Bug Receipt v1.4.1 — enforced version metadata

Version tracking guard

• adds metadata.version: "1.4.1" to the Agent Skill frontmatter;
• makes npm run skill:validate require semantic version x.y.z;
• makes validation fail when skill metadata and package.json versions differ.

This patch does not change the discovery description, skill instructions, receipt contract, or runtime behavior. The v1.4 robustness evidence therefore remains the applicable behavioral result.

Verification

• 33/33 tests passed
• full npm run check passed
• source and installed skills passed system quick_validate.py
• CI passed on Windows and Ubuntu with Node 20 and 22
• npm audit: 0 vulnerabilities

Bug Receipt v1.4.0 — robust, evidence-safe closeouts

Highlights

• Strengthens automatic bug and incident closeout with a mandatory complete BUG RECEIPT.
• Adds explicit executed now, supplied, or mixed evidence provenance.
• Covers security redaction, rollback recovery, diagnosis-only authority, and rolling version skew.
• Ships receipt schema v2 while retaining version 1 compatibility.
• Hardens installation with safe replacement, backup, rollback, and unsafe-destination refusal.

Measured result

In the pre-registered four-case v1.4 robustness cohort:

• Skills ON: 18/20 assertions (90%)
• Skills OFF: 5/20 assertions (25%)
• Lift: +65 percentage points
• Exact McNemar p = 0.000244
• Natural routing: 4/4
• Complete receipt and correct status: 4/4
• Secret leaks: 0
• Candidate actions: 0

This is a bounded comparison against the no-skill arm, not a claim against every alternative. Methodology and SHA-verified evidence are in benchmarks/RESULTS.md.

Verification

• 33/33 repository tests passed
• lint, TypeScript build, distribution validation, and packed-artifact smoke passed
• CI passed on Windows and Ubuntu with Node 20 and 22
• npm audit: 0 vulnerabilities

Install

npx --yes github:lMysticl/bug-receipt#v1.4.0 install

Bug Receipt v1.3.1 — verified, cross-platform audit gate

Bug Receipt v1.3.1 publishes the verified audit-gate benchmark and fixes cross-platform provenance verification.

Measured result:

• Skills ON: 18/20 (90%)
• Skills OFF: 7/20 (35%)
• Lift: +55 percentage points; exact paired McNemar p = 0.003418
• Complete receipts: 4/4; routing: 4/4; evidence-safety regressions: 0
• Independent corroborating run: 19/20 vs 7/20 (+60 points), p = 0.000488

The report SHA is now calculated over canonical LF text, so npm run benchmark:audit-gate is reproducible across Windows and Linux checkouts.

Scope: four frozen supplied-evidence closeout cases on gpt-5.6-sol/xhigh. This is a bounded result, not a universal ranking.

Bug Receipt v1.3.0 — verified audit gate

Verified impact

• Pre-registered A/B: 18/20 (90%) with Bug Receipt vs 7/20 (35%) without it (+55 percentage points).
• Exact paired McNemar p = 0.003418.
• Natural routing: 4/4; automatic receipts: 4/4; evidence-safety regressions: 0; candidate actions: 0.
• Corroborating run: 19/20 vs 7/20.

What changed

• Mandatory complete receipt for bug and incident closeout, even under concise prompts.
• SHA-verified raw benchmark reports, executable decision gate, and transparent quarantine history.
• Updated premium benchmark presentation and AI-readable metadata.

The measured claim is intentionally bounded to the published four-case coding-agent bug-closeout cohort.

Bug Receipt v1.2.0 — reproducible benchmarks

Measured evidence

• Deterministic receipt invariants: 20/20 passed.
• Fresh Codex routing probes: 3/3 selected Bug Receipt.
• Paired skills-ON quality: 12/12 assertions passed.
• Paired skills-OFF quality: 12/12 assertions also passed — this bounded sample does not prove a quality uplift.
• Raw telemetry, fingerprints, exact prompts, and the quarantined failed holdout are published in the repository.

Also included

• A more precise BLOCKED evidence-package rule for cross-system failures.
• A public benchmark section on the project site and llms.txt.
• Reproducible validator benchmark via npm run benchmark:validator.

Full report: https://github.com/lMysticl/bug-receipt/blob/main/benchmarks/RESULTS.md

v1.1.0

Full Changelog: v1.0.1...v1.1.0

Bug Receipt v1.0.1

Patch release: the lightweight CLI now rejects unknown receipt fields exactly like the published JSON Schema. CLI help also reads its version from package.json to prevent release drift. Verified by 13 automated tests, both skill validators, production build, CI, and GitHub Pages deployment.

Bug Receipt v1.0.0

No “fixed” without receipts.

Bug Receipt v1.0.0 ships a portable Agent Skill, deterministic receipt validator, JSON Schema, safe cross-agent installer, and a polished local-first web experience.

Highlights:

• VERIFIED, PARTIAL, and BLOCKED evidence contract
• Agent Skills support for Codex, Claude Code, Copilot, and project-local installs
• bug-receipt check and bug-receipt sample CLI commands
• 12 deterministic contract, schema, installer, and boundary tests
• No API key, telemetry, or hosted receipt service

Install directly from GitHub:

npx --yes github:lMysticl/bug-receipt install

The activation corpus is a maintained boundary surface; it is not a claim that every model or harness has been behaviorally exercised.