Active Stack (What to Actually Use)

The wiki is a second brain that informs development, not a storage locker. A captured skill / tool / loop is worthless until it is actively used. This page is the curated subset Kevin should pull into live workflows, by activity. A capture is not done until it lands here and in routing (see the Activation gate). Source: User, 2026-06-13

Inventory: 251 executable registry entries, 252 runtime links including .system, and 38 automation definitions. Most imported depth now lives as folded references inside parent router skills; the lists below are the active subset. Routing health after the 2026-07-06 agent-docs/slash-command pass: npm run routing-doctor = 0 dead pointers, 0 dark skills, 0 weak triggers; five discoverable entries remain intentionally outside direct trigger routing (copilotkit, illo, zero-tech-debt, pdf-operations, red-team-business). check-tool-discovery passes for 69/69 tool/design/skill pages with routing_summary frontmatter, discovered through central or category routers. npm run trigger-eval passes 170/170 (incl. autoreview, handoff, web-design-engineer, animated-component-libraries, find-skills, ai-seo, programmatic-seo, core-web-vitals, accessibility, webapp-testing, vibe-code-doctor, no-sus-code-doctor, loop-me, skill-creator no-op pruning, agent-eval-library, loopy, AI SDK testing, the Kevin engineering-flow skills, design-engineering-polish branches, img2threejs, hatch-pet, codex-pet fallback, the promoted SaaS/infra service skills, service-cli-registry, and graphify-sidecar).

How agents should use this page

Use this page when the task is "what should I be using?" or when a new capture might change the recommended workflow. This is not the full inventory. It is the curated working set that should shape defaults.

Decision rule:

1. Start with the workflow section that matches the task.
2. Load the named skill or loop before improvising.
3. If the tool family overlaps with another family, read Capability Routing Map.
4. If this task reveals a better default, update this page and the resolver in the same pass.

The active stack is intentionally opinionated. Long-tail tools still live in Skill Resolver, but this page names the tools Kevin's agents should reach for without needing a fresh debate every time.

When editing this page or any routing surface, close the loop with npm run build-index, qmd update && qmd embed, npm run routing-doctor, and a wiki/log.md entry. Routing changes that are not indexed and tested are just decorative confetti with a YAML header.

Sandbox comparison status

Use Agent Sandboxes for current provider selection. The recovered Daytona, Blaxel, Cloudflare Sandbox SDK, Vercel Sandbox, Docker Sandboxes, CubeSandbox, Smol Machines, AWS Fargate, Runloop Devboxes, Declaw, HopX, Fly.io Sprites, and CodeSandbox pages are reference candidates. The September 25 review checked documentation and package metadata, not deployed workloads. Keep the incumbent governed provider until a task-specific comparison proves a reason to switch.

The loops to run live (the spine)

By workflow

Choosing between overlapping tools. Start at Capability Routing Map when the question is "which browser/search/model/API tool should I use?" Duplicated skill/tool pages are allowed when they represent different layers: skill = procedure, tool/concept = capability facts and tradeoffs.

Building (Dedalus / loop / any repo). agent-iteration-loop (spine) · Cross-Harness Model Routing when Claude/Fable-class models should plan or judge while Codex executes, uses the computer, or verifies UI/UX · loopy for bounded feedback-loop design · agent-eval-library when the agent's behavior, evaluator, or self-improvement loop changes · conductor skill + Conductor app for local Mac parallel workspaces, run scripts, harness auth, checks, diff review, and PR flow · visual-plan for inspectable diagrams/storyboards before broad implementation · shadcn/improve / improve + Borrowed Intelligence (Drain It into Plans) (plan) · kevin-engineering-flow as Kevin's build-flow router · setup-kevin-engineering-flow once per repo when issue/domain config is missing · grill-with-docs -> Agent Engineering Skills -> Agent Engineering Skills -> Agent Engineering Skills -> Agent Engineering Skills -> Agent Engineering Skills for Kevin's source-derived build flow · Agent Engineering Skills for test-first vertical slices · Agent Engineering Skills for repro-first debugging · Agent Engineering Skills + Code Taste for deep modules, seams, and readable abstractions · Agent Engineering Skills for architecture-deepening reports · Agent Engineering Skills for messy issue/backlog intake · Agent Engineering Skills for merge/rebase conflicts · security-doctor-chain for pre-ship security composition · no-sus-code-doctor (strict PR quality gate to 100) · Security and Review Skills (pre-merge closeout, installed — "autoreview → merge") · gstack-review / gstack-qa / dogfood / Browser Testing Skills (general review + QA) · beta-dx-walk (first-run UX) · agent-browser (default live verify and authenticated exploration) · OpenCLI only as a guarded project-scoped adapter candidate after a repeated authenticated website/Electron route is proven; use an isolated Chrome profile, inspect exact commands/permissions/dependencies, and require explicit mutation authority · systematic-debugging · tests / refine · doctor-enforcement + react-doctor / Security and Review Skills (+ React Security Doctor) + Security and Review Skills for calibrated repo-wide vuln scans + Cnfast · RTK (Rust Token Killer) / lean-ctx (cut agent token cost 60–90%) · commit-and-push / yeet / babysit (ship) · Agent Operations Skills (pass a task to another agent/session) · cleanup-terminals-browsers / Agent Broom (hygiene).

