dumptriage-tools
Scripts and agents for capturing crash dumps on Windows and Linux (and crash reports on macOS) and sending them to
DumpTriage, which analyses crash dumps with AI and returns a root-cause
report: Windows BSOD memory dumps, minidumps (.dmp) and Linux core dumps.
The capture scripts work on their own. You don't need DumpTriage to use them: they only make sure a crash
leaves a dump you can open in WinDbg, gdb or crash.
Capture configuration
| Script | What it does |
|---|
windows/Set-WerLocalDumps.ps1 | Windows Error Reporting LocalDumps: full user-mode dumps of crashing processes to a folder you choose |
windows/Set-WerLiveKernelReports.ps1 | Live kernel reports: kernel dumps Windows takes while it keeps running (for example after a driver timeout), and how often it may take them |
linux/configure-dumps.sh | Raw ELF process cores, unlimited core size, retention, and optionally kdump for kernel crashes |
windows/Start-PerfTrace.ps1 | Performance traces: one WPR capture with PerfInsights.wprp while you reproduce the problem, zipped with its manifest, and -Upload to send it (no agent needed) |
linux/configure-tracing.sh | Performance traces: installs perf, checks tracefs and frame pointers, and --capture SECONDS takes one manual capture with the upload command (no agent needed) |
Collection scripts
Send dumps or reports without installing an agent.
| Script | What it does |
|---|
macos/dumptriage-collect.sh | Lists the last week's macOS crash reports, kernel panics and spin/hang reports, lets you pick, and asks before sending. --spindump samples a hanging app now; --profile, --sample and --xctrace send a timed performance capture (system-wide spindump, sample, Instruments' Time Profiler folded on the Mac) as a performance trace |
rmm/ninjaone/Invoke-DumpTriageCollect.ps1 | A NinjaOne Script Library script: sends each device's new minidumps on a schedule and writes the report link into a custom field (setup) |
Agents
Background agents that watch for new dumps, decide by policy whether to submit them (always, ask first,
notify only, or rules), and upload them to the DumpTriage API.
An agent needs an API key from DumpTriage; the
API docs describe the endpoints it uses.
Downloads
Every release carries each script
on its own, the Windows agent built (dumptriage-agent-windows-x64.zip: agent, tray app and installer), the Linux
agent with configure-dumps.sh and configure-tracing.sh (dumptriage-agent-linux.tar.gz), and SHA256SUMS. The newest of each is always at
https://github.com/performanceinsights-ai/dumptriage-tools/releases/latest/download/<file>, and listed on
performanceinsights.ai/downloads. A v* tag publishes a release
(.github/workflows/release.yml).
Each asset carries a build-provenance attestation from that workflow (#65). SHA256SUMS sits in the same release, so
it only catches a broken download; the attestation shows the file was built by this repository's release workflow:
gh attestation verify dumptriage-agent-windows-x64.zip -R performanceinsights-ai/dumptriage-tools
Tracing contracts
Automated performance-trace capture (website#1999) is defined once here so the three platforms cannot drift:
schemas/tracing-config.schema.json: the tracing config section (off by default,
backend, triggers[], limits, submission policy defaulting to AskFirst). Same keys in the Windows agent, the Linux
agent and ~/.config/dumptriage/tracing.json on macOS.schemas/trace-manifest.schema.json: the manifest.json uploaded with every trace.
It carries the host token, never the hostname.tests/fixtures/tracing/triggers.json: metric series in, fire / no_fire /
cooldown / capped out. Its semantics list defines the rules; the Python suite holds a reference evaluator, and every
platform's trigger loop is tested against every case.
Both schemas are release assets, and the website copies them for its own validation.
Tests
python3 -m unittest discover -s linux/dumptriage-agent/tests
dotnet test windows/dumptriage-agent/tests/DumpTriage.Common.Tests
dotnet test windows/dumptriage-agent/tests/DumpTriageAgent.Tests
pwsh -NoProfile -File windows/dumptriage-agent/installer/tests/Test-Install.ps1
bash macos/tests/test_collect.sh
bash linux/tests/test_configure_tracing.sh
bash linux/tests/test_install_perfetto.sh
pwsh -NoProfile -File rmm/ninjaone/tests/Run-Tests.ps1
.github/workflows/macos-trace-capture.yml (run by hand from the Actions tab, and on a PR that changes it or
macos/tests/burn.c) captures real spindump, sample and
xctrace output on GitHub's macOS 14, 15 and 26 runners and keeps it as a build artifact for 30 days, for testing the
service's macOS trace parsers on current macOS output (#41).
Security
Report vulnerabilities to [email protected] (subject "Security"), not in a public issue.
License
MIT
