Releases: pyjeebz/PreScale
Releases · pyjeebz/PreScale
Release list
v0.2.1
A small patch release.
Fixed
prescale mcpwithout the optional extra printed the wrong install command (pip install 'prescale'instead of'prescale[mcp]') — rich was interpreting[mcp]as console markup. It now prints the correct command, with a regression test.
Docs
- Rewrote the README and refreshed the PyPI description for the full 0.2.x toolkit: load test → confidence band →
investigate→ MCP server → regression/CI → launch profiles.
Install: pip install prescale (add [mcp] for the agent server).
Full Changelog: v0.2.0...v0.2.1
v0.2.0
PreScale grew from a single load-test command into a full toolkit.
Added
- Saved runs & a versioned result contract — every run is stored under
.prescale/runs/;prescale history/showre-open them,prescale schemadocuments the format.run --jsonand the saved file share one shape. - A trustworthy verdict — a deterministic confidence band ("survives ~90 (80–110)") with a stable/uncertain flag, at no extra cost. Warmup runs by default (
--no-warmup), plus--repeat Nand--think-time S. prescale investigate— finds the route that breaks first, then probes it to diagnose why (a bottleneck class with confidence + evidence) and prescribe a fix. Fully local, no LLM.- MCP server (
prescale mcp) — load-test from a coding agent (Claude Code, Cursor).pip install 'prescale[mcp]'. Safe by default — local hosts only unless allowlisted. - Regression + CI —
--fail-under N,prescale compare(capacity diff of two runs), and--fail-on-regression(band-aware), plus a copy-paste GitHub Actions workflow. - Launch profiles —
prescale run --profile product-huntframes the verdict as "would survive a Product Hunt #1 launch".
Changed
--jsonnow emits the full versioned Result and is robust to long lines.
Install: pip install prescale (add [mcp] for the agent server).
Full Changelog: v0.1.0...v0.2.0