Frontend / design engineering. taste (router with folded visual-mode references) · design-engineering-polish (motion build/review/audit/opportunity/vocabulary parent with pinned Emil branches) · StyleX when agent-authored UI needs typed token/design-decision constraints · frontend-design (UI direction/build) · web-design-engineer (polished browser artifacts in a named style + anti-cliché, see Refined Interface Design (Avoiding AI-dashboard Slop)) · DESIGN.md for persistent agent-readable visual identity files · Impeccable as a reviewed external project-design-language and deterministic detector route after local skills · Kive for brand-controlled physical-product representations, reusable studio direction, and campaign variants that must not look generically AI-generated · animated-component-libraries + Component Library Sources (Radix/Base primitives, shadcn, UI Skills Pack, Astryx Design System, Cloudflare Kumo, Blocks.so, Dashboard Blocks / Efferd, Kokonut UI, Bklit UI, Chanh Dai, Plotly/Evil Charts by analytical-engine versus product-treatment need, Sonner, Skiper UI, Cult UI, Magic UI, Motion Primitives, React Bits, Fancy Components, Origin UI, Shaders.com, and one-inspected-item-at-a-time Componentry before hand-rolling; Iconsax is a proprietary-license reference, not a bulk source) · Design Effects and Motion Sources for bounded CSS-cursor and effect studies · Design Inspiration Galleries / Recent for current source-linked discovery before following the original creator · Motion Library Ranking / Motion.dev for Motion.dev, Framer Motion, Anime.js, GSAP, and copy-paste effect routing · Web3D Library Ranking for img2threejs image-to-procedural reconstruction, Three.js, WebGL/WebGPU, R3F, shaders, no-code WebGL, and asset/mockup routes · Modern Web VPS Stack when the frontend stack is TanStack Router/Start + Tailwind + Vite/Rolldown and deployment posture matters · CSS Field Sizing for Baseline 2026 form-control autosizing · Design Arena for directional human-preference benchmark checks on visual/code-generation models · Realtime Colors for in-context palette/type preview · Haikei for exported SVG assets · PhotoGIMP only for a human Photoshop-to-GIMP transition after backing up the GIMP profile · gstack-design-review (end-to-end audit after craft/design routing) · make-interfaces-feel-better · frontend-frontier · web-design-guidelines · vercel / nextjs folded React/Next performance references · Browser Testing Skills (LCP/INP/CLS) · Frontend and Design Skills (WCAG 2.2) · shadcn · motion-framer / gsap-scrolltrigger · loading-screens · stitch when the Stitch MCP/workflow is explicitly in play · excalidraw-diagram-generator / Hand-Drawn Diagrams for Excalidraw-style diagrams.

SaaS / infra service work. Use Capability Routing Map first, then service-specific parent router skills only when the task touches that service: cloudflare, supabase, clerk, tanstack, nextjs, vercel, stripe, posthog, sentry, firebase, azure, aws, terraform, gws, plus focused singletons such as kafka, hetzner, vite, astro, zustand, rust, and vue. For TanStack Router/Start + Tailwind + Vite/Rolldown apps that may run on Hetzner/Coolify with Cloudflare in front or as edge sidecars, route through Modern Web VPS Stack before choosing individual service skills. Each parent router owns its folded plugin/upstream references under references/skills/, so load only the exact child reference matching the service, framework, or failure mode. For GitLab, Google Cloud, Docker, Kubernetes, Datadog CI, warehouses, CMS/search, comms, and other SaaS/infra products without a specific skill, load Cloud, Data, and Service Skills and Service CLI Coverage. Production, auth, email, recovery, OAuth, and package identity use a reputable paid registrar with named renewal ownership; DigitalPlat FreeDomain is a bounded demo/learning route only. Mutating cloud/auth/workspace state still requires target/account confirmation.

