The packages in this repository (pivotkit-core, pivotkit-react, pivotkit-element,
pivotkit-xlsx, pivotkit-report, pivotkit-cli, pivotkit-server, pivotkit-mcp) and the
container image built from it. A view (report template), the data, a loaded file
(CSV, JSON, a workbook read by pivotkit-xlsx), a print template and a host's theme tokens can come
from users; anything that lets them run code, load or read something they should not (a file outside
the directories pivotkit-cli --allow-dir names, a resource from an HTML report), or makes an
exported CSV, XLSX, HTML report or SVG dangerous to open is in scope.
For the REST service (pivotkit-server) and the MCP server (pivotkit-mcp) the callers, their
views, parameters and data, and an AI model are untrusted too. In scope are, among others: a request
the server makes where it should not (SSRF through an HTTP source, a parameter, an upload store or
the issuer's metadata, past the address rules in Security);
a token, a backend's secret or a download link that leaks into a result, a log line, an error or a
page; a download link that someone other than the person it is bound to can open, or that works after
it expired or was used; and a field's values or rows that reach a caller or a model above the level
the policy gives them.
A view cannot run code, but it decides how much work the engine does. The limits in LIMITS
(Limits on the documentation site) bound it: the size of a view (fields per axis, values), the rows
encode accepts, the cell updates computePivot does (rows x subtotal levels x values, heavier for
medians and unique counts), the nodes of its key trees and the cells it aggregates, and the cells of
the dense grid and chart. Each is counted before the memory it guards is allocated, or while it grows. The defaults are sized
for a browser tab: a pivot at the defaults can still take seconds and up to about 1.5 GB of memory,
and the data itself (rows x columns) is the host's to bound, since only its rows are limited. A host
that takes views or data from untrusted users should lower the limits to what its reports need and
catch PivotkitError. A way to exhaust memory or CPU around the limits, or far beyond what the
documentation site says the defaults allow, is in scope. Vulnerabilities in a dependency (for example Apache
ECharts or fflate) belong to that project; tell us too if a package ships an affected version.
Please report privately through GitHub's
Report a vulnerability
(Security tab, Advisories), not in a public issue. If you cannot use GitHub, email oss@eize.dev.
Include the package and version, the host (browser, framework), steps to reproduce and what you expected.
The latest released version of each package.