Changelog

• Hardened the automated editorial ingestion pipeline for daily production use.
• Added ingestion health reporting, stronger generated article validation, image retry/backoff, and Blob image smoke checks.
• Cleaned Read/Feed visibility so validated generated articles can show source-backed provenance without making the UI noisy.
• Preserved legacy article access by slug while keeping default latest surfaces focused on validated generated content.

+10 more

• Harden editorial ingestion
• Add release changelog sync
• Fix feed Agent Mag AI panel clipping
• Add PR Security Auto-Fix to tools and CLI

• fix(seo): auto-fix SEO issues from GPT-5.4 review

• Split fast model catalog sync from slow Firecrawl enrichment.
• Use OpenRouter API and Ollama catalog as the source of truth for active models.
• Filter malformed model rows from the public models API and prevent enrichment-only inserts.
• Stop the /models page from live-fetching Ollama during render and reduce the initial server payload.

+10 more

• Fix models catalog refresh pipeline
• Add GitHub PR screenshot workflow skill
• Harden editorial ingestion

• fix(seo): auto-fix SEO issues from GPT-5.4 review

• This update turns the Feed and Read surfaces into a real article engine instead of a mostly MDX-only display path. It adds a unified published-article reader that prefers active Cosmos articles, falls back to local MDX content, dedupes by slug/source URL, and feeds the homepage/feed helpers, /read, article detail pages, API responses, RSS, sitemap, AI sitemap, llms feeds, crawl manifest, and global search.
• It also adds ingestion status notifications so scheduled/manual article runs can email a scrape/write/publish summary to configured recipients.
• Added a Cosmos-first published article layer with MDX fallback and shared filtering/sorting helpers.
• Added a Firecrawl /v1/search + Azure OpenAI ingestion engine that discovers recent AI-agent sources, validates scraped markdown, rejects thin/duplicate/off-topic content, generates original cited Agent Mag articles, and records ingestion run/source metadata.

+10 more

• Add feed/read article engine
• Fix click navigation errors
• Add read editorial pipeline
• Productionize editorial automation

• fix(seo): auto-fix SEO issues from GPT-5.4 review

• Hosts the private sponsor workbench on partners.theagentmag.com and keeps /partners as the public sponsorship sales page.
• Added a shared partner host helper for partners.theagentmag.com and partners.staging.theagentmag.com.
• Updated proxy host detection to use the shared helper.
• Redirects non-local /partner-dashboard/ requests to https://partners.theagentmag.com/, so the dashboard is exposed through the partners subdomain instead of a public app path.

+10 more

• Host partner dashboard on partners domain
• Improve Agent Mag AI QA reliability
• Add feed/read article engine

• fix(seo): auto-fix SEO issues from GPT-5.4 review

• This update completes the Agent Mag PostHog integration on a dedicated branch. It moves PostHog client initialization into the Next.js 16 instrumentation-client path, expands the event taxonomy for core product funnels, adds non-blocking server-side capture across key API routes, and gives the CLI a persistent anonymous install ID so successful skill/model installs can be attributed without requiring an account.
• Added early PostHog browser initialization with /ingest reverse proxying, uihost, defaults: "2026-01-30", debug-friendly replay, exception autocapture, web vitals, dead-click capture, and masking for password/token/secret fields plus data-mask-recording.
• Updated PostHog rewrites so static assets, /ingest/array/:path, /ingest/decide, /ingest/flags, and the catch-all route are ordered correctly.
• Standardized client events around lowercase snakecase names and stable properties such as surface, source, slug, category, rank, query, resultcount, plan, provider, UTM fields, and referrer.

+10 more

• Integrate PostHog analytics
• Fix skills metadata build
• Simplify sponsor workbench
• Add Agent Mag AI product brain

• fix(seo): auto-fix SEO issues from GPT-5.4 review

• Replaces shadcn Button, Badge, and Input components on the /models page with raw HTML elements matching the editorial style used on /skills and /jobs
• Adds a proper hero section with serif heading, description, and CLI install box
• Removes the FAQ-style "What is..." info block and redundant "Local model install" box
• | File | Change |

+10 more

• fix: align /models page with editorial design system
• Add contextual Agent Mag AI model recommendations
• Integrate PostHog analytics

• fix(seo): auto-fix SEO issues from GPT-5.4 review
• fix: align /models page with editorial design system

• Fixes the individual model detail pages and keeps the model list cards easy to use after adding install command copy controls.
• Replaced the incompatible RelatedContent usage on /models/[slug] with local related-content cards that match the page data shape.
• Limited related Ollama model lookups on detail pages so opening a single local model page does not wait on the full Ollama catalog.
• Restored full-card navigation for model rows while keeping the install command copy control independently clickable.

+10 more

• Fix model detail pages
• ci: auto tag releases after merged PRs
• fix: restore original skills page UI
• fix: add missing space in footer copyright text

• fix: add missing space in footer copyright text
• fix: restore original skills page UI broken by Codex

• When reading an article in focus mode and clicking "Back to Feed", focus mode stayed on — the feed rendered without the sidebar
• Now both "Back to Feed" links (top and bottom of article) disable focus mode on click before navigating home
• Extracted a reusable BackToFeed client component that accesses the focus mode context
• Page alignment: about, authors, partner pages now match masthead/contact/feedback (max-w-2xl, gap-10, consistent heading sizes and spacing)

+10 more

• fix: turn off focus mode when navigating back to feed
• fix: page alignment, focus mode nav, article breadcrumbs
• fix: theme toggle shows target mode, not current
• feat: professional footer with 6-column navigation

+10 more

• fix(seo): auto-fix SEO issues from GPT-5.4 review
• fix(models): move metadata from client page to server layout
• fix(ci): pass PostHog env vars at build time in deploy workflows
• fix(analytics): wire PostHog provider into app and add /ingest proxy

+57 more

• Initialized the public Agent Mag skills registry with README, templates, example skill structure, contributing guidance, and MIT license.
• Established the skill format as auditable config bundles containing manifests, prompts, tool schemas, and optional settings.
• Synced the registry from the skills.sh ecosystem, growing from the first import of 264 skills to the current 492-skill registry snapshot.
• Connected the registry to the Agent Mag install flow through npx agentmag add skill-name.

• Added Agent Mag branded banners, footer links, and registry guidance for contributors.
• Documented valid categories, manifest expectations, PR workflow, and security expectations for non-executable skills.
• Clarified when a contribution belongs in Agent-mag/skills versus Agent-mag/tools.

• Initialized the public Agent Mag tools repository with an MIT license, contributing guide, and repo-level documentation.
• Added SEO/GEO Auto-Fix Reviewer, the first open-source Agent Mag tool for reviewing PRs for SEO issues and automatically proposing fixes.
• Added Agent Sounds, a Claude Code sound effects plugin with original royalty-free sound packs for task lifecycle cues.
• Added npx agentmag install instructions so tools can be discovered and installed through the Agent Mag CLI path.

• Reworked tool READMEs with Agent Mag branded structure, full-width banners, and consistent footers.
• Clarified the distinction between installable config skills and executable tools.
• Documented contribution requirements for community-submitted tools.