Releases: fmind/www
Release list
v2.2.2
This release reduces repeated startup work and fixes deployment, calculator, and HTTP caching behavior.
- Render each article's Markdown once for HTTP, LLM text, and MCP; allow a 60-second startup probe budget while preserving scale-to-zero.
- Verify a named, no-traffic revision before manual promotion, and prevent generated revision names from causing infrastructure drift.
- Keep calculator pilot evidence specific to its model, node, and precision.
- Return 304 for valid weak or strong stylesheet cache validators.
- Refresh locked Python dependencies and pinned uv, Node, and Tailwind build tools.
- Include MCP Registry publication and verification in the standard release process.
Validation: 658 Python tests with 91.19% coverage, 74 local Chromium journeys, four infrastructure tests, image security/runtime checks, and link checks. Production qualification covers 90 browser journeys, ten Lighthouse audits scoring 100 in every category, public API/MCP checks, and CLI compatibility.
The serving image retains 44 HIGH advisory instances across eight upstream CVEs, with no available fixes and no CRITICAL findings. The exposure assessment and review deadline remain documented in SECURITY.md.
v2.2.1
MCP GET and HEAD requests now return 405 with Allow: POST immediately. The portfolio has no notification subscriptions, so idle streams no longer run into Cloud Run's 30-second timeout. MCP POST operations remain supported, and legacy capabilities correctly advertise that catalogs do not emit change notifications.
Updated the public agent skill, agent guidance, release procedure, README, and security review; corrected the 2.2.0 publication date.
Validation: 644 Python tests, container scan and runtime smoke, 74 local browser journeys, and link checks passed. The exact release commit was deployed and independently verified at 100% traffic. Production browser/Lighthouse qualification is recorded in GitHub Actions. Independent DNS/TLS checks and the published CLI's 52 live checks also passed. The serving-image scan has no fixable HIGH/CRITICAL findings; the existing unfixed advisories remain documented in SECURITY.md.
v2.2.0
🚀 Features
- (articles) Add section navigation and feed discovery
- (articles) Publish agents-md-speaks-unix (#31)
- (delivery) Promote verified candidates and scope CI identities
- (site) Link a cached stylesheet and describe pages accurately
🐛 Bug Fixes
- (sites) Keep calculator scenarios exact and accessible
🧪 Testing
- (browser) Wait for article rail resize layout
v2.1.0
[2.1.0] - 2026-09-19
🚀 Features
- (finops) Add billing export configuration and analytics guidance
🐛 Bug Fixes
- (contact) Show Luxembourg location on contact and QR pages
- (delivery) Verify ready releases and parallelize browser journeys
- (ci) Qualify production on isolated browser runners
📚 Documentation
- (release) Preserve published changelogs during generation
Validation
- Exact-commit CI/CD: canonical gate, immutable image scan/smoke, and verified Cloud Run rollout.
- Production qualification: 62 Chromium checks, 16 Firefox/WebKit checks, and ten Lighthouse audits scoring 100 in all five categories.
- Serving-image scan: no fixable HIGH/CRITICAL findings; eight upstream CVEs without fixes remain documented in SECURITY.md.
v2.0.1
Improve discovery and attribution for Médéric Hurier (Fmind), freelance AI architect based in Luxembourg.
- Make Luxembourg visible in the homepage, biography, profile metadata, and agent interfaces.
- Include service descriptions, availability, and contact links in LLM summaries.
- Add article author links and retain author, date, and canonical URL information in Markdown exports.
- Keep website, contact-card, manifest, and MCP identity information consistent.
Validation:
- Full local qualification: 555 Python tests, 90.74% coverage, container scan/smoke checks, and 62 browser tests passed. Link checks passed.
- CI/CD and browser/Lighthouse qualification passed for commit
54026bad217b22c223b1cae44c0dcb0d20fa5d4f, including 16 Firefox/WebKit tests and ten local Lighthouse audits at 100 in every category. - Production: all 62 desktop/mobile browser tests passed, along with 18 endpoint checks, canonical redirects, and both supported MCP discovery paths. The ready revision serves the verified image with 100% of traffic, zero recent errors, and scale-to-zero preserved.
- Live Lighthouse scored 100 for SEO, accessibility, best practices, and agent browsing throughout. The updated article scored 100 in every category on desktop and mobile. The strict live portfolio performance gate did not pass: the initial matrix was 99–100; a confirmation under heavy workstation load ranged from 87–100. Server response timing stayed steady, but these measurements do not establish a stable perfect live performance score.
v2.0.0
The v2 release aligns the freelance AI architecture profile across portfolio, contact, manifest, and discovery metadata, and refreshes the delivery stack and operational guidance.
- Fix research links and extend link checking to the Python portfolio source.
- Update Python dependencies, the pinned build backend, Docker base image, Google providers, formatter plugins, and GitHub Actions.
- Remove the obsolete Debian patch layer and redundant range-response logic.
- Reduce README, AGENTS.md, SECURITY.md, and project skill text by about 58% while preserving release and privacy requirements.
- Enforce Cloud Run scale-to-zero and a service-wide maximum of five instances, with mocked infrastructure regression checks.
- Fix sandboxed Chromium startup in the Ubuntu Lighthouse workflow and require CI/live proof before tagging.
- Preserve existing article and decision-tool content.
Cloud configuration was reconciled before application deployment. State versioning and public-access prevention are enabled. Exact-commit CI scans and smoke-tests the pushed image digest before rollout.
See SECURITY.md for the remaining unfixed upstream Debian advisories. The unrelated vscode.email links in an unchanged article currently time out; the portfolio and documentation link audit passes.
Validation: 554 Python tests with 90.74% branch-aware coverage, 60 Chromium journeys both locally and against production, 16 Firefox/WebKit checks, mocked infrastructure tests, and portfolio/documentation link checks. All ten portfolio Lighthouse audits scored 100 in every category locally, in CI, and in production. The post-deployment OpenTofu plan has no drift.
Release-commit evidence: CI and deployment, browser and Lighthouse qualification, infrastructure, and serving-image security scan.
v1.8.1
v1.8.0
v1.7.0
🚀 Features
- (portfolio) Refine branding, contact export, and development reload
- (portfolio) Polish contact flow and reproducible assets
- (site) Publish agent research tools and reviewed portfolio improvements
⚡ Performance
- (fonts) Avoid unused code-font preloads
Reviewed agent discovery and research tools, portfolio/contact presentation, privacy disclosures, and maintenance skills. Fixed heading-anchor collisions, API catalog discovery, and broken article links.
Validation: full local delivery gate, external-link and infrastructure checks, Chromium journeys, Firefox/WebKit journeys, and the ten-audit Lighthouse portfolio matrix.