vakovalskii

vakovalskii / phantom-agent

Public

Autonomous AI agent for BitGN PAC1 Challenge — ~86% score with 12 hot-reloadable skills, live dashboard, and self-correcting classification

16
7
100% credibility
Found Apr 14, 2026 at 16 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

An open-source AI agent that autonomously solves benchmark tasks involving file operations, CRM, emails, and security checks in isolated virtual workspaces, complete with a real-time dashboard.

How It Works

1
🔍 Discover the agent

You find this project on GitHub, an smart helper that tackles real-world tasks in a safe testing playground.

2
🔑 Get your passes

Sign up for the challenge playground and link your thinking AI service so it can get to work.

3
💻 Launch the viewer

Start the colorful dashboard screen to watch everything happen live.

4
🚀 Kick off a test run

Hit the run button, pick how many helpers to use at once, and see it explore, think, and solve tasks right before your eyes.

5
📊 Review the results

Check live scores, tool uses, and times for each task, plus compare past runs with heatmaps.

🏆 Top the charts

Celebrate your agent's high score on the public leaderboard after submitting the best run.

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 16 to 16 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is phantom-agent?

Phantom Agent delivers autonomous agent capabilities for the BitGN PAC1 benchmark, solving 86% of 43 sandboxed file-system tasks like CRM lookups, invoice creation, secure inbox processing, and knowledge capture. Built in Python with OpenAI Agents SDK, it provides CLI benchmark runs, leaderboard submission, and a live React dashboard streaming tool calls, scores, and token usage. Users get hot-reloadable skill prompts—edit Markdown files to tweak behaviors without restarts—making it a practical autonomous agent framework for testing AI in isolated VMs.

Why is it gaining traction?

Its self-correcting classifier picks from 12 specialized skills, with fallback retries and auto-grounding file refs, boosting reliability on tricky tasks. The dashboard's real-time SSE streams, heatmaps for run comparisons, and skill browser expose autonomous agentic AI internals transparently. Developers grab it from GitHub autonomous agents repos to iterate fast on prompts and hit leaderboards.

Who should use this?

AI engineers benchmarking autonomous agents in AI on PAC1's CRM, security, and inbox challenges. Researchers prototyping github autonomous coding agents or autonomous agents and multi-agent systems for file ops. Teams inspired by autonomous agents Microsoft visions, testing against real traps like prompt injections.

Verdict

Grab it for PAC1 experiments—clear README, uv sync setup, and dashboard make tuning straightforward despite 16 stars and 1.0% credibility. Maturity shows in 86% score, but add tests for production; strong base for custom autonomous agents GitHub projects.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.