{"slug": "agent-harness-hackathon-2026-what-i-built-how-i-used-trueforge-and-what-i", "title": "Agent Harness Hackathon 2026 — What I built, how I used TrueForge, and what I learned", "summary": "A developer built AutoVault, an autonomous AI security operations agent using TrueFoundry's open-source TrueForge harness, which detects ransomware in 0.3 seconds and spawns parallel subagents for investigation while requiring human approval for irreversible actions. The project, entered in the Agent Harness Hackathon 2026, leverages all 12 TrueForge features including MCP tools, subagents, sandboxing, and persistent sessions, and incorporates 13 innovations such as predictive defense and self-healing file systems.", "body_md": "THE PROBLEM\n\nRansomware attacks cost organizations over $20 billion every year. Security teams are drowning in 11,000+ alerts per day, most of which are false positives. By the time a human analyst investigates, the encryption has already spread to hundreds of files.\n\nThe average response time? Hours. The average recovery cost? $2.73 million.\n\nI wanted to build something that could change that — an AI agent that doesn't just detect threats, but investigates and responds to them autonomously, while still keeping a human in the loop for irreversible actions.\n\nThat's how AutoVault was born.\n\nWHAT I BUILT\n\nAutoVault is an autonomous AI security operations agent that runs on TrueForge, the open-source agent harness by TrueFoundry. It detects ransomware in 0.3 seconds, spawns 5 parallel subagents to investigate, and responds autonomously — but always stops to ask a human before doing anything irreversible.\n\nHOW I USED TRUEFORGE\n\nTrueForge isn't just a dependency — it's the brain of the entire system. I used every single TrueForge feature, all 12 of them.\n\nMCP TOOLS\n\nI built 4 Python MCP servers using the MCP SDK v2:\n\nEach tool has full input schemas, proper error handling, and returns structured JSON.\n\nSUBAGENTS\n\nTrueForge's subagent system was a game-changer. I spawned 5 specialized agents that run in parallel:\n\nThe speed improvement was dramatic — what took hours now takes seconds.\n\nSANDBOX (DAYTONA)\n\nI wrote 20 Python scripts that run safely in TrueForge's Daytona sandbox:\n\nHUMAN APPROVALS\n\nTrueForge's approval system lets me define 10 rules with risk scoring:\n\nThe 7-step workflow ensures nothing irreversible happens without a human saying yes.\n\nSKILLS\n\nI created 4 instruction packs that TrueForge loads dynamically:\n\nCONTEXT ENGINEERING\n\nTrueForge's context engineering kept the agent efficient:\n\nPERSISTENT SESSIONS\n\nThe agent saves its state to SQLite every 30 seconds. If the connection drops mid-investigation, it resumes exactly where it left off.\n\nTHE 13 INNOVATIONS\n\nHuman-Like Memory Architecture — Episodic, semantic, procedural, and working memory. The agent remembers past attacks.\n\nPredictive Ransomware Defense — Predicts attacks 30 minutes before they happen.\n\nRecursive Self-Improvement — The agent improves its own code and algorithms.\n\nCollective Intelligence Swarm — Hive mind of agents making decisions together.\n\nNeuromorphic Processing — Brain-inspired neural networks with sub-millisecond latency.\n\nSelf-Healing File System — Automatically restores encrypted files from blockchain backups.\n\nAI Red Team vs Blue Team — Two AI teams compete and improve through adversarial training.\n\nKnowledge Graph — Graph database mapping relationships between threats, actors, tools, techniques.\n\nFederated Learning — Multiple organizations train models without sharing sensitive data.\n\nDigital Twin Simulation — Virtual replica for risk-free security testing.\n\nAdversarial Robustness — Defends against attacks designed to trick the AI.\n\nNatural Language Threat Intelligence — Query threats using plain English.\n\nSelf-Learning Security Evolution — Detection models evolve using genetic algorithms.\n\nTHE QODO CODE REVIEW EXPERIENCE\n\nEvery change goes through Qodo review via GitHub Actions. On PR #1, Qodo found:\n\nAll three were real issues. I fixed all HIGH findings and documented everything in the README.\n\nWHAT I LEARNED\n\nTrueForge Is More Than a Wrapper — The MCP tool system, subagent orchestration, approval workflows, and context engineering are all deep, production-grade features.\n\nParallel Subagents Change Everything — Running 5 investigations simultaneously isn't just faster — it's qualitatively different.\n\nHuman Approval Is Non-Negotiable — For a security tool, having human approval is critical. TrueForge made this trivial to implement.\n\nQodo Actually Finds Bugs — It understands context and finds issues that surface-level analysis misses.\n\nThe Sandbox Is Your Safety Net — When running scripts that analyze potentially malicious files, you need a sandbox.\n\nDocumentation Matters — Writing the README forced me to think about what I actually built and why.\n\nTRY IT YOURSELF\n\ngithub.com/ItsParthPinjarkar/autovault\n\ngit clone [https://github.com/ItsParthPinjarkar/autovault.git](https://github.com/ItsParthPinjarkar/autovault.git)\n\ncd autovault\n\npip install -r agent/requirements.txt\n\npip install -r mcp-server/requirements.txt\n\nbash setup-trueforge.sh\n\npython demo_advanced.py\n\nACKNOWLEDGMENTS\n\nBuilt for the Agent Harness Hackathon (August 24-30, 2026).\n\nThis project was built with AI assistance. All code has been reviewed, tested, and understood by the developer.", "url": "https://wpnews.pro/news/agent-harness-hackathon-2026-what-i-built-how-i-used-trueforge-and-what-i", "canonical_source": "https://dev.to/parth_pinjarkar_86f7f3ec2/agent-harness-hackathon-2026-what-i-built-how-i-used-trueforge-and-what-i-learned-5fh6", "published_at": "2026-08-30 09:08:14+00:00", "updated_at": "2026-08-30 09:52:39.033125+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-agents", "ai-tools", "ai-infrastructure", "developer-tools"], "entities": ["AutoVault", "TrueForge", "TrueFoundry", "Qodo", "Daytona", "MCP", "GitHub Actions"], "alternates": {"html": "https://wpnews.pro/news/agent-harness-hackathon-2026-what-i-built-how-i-used-trueforge-and-what-i", "markdown": "https://wpnews.pro/news/agent-harness-hackathon-2026-what-i-built-how-i-used-trueforge-and-what-i.md", "text": "https://wpnews.pro/news/agent-harness-hackathon-2026-what-i-built-how-i-used-trueforge-and-what-i.txt", "jsonld": "https://wpnews.pro/news/agent-harness-hackathon-2026-what-i-built-how-i-used-trueforge-and-what-i.jsonld"}}