 500 characters for input and 2,000 characters total, supports Slack, Discord, and Mattermost rendering, and introduces an admin toggle switch in the Switch Gateway dashboard with zero breaking database migrations.

    BitBuddies@mayank_cs50

  • GitOS: Graph-Powered Ecosystem Defense & Intelligence Layer for GitHub

    Graph HacksOct 2026

    Open-source maintainers are drowning in "AI Slop"—untested, hallucinated pull requests generated by LLMs that introduce subtle multi-hop breakages, dead code, and cosmetic churn. Meanwhile, contributors lack visibility into downstream blast radius when modifying shared functions. GitOS solves this by deploying a graph-powered defense engine backed by FalkorDB and AST parsing. Before spending any LLM tokens, GitOS deterministically maps incoming PRs against the full repository topology in FalkorDB, calculating multi-hop blast radius, detecting dead code (in-degree = 0), identifying untested callers, and enforcing maintainer architectural boundaries. A 5-agent swarm then synthesizes targeted triage decks, AST semantic release notes, and single-shot micro-patches—all executed through authenticated GitHub OAuth with zero unconsented spam.

    BitBuddies@mayank_cs50,@harshit01,@jagrati_gupta