Repository navigation
Releases
v0.4.0 — SQL optimization evidence and local dashboard
SQLHarness v0.4.0
SQLHarness now brings SQL optimization evidence into a local browser dashboard, alongside its CLI and MCP server. Compare proposed rewrites, check their results for the tested inputs, and follow the attempts that led to a selected query version.
Highlights
- Local dashboard: browse activity and sessions, compare retained SQL versions, review elapsed time, CPU and IO, and copy comparisons as Markdown.
- Clearer optimization evidence: result equivalence, IO changes and policy outcomes are separate; selected cases and incomplete evidence retain their context.
- Coding-agent workflows: MCP support and a bundled skill connect SQL inspection and measurement to the optimization workflow.
- Guarded mutations: persistent SQL changes use the mutation workflow and the operator's dashboard approval policy.
- SQL Server, Azure SQL and PostgreSQL: engine-specific measurements, compact execution-plan inspection, and bounded output.
- Demo walkthrough: the README now shows a real dashboard populated with explicitly synthetic data, with a reproducible source-checkout fixture.
Downloads
Choose sqlharness-win-x64.zip, sqlharness-linux-x64.tar.gz, or sqlharness-osx-arm64.tar.gz. Each archive contains a self-contained binary with the .NET 10 runtime included. Verify its SHA-256 digest against SHA256SUMS, extract it, and put sqlharness on PATH. Unix users should ensure the extracted binary is executable.
sqlharness-skill.zip contains the SQLHarness skill. Extract its sqlharness folder into the skills directory supported by your coding agent. Configure the MCP server separately using the README instructions.
Notes
- Upgrade from v0.3.1: the CLI surface has changed. The old
ping,schema,counts,space, andgaincommands are no longer registered. Usequeryfor inspection SQL,doctorfor local installation/configuration checks, andcapabilitiesto discover the current interface. Update scripts and agent instructions against--helpbefore upgrading. - Comparisons require
--subjectand--scopeby default; use a stable subject label and--scope wholefor a whole-query comparison. Thecompare.requireContextsetting controls this policy. Restart MCP after changing it. - Start from the example target configuration and choose the correct engine for each profile. Existing users should review the current README before upgrading across the changes since v0.3.1.
- Result equivalence applies to the supplied test inputs and configured comparison mode; it is not a proof for every possible input.
- SQL, parameter values and execution plans stored in the optional local journal may contain sensitive data. Review journal settings before enabling that storage.
- The screenshot metrics are synthetic demo data, not performance claims.
Validation
- Release tag and binary build:
4ce7f7f16b684dde88905d71fa674e854168d1fd. - Local Windows verification: UI 570 passed; Core 4,919 passed / 11 skipped; MCP 537 passed / 4 skipped. The full
just verifygate passed. Core also passed with two logical processors, and all 29 MCP reader tests passed with a deliberately long temporary-directory prefix. - GitHub CI: passed on Windows and Linux (run).
- Windows download checks: archive extraction, executable SHA-256, version/build ID, embedded dashboard, and MCP initialization/tool discovery/capabilities passed. No database connection was used for the download smoke checks.
- Self-contained, single-file, untrimmed builds succeeded for all three platforms. Unix archives preserve executable mode 0755. SHA256SUMS covers the four downloadable archives.
- Linux/macOS archives were cross-built on Windows. Linux CI additionally runs the native code/tests and its own published-server smoke; these exact Linux/macOS download binaries were not launched on those operating systems. macOS runtime acceptance remains unverified.
v0.3.1
v0.3.0
v0.2.0
v0.1.0
Full Changelog: https://github.com/rgonek/sqlharness/commits/v0.1.0