File artifacts / hosted sites. Use the imported Codex primary-runtime skills when the deliverable is a file artifact: documents, spreadsheets, pdf, presentations, and template-creator. These must travel with their bundled scripts, references, assets, templates, and agents/openai.yaml. For OpenAI Sites, route sites-building before sites-hosting; read .openai/hosting.json, reuse project IDs, save/version source state, then deploy only saved versions.

Harness-specific Cursor built-ins. Keep these visible but narrow: onboard for Cursor's focused onboarding interview, automate for Cursor Automations editor handoff, and review / review-bugbot / review-security for Cursor's built-in readonly review subagents. Kevin's generic review spine stays no-sus-code-doctor -> autoreview -> gstack-review.

Second brain (this wiki). Capture Ingest Protocol (every "add it") · x-bookmark-absorb + X Bookmark Artifact Audit (npm run audit:x-bookmark-artifacts) for one-by-one source/media/object/workflow review · Wiki Brain Operating Model as the public graph interface · graphify-sidecar for repo topology/path/affected-node questions without replacing qmd · agent-eval-library (npm run agent-self-eval; npm run agent-self-eval:test) for receipt-based self-improvement proof and controlled evaluator epochs · learn-from-projects · loop-me (interview Kevin for recurring work loops and specs) · loopy (turn repeated workflows into bounded loop skills) · skill-creator (create skills, then prune no-op instructions and token-heavy sediment) · liteparse / LiteParse for fast local PDF/layout extraction and parse-once document QA · MarkItDown - Universal File-to-Markdown Converter / MinerU / Defuddle - Web Content Extraction for source extraction by medium and difficulty · Agent Operations Skills + the npx skills CLI (find/add/check/update) — search skills.sh before hand-rolling any capability; keep installed skills current with npx skills check · Capability Harvest Pattern (npm run harvest:capabilities) + npm run promote:capabilities for broad skills/MCP/CLI/pattern/source refreshes and promotion audits · Wiki Quality Audit (npm run wiki-quality, optional -- --network --limit=N) for read-first/router/source/tweet-artifact/skill-load corpus audits · wiki-doctor · signal-detector (always-on rule).

Content / career. content-strategy · Writing and Content Skills -> Writing and Content Skills / Writing and Content Skills for long-form article drafting · social-draft -> humanizer -> kevin-voice (publish chain) · HyperFrames for deterministic code-generated launch/demo clips · ReClip - Self-Hosted Video Downloader only as the loopback human batch UI for rights-cleared source intake (agents use direct yt-dlp) · slack-voice · career-ops (upstream career-ops-v1.15.0, 56K+ stars, checked 2026-06-30) · Skills as Onboarding for compiling top-performer workflows into skills · last30days · Launch Post Playbook (launches). SEO/GEO: Writing and Content Skills (get cited by AI search - GEO/AEO) · programmatic-seo (pages at scale) · seo-audit · og-metadata-audit (incl. runtime check for OG images that don't load).

Research / external. last30days · agent-reach · find-skills · nia-docs · Self-Hosted Research Workspaces when a bounded source-grounded project genuinely benefits from a notebook UI/API (Open Notebook first; Odysseus is a guarded full-operator evaluation) · Reading and Reference Shelf for Kevin's curated source shelf, beginning with Stripe Press · The Bitter Lesson for the general search/learning versus hand-engineered-domain-knowledge distinction · Awesome Evals when the research is about agent evals, benchmarks, verifiers, or RL environments · Design Arena when benchmarking visual/code-generation model taste from human-preference arenas · Red Queen Gödel Machine when self-improvement loops risk evaluator saturation or reward hacking: freeze the active judge within an epoch, propose harder challengers, promote only at anchored independent checkpoints, and invalidate displaced-evaluator scores. Reference/watch-only: the unaffiliated observeco/rqgm-core package and OpenEnv for open-source agentic RL environment standards. Track via Signal Radar Workflow and Source Compile Workflow, but do not make either a default project dependency until Kevin is building an agentic RL training/eval environment.

Codex pets / generated sprites. hatch-pet is the default official $imagegen route for Codex-compatible 8x9 pet atlases, contact sheets, validation, and pet.json packaging. codex-pet remains the RunComfy-backed single-reference route when that workflow is explicitly desired or $imagegen is unavailable.

Coverage (every skill is reachable)

After the 2026-06-28 compaction pass, the active executable skill surface is routed through parent routers and direct singletons, while a small description-only tail stays discoverable without being recommended as a default. The big clusters that used to be description-only are now slug-routed through parent routers such as taste, stripe, vercel, aws, gws, rust, and vue, plus the design, animation/3D, testing, lint, security, and ops singletons listed in SKILL-RESOLVER.md. New captures must either join a route or be explicitly left as description-only by the Activation gate below.

