What pgrundev/pgbot shipped
Written by FoxPlug from the repository's own releases; not affiliated with Pgbot. Paste your own repository in the box below and FoxPlug writes the same weekly update for your product. This page is an automatic summary of the releases, pull requests and commits in github.com/pgrundev/pgbot; Pgbot did not write it and does not use or endorse FoxPlug. Every line links to the change it describes.
Get a weekly update like this for your product, free
Week of October 12, 2026
What shipped
- Ported the replica-identity integration fixture to pggo, fixing the collect package, which had been unbuildable since pgx was removed from go.mod. Pull request #119
- Stopped baseline comparisons from producing deltas or a nothing-material-changed message after a statistics reset or server restart. Pull request #64
- Changed the health collector to discard rate calculations that span a counter reset, including resets that a negative-delta check would miss. Pull request #73
- Kept configuration comparison events correct by distinguishing an unavailable settings read from a successfully collected empty override map. Pull request #75
- Separated unavailable query statistics from empty workloads, so a failed statistics read no longer prints an empty success table, and MCP
top_queriesnow reports an available flag. Pull request #79 - Required at least four snapshots before regression analysis, since three interval observations are needed by the sustained-shift detector, and shared that minimum across analysis, CLI rendering, guidance and the MCP description. Pull request #81
- Stored unavailable table and activity measurements as gaps rather than numbers, kept observed zero connection counts, and preserved the positive connection gauge when a reset invalidated rates. Pull request #87
- Measured archive age from the collected snapshot time instead of wall-clock time, so the same stored input gives the same finding when recomputed. Pull request #89
- Serialized first-use store initialization and made fingerprint metadata insertion race-tolerant, so only one initializer creates the legacy notice. Pull request #91
- Returned text report write failures, including short writes, to command callers so an incomplete or rejected report no longer reports success. Pull request #93
Why it matters
This week fixed several cases where stored or displayed results could be wrong after a statistics reset, a restart, or a failed collection instead of showing that data was unavailable. It also made report write failures surface to callers and fixed the collect package build, which had been failing in CI. For developers using the project, the output should be more trustworthy and the build should pass again.
Changelog entry
- Ported the replica-identity integration fixture to pggo, fixing the collect package build. Pull request #119
- Suppressed baseline comparison deltas and messages after a statistics reset or server restart. Pull request #64
- Discarded health rates that span a counter reset. Pull request #73
- Preserved configuration events across empty and unavailable settings. Pull request #75
- Distinguished unavailable query statistics from empty workloads and added an available flag to MCP
top_queries. Pull request #79 - Required at least four snapshots for regression analysis across analysis, CLI, guidance and the MCP description. Pull request #81
- Stored unavailable table and activity observations as gaps and kept the positive connection gauge after a counter reset. Pull request #87
- Measured archive age at the collected snapshot time. Pull request #89
- Serialized first-use store initialization and made fingerprint metadata insertion race-tolerant. Pull request #91
- Returned text report write failures to command callers. Pull request #93
- Added an opper provider alias for explain and ask, reusing the existing chat completions provider. Pull request #118
This week: baseline diffs no longer compute deltas across statistics resets, health rates drop reset-spanning values, unavailable query and configuration stats stay distinct from empty ones, regression analysis requires four snapshots, and text report write errors now reach calle
This week's changes focus on results you can trust. Baseline comparisons stop after a statistics reset or server restart instead of printing deltas. Health rates are discarded when they span a counter reset. Unavailable query and configuration statistics are now kept distinct from genuinely empty ones, and unavailable measurements are stored as gaps rather than numbers. Regression analysis requires at least four snapshots, archive age is measured from the collected snapshot time, first-use store initialization is serialized, and text report write failures are returned to callers. The collect package build is fixed.
Week of September 14, 2026
What shipped
- SSH tunnel connections now resolve database hostnames on the bastion instead of locally, allowing connections to databases only reachable from the tunnel host. Pull request #47
- Connection strings with password query parameters are now redacted in logs, closing a gap in the privacy boundary for URL-form postgres:// strings. Pull request #53
- Windows targets now compile in CI alongside Linux and Darwin, catching platform-specific breaks before release. Pull request #44
- Pack smoke tests now run against both npm and bun, preventing wrapper changes that break bun from passing CI. Pull request #52
- Documentation now lists pnpm, Yarn, and bun as supported package managers alongside npm. Pull request #51
Why it matters
This week fixes two privacy and connectivity gaps—SSH tunnel hostname resolution and connection string redaction—while improving test coverage to catch regressions across package managers and platforms before release.
Changelog entry
- Fix(conn): resolve tunneled database hosts on SSH server Pull request #47
- Fix(conn): redact password query parameters in URL-form connection strings Pull request #53
- Ci: build Windows targets in CI matrix Pull request #44
- Test(npm): run pack smoke tests with bun as well as npm Pull request #52
- Docs(install): document pnpm, Yarn and bun alongside npm Pull request #51
Pgbot now redacts password query params in logs, resolves tunneled hostnames on the bastion, builds Windows targets in CI, and tests across npm and bun.
Pgbot's latest updates improve security and reliability: connection strings now properly redact password parameters, SSH tunnels resolve database hostnames on the bastion host instead of locally, Windows targets compile in CI to catch platform-specific breaks early, and test coverage extends to both npm and bun package managers. Documentation also clarifies support for pnpm and Yarn.
More from pgrundev/pgbot on GitHub
Merged pull requests
- chore(nix): update pgbot to 0.9.0 (#125, merged October 28, 2026 by nully0x) (source)
- feat:
pgbot sql explain— a query plan, read in plain English (#123, merged October 19, 2026 by alexshapalov) (source) - fix: stop idle MCP input on cancellation (#104, merged October 19, 2026 by DivyamTalwar) (source)
- fix: suppress baseline comparisons across statistics resets (#64, merged October 12, 2026 by DivyamTalwar) (source)
- feat(ai): add an opper provider alias (#118, merged October 12, 2026 by Felixkw12) (source)
- fix: discard health rates across counter resets (#73, merged October 12, 2026 by DivyamTalwar) (source)
- fix: preserve configuration events across empty and unavailable settings (#75, merged October 12, 2026 by DivyamTalwar) (source)
- fix: distinguish unavailable query statistics from empty workloads (#79, merged October 12, 2026 by DivyamTalwar) (source)
- fix: require enough snapshots for regression analysis (#81, merged October 12, 2026 by DivyamTalwar) (source)
- fix: preserve unavailable observations as gaps in stored trends (#87, merged October 12, 2026 by DivyamTalwar) (source)
- fix: measure archive age at the collected snapshot time (#89, merged October 12, 2026 by DivyamTalwar) (source)
- fix: serialize first-use store initialization (#91, merged October 12, 2026 by DivyamTalwar) (source)