📋 Changelog

Run generate_soft404_artefacts.py and stamp_worker_manifest.py against

Introduce a soft-404 protection system to prevent SPA fallbacks from being indexed as real pages. Adds scripts to generate artefacts (scripts/generate_soft404_artefacts.py), stamp the worker with a build-time path manifest (scripts/stamp_worker_manifest.py), and perform a targeted KV purge (scripts/purge_soft404_kv_cache.py). The deploy workflow now runs generation and stamping before publishing the worker. Adds public/path-manifest.json, a minimal public/404.html, and appends idempotent legacy `/slug/ -> /YYYY/MM/slug/` redirects to public/_redirects. _worker.template.js and the staged public/_worker.js are updated to consult the manifest and refuse to serve or cache unknown paths (returning a real 404 with noindex headers). Docs (docs/SEO.md) were updated with rationale and a manual runbook (including a one-shot KV purge) to complete the rollout.

Analytics are handled by Plausible, so the KV-metadata view counter had no

The old README described the html-cache worker and a now-missing

Wrangler v4 removed the colon form `kv:key put` in favour of `kv key put`.

Wrangler requires Node 20+, so the Node 18 pin caused the KV search index

Change the workflow to probe for required system binaries and Python modules first, accumulating any missing packages into MISSING_PKGS. If nothing is missing, skip apt entirely; otherwise run apt-get update/install only for the missing items. This avoids unnecessary dpkg lock waits on long-lived self-hosted runners and speeds up the job startup. The script checks jq, bc, avifenc, optipng, jpegoptim, cwebp, python3-pip and python3-venv and still prints versions after installation.

Add new probes to .claude/settings.local.json (Bash(xxd) and WebFetch for www.bing.com) and update the stored IndexNow key in .indexnow_key (key rotated). These changes enable additional local checks and refresh the indexing key without altering other configuration.

Include a keyLocation field in IndexNow submission payload so the API knows where the verification key is hosted (scripts/submit_indexnow.py). Also add WebFetch(domain:www.indexnow.org) to local Claude settings to allow fetching IndexNow resources during analysis (.claude/settings.local.json).

Upgrade action usages across workflows to newer releases (actions/cache -> v5, actions/cache/restore/save -> v5, actions/upload-artifact/download-artifact -> v7, actions/setup-node -> v6, slackapi/slack-github-action -> v3.0.1) to keep CI tooling up-to-date and compatible. Also add Bash patterns for gh run/api to .claude/settings.local.json to include gh CLI checks in local scans.

Upgrade GitHub Actions used in workflow files: actions/checkout@v4 -> actions/checkout@v5 across all workflows, and actions/setup-python@v4/@v5 -> actions/setup-python@v6 where present. Updated workflows: deploy-static-site.yml, enable-cloudflare-indexing.yml, quality-checks.yml, rollback-site.yml, secret-scan.yml, spell-check-consolidated.yml. No behavioral changes aside from using newer action releases; existing fetch-depth and Python version settings are preserved.

Add 'consolidated-inline-styles' to the EXCLUDED tuple in scripts/extract_critical_css.py to prevent double-counting of inline noscript styles during critical CSS extraction, matching the handling of other excluded assets like 'brutalist-theme' and 'fonts.css'.

Add a new scripts/html_transformer.py that performs SEO fixes, image-to-<picture> conversion, performance hints, critical CSS extraction/inlining, head deduplication and HTML minification in a single parse cycle per file. Update .github/workflows/deploy-static-site.yml to wait for dpkg locks during apt operations and to replace the previous multi-step image/SEO/perf/CSS/HTML pipeline with a streamlined CSS optimization step and a single-pass HTML transformer step. Simplify Makefile targets to call the new transformer and to consolidate the CSS pipeline (remove separate critical extraction step). These changes reduce repeated HTML parsing, simplify CI steps, and add robustness to runner package installs.

Rework the GitHub Actions deploy flow and staging URL conversion:

Replace selective file-based purge with a purge-all workflow step that POSTs /.purge?all=true and requires CACHE_PURGE_TOKEN; step name and GitHub summary fields updated and response count parsed with jq. Update worker handler to support the all=true query: iterate HTML_CACHE.list to delete KV keys and corresponding Cache API entries, return JSON with purged count and timestamp, and improve missing-parameter error and cache request origin handling.

Refine the regex used to extract taxonomy names from page titles. Use a non-greedy match for patterns like "Docker Archives - James Kilby" and add a fallback to the previous pattern when the 'Archives' token isn't present. Also update the comment/example. This prevents incorrect captures for taxonomy names that include hyphens or other characters.