Loops on a cadence (automations - wire the never-run ones)

project-briefing (daily) · source-compile (evening) · daily-work-summary · frontend-frontier-radar / signal-radar (weekly) · skill-maintenance · capability-harvest (weekly skills/MCP/CLI/pattern/source refresh) · cross-project-learning. npx tsx scripts/check-freshness.ts shows which are overdue; the never-run items are the next operational gap to close.

Activation gate (so capture -> use, every time)

When any pattern / tool / skill is added to the wiki it is not done until:

1. Routed - capability-routing-map.md and SKILL-RESOLVER.md row (+ harness projection if a compact loader needs it; hub task row if it changes how agents work).
2. Recommended - added to the right section of this page (or explicitly marked reference-only).
3. Proven - a config/trigger-evals.json phrasing so npm run trigger-eval fires it.
4. Surfaced - cadence work -> an automation; event-driven -> noted in the loop family.

Enforced by Capture Ingest Protocol (Activation step) and surfaced at session start (Operational Heartbeat). The brain recommends; it does not just store.

---

Timeline

• 2026-09-25 | Recorded sandbox additions as references without changing the default execution provider. Source: User request; Agent Sandboxes primary-source review, 2026-09-25

• 2026-08-10 | Refreshed inventory/routing counts and activated the reviewed utility-roundup decisions without making every repository a default: Open Notebook for bounded research projects, Odysseus as a guarded operator lab, ReClip for authorized human batch intake, PhotoGIMP for opt-in GIMP familiarity, and FreeDomain for noncritical demos only. Source: X/@DivyanshT91162; current first-party sources

• 2026-07-14 | Added Impeccable, the Plotly/Evil Charts chart boundary, Reading and Reference Shelf with Stripe Press, The Bitter Lesson, and Is This Tech Dead? to the active/reference routing surface. Source: User request, 2026-07-14

• 2026-07-13 | Added Kive as the active physical-product representation route: preserve product invariants, encode a reusable photographic system, use reference-led art direction, and reject generic AI gloss before accepting campaign imagery. Source: User request, 2026-07-13; Kive official site and docs

• 2026-07-07 | Refreshed frontend/design source routing after the component-library audit: Kokonut UI, Bklit UI, Magic UI, Motion.dev, Motion Library Ranking, and Web3D Library Ranking are explicit active-stack routes beside Astryx, Kumo, Skiper, Cult, React Bits, Three.js/WebGL, and R3F coverage. Source: User request, 2026-07-07

• 2026-07-07 | Added Sonner to active frontend/design component-source routing so focused component libraries like toast/notification primitives are discoverable beside chart sources such as Evil Charts. Source: User request, 2026-07-07; Component Library Sources

• 2026-07-06 | Promoted Cross-Harness Model Routing into the active build stack so Claude/Fable-class models own taste/judgment while Codex handles clear-spec execution, computer use, and UI/UX verification. Source: User-provided Theo X post and screenshot, 2026-07-06

• 2026-07-06 | Promoted Modern Web VPS Stack into active frontend and SaaS/infra routing for TanStack Router/Start, Tailwind, Vite/Rolldown, Hetzner/Coolify, and Cloudflare architecture decisions. Source: User request, 2026-07-06

• 2026-07-02 | Promoted graphify-sidecar into the active second-brain stack for repo topology/path/affected-node work, with qmd remaining the default compiled-wiki retrieval layer. Source: User request, 2026-07-02

• 2026-06-30 | Deep-reviewed the UI Skills and Dashboard Blocks bookmark artifacts: UI Skills Pack is now an active external UI-skill discovery shelf, and Efferd Dashboard Blocks is routed through Shadcn Registry Directory (Top Blocks) for dashboard composition references. Source: X bookmark artifacts 2036445894426374213, 2057754739392291170

• 2026-06-30 | Deep-reviewed career-ops and Anthropic sales-ramp artifacts: refreshed career-ops to upstream career-ops-v1.15.0 / 56K+ stars and promoted Skills as Onboarding into content/career capability routing. Source: X bookmark artifacts 2040820834428658008, 2058247099544948837

• 2026-06-30 | Deep-reviewed the Design Arena and field-sizing bookmark artifacts; promoted Design Arena into eval/frontend routing and CSS Field Sizing into frontend CSS routing with current support caveats. Source: X bookmark artifacts 2066940737011560652, 2031672243630293137

