v1.0.0 — bounded failure experiments and reviewable reproduction
FailTrace 1.0.0 provides local failure experiments for developers and coding agents: reproduce a known failure, compare evidence, isolate a revision, minimize an input, verify a proposed fix and package a replayable reproducer.
FailTrace 1.0.0 is available on npm and in the official MCP Registry. Try the published package with npx --yes failtrace@1.0.0 demo, or use the verified GitHub archive below.
Try this exact archive with Node.js 22.12.0 or later:
npm exec --yes --allow-remote=root --package=https://github.com/LBarimi/FailTrace/releases/download/v1.0.0/failtrace-1.0.0.tgz -- failtrace demoThe demo measures a controlled failure, reduces its input, rejects an unrelated crash during verification, checks a healthy fixed sample and creates a bundle that reproduces the original failure. Replaying the intentionally failing bundle exits 1. The command-scoped URL option is for npm 12; older npm versions that do not recognize it can omit it.
Changes since 0.6.0:
- Bound command output, retained input copies, input complexity and metadata. Incomplete experiments preserve evidence and return an inconclusive outcome.
- Keep compact schema-2 bisect results with complete counts and saved-run references. Use
loadRun(candidate.run.metadataPath)or MCP inspection to retrieve full evidence. - Exclude original logs/metadata and captured environment values from new bundles by default. Explicit sharing choices, environment prerequisites and a relative file/hash manifest make the contents reviewable.
- Define the supported 1.x Core, CLI, JSON and MCP interfaces and stored formats. Offline fixtures from verified public 0.3.1 through 0.6.0 packages check older records, interrupted recovery, Verify migration and new bundle replay.
Read the migration guide, compatibility contract, resource limits and bundle sharing guide.
Release gates cover Windows, macOS and Linux with Node 22 and 24, plus package installation on the minimum Node 22.12.0. The attached release.json identifies the exact source commit and CI run; SHA256SUMS identifies the archive inspected and installed before publication.
Commit 8a6c0ed6bbcfec56be1c99d3123bf8dd6a02cc59 passed all seven CI gates. Public GitHub and npm archive downloads matched SHA-256 f7793d39ce5083c7030d9dc1da2a189b91e922bec65684f52366a8dd9fe4d913 and the reviewed SHA-512 integrity. Independent GitHub-URL and fresh-cache npm installations passed the demo, Verify controls, bundle manifest checks and replay. The installed MCP server listed seven tools; run-limit, Verify and saved-evidence inspection calls passed. Its versioned Registry record is active after the publication workflow. Public npm metadata was checked for private paths and secrets. These are maintainer distribution checks, not evidence of independent adoption.
Commands run locally with the caller's permissions. Review selected source, input and command text before sharing; the manifest does not sanitize content. target_not_observed describes a healthy finite sample and does not prove that a defect was eliminated.