FAQ¶
Does TRACE send my code anywhere?
No. Audits run on local CPU with zero tokens. Only live probes you explicitly
target receive HTTP requests, and ScopeGuard refuses non-local targets.
Why do I need the full install?
You usually don't. Lite (pip install -e .) covers scanning, findings, MCP,
and verification. Full (requirements.txt) adds neural scoring.
Which devices are supported? Windows x64, Linux x64, macOS Intel and Apple Silicon; Python 3.11–3.13; Node 18/20 for the NPM shim. See Installation.
Where do results live?
.trace/findings.json, .trace/findings.md, .trace/graph.db,
.trace/reports/. Never commit these. Each scan overwrites the findings file.
Which files does TRACE read? Source code in Python, JavaScript/TypeScript, Java/Kotlin, Go, PHP, Ruby, Rust, C# and Dart. HTML, CSS, JSON, YAML, TOML, SQL, Swift, C/C++, assets and build output are skipped. See What TRACE scans.
Why did a scan find 0 endpoints?
Either the framework is not supported yet or the path is wrong. Zero
endpoints is a coverage gap, not a clean result. Check
trace endpoints <path> and your --scope.
Does a short findings list mean the code is secure?
No. TRACE checks routes for ten vulnerability classes. Secrets, dependency
CVEs, cloud/IaC configuration and business logic are not assessed. The
trace-fix skill lists the gaps and complementary checks.
Why is a finding POTENTIAL or MEDIUM instead of CONFIRMED?
CONFIRMED needs a live probe to reproduce the exploit. Static-only runs
(no reachable --target) top out below that. See the
trace-fix skill for how to read confidence versus severity.
trace verify says INCONCLUSIVE. Is the fix good?
Unknown. The target was down or refused, the finding is no longer in the
stored scan, or static analysis still flags the code. Restart the target and
rerun verify before rescanning, because scans renumber finding IDs.
How do I give my AI agent TRACE?
Install the trace skill: npx skills add VK-Amogh/TRACE --skill trace for any
harness. The agent then installs and configures TRACE itself. See
Agent skills.