HN Showcase
Back to browse
GitHub Repository

👻 See what your AI agent almost did. Record agent decisions including Phantom Branches — the actions considered but rejected.

2 starsPython

GhostTrace – See rejected decisions in AI agents

by AhmedAllam0·Feb 18, 2026·1 point·1 comment
Visit ProjectView on HN

AI Analysis

●●SolidNiche GemShip It
The Take

Recording what an agent considered — not just what it executed — is a tidy, concrete insight. GhostTrace already gives record/replay commands, a .ghost.json schema and a --show-phantoms terminal replay so you can inspect rejected actions and the agent's reasoning. The thing that will decide if this takes off is integrations (LangChain/OpenAI Agents/CrewAI) and the promised web/VS Code UIs; without those it's a very useful niche tool, not yet a platform.

Target Audience

ML engineers, LLM/agent developers, observability/debugging engineers, and anyone building with LangChain or other agent frameworks

Post Description

Hey HN,

When an AI agent makes a decision, it evaluates several options and picks one. The rest disappear forever – you never see what it almost did or why it rejected the alternatives.

I built GhostTrace to fix that. It captures "Phantom Branches": the actions your agent considered but rejected, with the reasoning for each rejection. All saved to a .ghost.json file you can replay and inspect.

Quick demo:

ghosttrace record Recorded 4 decisions with 5 phantom branches Saved to gt_a1b2c3d4.ghost.json

ghosttrace replay gt_ghost.json --show-phantoms

Step 1: read_file → src/auth.py REJECTED: write_file (premature) REJECTED: search_codebase (too broad)

pip install ghosttrace

PyPI: https://pypi.org/project/ghosttrace/

It's framework-agnostic for now, but what agent frameworks should I integrate first? LangChain? CrewAI? OpenAI Agents SDK?

Feedback very welcome!

Similar Projects