Installation¶

Requirements¶

Tool	Supported	Notes
Python	3.11, 3.12, 3.13	`requires-python = >=3.11`
Node	18.x, 20.x	Only for the `npx trace-sec` shim
OS	Windows x64, Linux x64, macOS Intel + Apple Silicon	CI-tested
RAM	4 GB lite / 8 GB full	Full = AI inference with torch

Lite install (recommended for contributors)¶

No torch, no model downloads — everything needed for tests and audits:

WindowsLinux / macOS

python -m venv .venv
.venv\Scripts\activate
python -m pip install --upgrade pip
pip install -e ".[dev]"

python3 -m venv .venv
source .venv/bin/activate
python -m pip install --upgrade pip
pip install -e ".[dev]"

Full AI install (inference on your machine)¶

pip install -r requirements.txt

This pulls torch, transformers, and onnxruntime (hundreds of MB). On low-RAM or metered devices, stay on lite — detection still works, neural scoring gracefully degrades. See Lite or Full.

NPM shim¶

npm install -g trace-sec
trace

or zero-install: npx trace-sec audit .

The install step provisions a Python environment under ~/.trace/venv. Set TRACE_SKIP_POSTINSTALL=1 (it is skipped automatically when CI=true) to skip it, for example when you only want the agent skills from the package.

Agent skills¶

npx skills add VK-Amogh/TRACE --skill trace   # universal: 75+ agent harnesses
npx trace-sec install-skill                   # trace + trace-fix + MCP config

With the trace skill installed you do not need to install anything else yourself: the agent installs and configures TRACE.

See Agent skills for global installs, per-harness options and npm-based linking.

Verify the install¶

trace doctor
trace audit ./testbed_hardcore --format table

If python -m trace_engine.cli says No module named 'trace_engine', set PYTHONPATH=src (or use the repo launchers trace.bat / trace.cmd / trace.ps1, which do it automatically). More in Troubleshooting.