v0.6.0 — inspect evidence and try the complete debugging workflow

FailTrace 0.6.0 makes a complete failure investigation easier to try and its saved evidence easier for coding agents to inspect.

• Try the complete workflow: the guided demo measures 7 passing and 3 failing trials, reduces input to ["BUG"], rejects an unrelated crash during Verify, checks a healthy candidate, and packages the original failure for replay.
• Inspect saved evidence through MCP: the new read-only failtrace_inspect_run tool and Core inspectRunEvidence API page trial records and bounded stdout/stderr chunks without executing saved commands.
• Connect with one command: version-pinned setup examples cover Codex, Claude Code, Cursor, and generic stdio clients. The installed package exposes seven MCP tools.
• Reproduce a real race: npm run case:p-memoize runs the pinned affected/fixed case and its negative controls from a source checkout.
• Review measured performance: the README links sanitized five-sample benchmark evidence, timing ranges, and the limits of concurrency and early classification.

Run the exact GitHub archive with Node.js 22.12+:

npm exec --yes --allow-remote=root --package=https://github.com/LBarimi/FailTrace/releases/download/v0.6.0/failtrace-0.6.0.tgz -- failtrace demo

The command-scoped remote option is for npm 12's URL installation policy. The verified npm installation is:

npx --yes failtrace@0.6.0 demo

The public npm archive matches the attached GitHub archive byte for byte. A fresh-cache public installation passed the demo, bundle replay, Core and CLI Verify, saved-evidence inspection, and all seven tools through an npx-launched MCP server.

Verify reports observations from the recorded sample. A healthy candidate with no target match does not prove elimination or statistical improvement. The published performance measurements are from one Windows/Node host and are not portable timing guarantees.

Validation: 301 tests passed locally, along with type checking, build, fresh installed-package checks, demo and bundle replay, and the p-memoize case. All six Windows/macOS/Linux and Node.js 22/24 CI jobs and all three performance jobs passed on the release commit. The release workflow independently installed and tested the prepared archive.

The attached SHA256SUMS and release.json identify the exact archive and source commit. npm publication uses this public HTTPS archive without repacking it.

v0.5.0 — verify a proposed fix

FailTrace 0.5.0 connects a reproduced failure to a proposed code change with durable, inspectable evidence.

• failtrace verify and Core verifyFix run a fixed candidate budget against an eligible captured baseline.
• The failtrace_verify MCP tool exposes the same Core contract to coding agents.
• Results distinguish a target that remains, a healthy sample where it was not observed, unrelated command failures, infrastructure failures and inconclusive evidence. They do not claim elimination or statistical improvement.
• Baselines can hash declared input, setup and source files or capture bounded Git identity with raw tracked-file hashes. The current command, canonical working directory and intentional changes remain explicit.
• Installed-package tests cover Core, CLI and MCP Verify, cancellation and the unrelated-error guard.
• Real affected/fixed workflows cover a Prettier formatting defect and the p-memoize pending-call race. The race schedules are controlled, not presented as a natural failure-rate estimate.

Try the guided existing demo:

npx --yes failtrace@0.5.0 demo

Then follow the Verify workflow to capture your own baseline before editing code. Requires Node.js 22.12+ and npm.

The attached release.json records the exact source commit and six-job CI run. The archive was built and installed outside the checkout by the release workflow; SHA256SUMS identifies the reviewed bytes.

v0.4.0 — bounded execution and durable evidence

FailTrace 0.4.0 adds execution controls and more efficient durable evidence for repeated debugging experiments.

• Opt-in run --concurrency N, also available through Core and failtrace_run; the default remains sequential.
• Bisect/minimize stop after their repeated-trial threshold is decided, preserving independent final minimization checks.
• Incremental statistics and authoritative per-trial records avoid repeated growing metadata writes. loadRun reconstructs large compact summaries and recoverable records after interruption.
• File-set minimization requests copy-on-write copies with an ordinary-copy fallback. Bundles preserve concurrency and replay the original full trial budget.
• Reproducible benchmarks document measured improvements and remaining costs. Concurrency can change failure behavior through shared resources; early classification is not a full-budget frequency estimate.

Try the installed package:

npx --yes failtrace@0.4.0 demo

Requires Node.js 22.12+ and npm. CLI/Core/MCP, package installation and the real Prettier investigation are gated by CI on Windows, macOS and Linux with Node.js 22 and 24. The attached release.json records the exact source commit and CI run; SHA256SUMS identifies the reviewed archive.

Dedicated code-fix verification remains planned in this release. Performance scope · Changelog

v0.3.1 — a first debugging result in one command

FailTrace is now available from the public npm registry. Try it from any directory:

npx --yes failtrace demo

For regular use:

npm install --global failtrace

This patch updates the package README and agent setup instructions to use the shorter registry installation route. It also includes the documented historical Prettier case and packaging checks that keep local investigation evidence and case dependencies out of published archives. The Core investigation behavior is unchanged.

The package includes mcpName and server.json metadata identifying the existing stdio server as io.github.LBarimi/failtrace, ready for official MCP Registry publication.

The Prettier case uses pinned affected/fixed releases and reduced a maintainer-authored 464-character fixture to 11 characters while preserving the upstream formatting defect. The fixed release succeeds on the reduced input. Its relocated bundle was replayed after installing the target dependencies in a fresh directory.

Validation includes 191 Core/CLI/MCP tests, four case-oracle tests, the complete real-case investigation, typecheck/build, and installed-package checks on Windows, macOS, and Linux with Node.js 22 and 24. The attached release.json identifies the exact source commit, successful CI run, and SHA-256 digest; SHA256SUMS covers the package archive.

The same prebuilt archive is available here for users who prefer a versioned GitHub asset. Node.js 22.12+ and npm are required. Runtime dependencies are installed by npm; no source checkout or compiler is needed.

Agent setup · Changes · Results and limits

Successful installation and these maintainer-run examples establish distribution and compatibility. Independent usage and repeat adoption remain to be established.

v0.3.0 — a first debugging result in one command

FailTrace turns repeated debugging experiments into inspectable evidence for developers and coding agents. This is the first prebuilt GitHub release.

Try the real demo from any directory:

npm exec --yes --allow-remote=root --package=https://github.com/LBarimi/FailTrace/releases/download/v0.3.0/failtrace-0.3.0.tgz -- failtrace demo

The demo records 7 passes and 3 failures, minimizes six JSON elements to ["BUG"], and creates a replayable bundle. Evidence stays local. It exits 0 when the expected results are verified.

For regular use:

npm install --global --allow-remote=root https://github.com/LBarimi/FailTrace/releases/download/v0.3.0/failtrace-0.3.0.tgz
failtrace demo

Requires Node.js 22.12+ and npm. The archive includes compiled code; no clone or TypeScript build is required. Runtime dependencies are installed by npm. The npm registry package itself has not been published.

Included: repeat/predicate/compare, repeated-trial Git regression isolation, text/JSON/files/environment minimization, reproduction bundles, and five official-SDK MCP tools. New in 0.3.0: guided demo, complete MCP predicate-match counts, agent setup guides, and installed-package CI.

Validation: 191 tests plus typecheck/build and installed-package checks across Windows/macOS/Linux with Node 22 and 24. Exact commit CI. Commit: 823163a.

Check SHA256SUMS before installation if you download the asset separately. Read agent setup, changes, and limits.

Share a real debugging result or first-install problem through the optional workflow report. No telemetry or account is required to use FailTrace.

The command-scoped --allow-remote=root option permits this explicitly requested archive on npm 12, which blocks URL installs by default. It does not change npm settings. Older npm versions that do not recognize it can omit it. See npm's URL install policy. This corrects the install commands recorded in the original archive's documentation and build metadata; the package and checksum are unchanged.