• 2026-06-30 | Refreshed routing health claims after routing-doctor: 57/57 routing_summary tool/design pages are discoverable, while copilotkit, illo, and red-team-business remain description-only skills rather than active defaults. Source: routing-doctor, 2026-06-30

• 2026-06-29 | Refreshed React Bits as a first-class component-library source with official install/docs details and linked it directly from the active frontend/design stack. Source: User request, 2026-06-29; https://reactbits.dev/

• 2026-06-29 | Added security-doctor-chain as the explicit composition route for pre-ship security work: React Security Doctor, deepsec, threat model, secure-coding review, adversarial bug hunt, and final no-sus gate now route as one chain when needed. Source: User request, 2026-06-29

• 2026-06-29 | Promoted react-doctor and visual-plan into executable skill routes, keeping React Doctor / React Security Doctor as active build gates and BuilderIO-style visual plans as inspectable pre-implementation artifacts. Source: User request, 2026-06-29

• 2026-06-29 | Added the requested doctor/source stack confirmations: Security and Review Skills, no-sus-code-doctor, Security and Review Skills / React Security Doctor, and shadcn/improve remain active build/review gates; Astryx Design System, Cloudflare Kumo, Motion Primitives, Haikei, and Realtime Colors are now active frontend/design sources. Source: User request, 2026-06-29

• 2026-06-29 | Promoted Loopy as the active route for bounded loop-skill crafting and Agent Self-Improvement Eval Library / agent-eval-library as the active self-improvement benchmark route, with Red Queen Gödel Machine as the evaluator-hardening principle. Source: User request, 2026-06-29

• 2026-08-10 | Routed the second-brain loop through Wiki Brain Operating Model and X Bookmark Artifact Audit so evidence feeds durable objects, workflows, skills, tools, playbooks, and diagrams. Source: User request, 2026-08-10

• 2026-08-11 | Added the pinned img2threejs external skill as the active reference-image-to-procedural-Three.js route after 673 passing tests, exact license/release audit, and rendered-gallery review; ordinary Three.js/R3F integration and Blender/asset workflows retain their existing jobs. Source: X 2079267787298746709; img2threejs/img2threejs@d667338

• 2026-08-11 | Revalidated design-engineering-polish as the single active parent for Emil Kowalski's now-ten-skill design-engineering source. Folded build, review, audit, opportunity, vocabulary, and guarded Apple-motion branches; delegated prototype, library selection, and Sonner overlaps to existing owners; and added five passing route cases instead of globally installing every upstream command. Source: X 2069742622252445866; emilkowalski/skills@78761e1; current skills.sh replay

• 2026-08-11 | Corrected the active Red Queen route to controlled evaluator epochs and added executable invariant tests: no mid-epoch mutation, anchor-backed independent checkpoint promotion, selective invalidation, fixed-anchor-only cross-epoch comparison, and explicit limitations. Kept the unaffiliated observeco/rqgm-core package watch-only rather than presenting it as paper code. Source: X 2071285506630160761; arXiv:2606.26294v2; CaMLSys explainer

• 2026-06-28 | Compacted noisy imported and compatibility skill families into parent routers: aws, azure, firebase, gws, stitch, posthog, sentry, clerk, cloudflare, tanstack, stripe, supabase, terraform, vercel, nextjs, taste, pixel-art, rust, and vue. Runtime skill links dropped to 214 while folded references preserve the child prompts. Source: User request, 2026-06-28; scripts/compact-skill-families.ts

• 2026-06-28 | Integrated the fresh source-compile delta: HyperFrames became an active content/demo-video route after 3D keyframing, Security and Review Skills was reinforced as the complete repo security sweep, and OpenClaw Harness stayed watch/reference-only for harness architecture after the GitHub stars diff. Source: X Bookmark Ingest 2026-06-28; GitHub Starred Repos Harvest - 2026-06-28

• 2026-06-29 | Renamed the live imported engineering router to kevin-engineering-flow and the setup route to setup-kevin-engineering-flow so Kevin's stack exposes Kevin-owned route names while preserving Matt Pocock as upstream provenance. Source: User correction, 2026-06-29

• 2026-06-27 | Completed the source-imported engineering flow: kevin-engineering-flow, setup-kevin-engineering-flow, Agent Engineering Skills, Agent Engineering Skills, and Agent Engineering Skills joined the existing domain/prototype/PRD/issues/TDD/debug/design/triage stack. Source: User request, 2026-06-27; github.com/mattpocock/skills commit 5d78bd0903420f97c791f834201e550c765699f8

• 2026-06-27 | Promoted the local conductor skill into the active build stack for Conductor app setup, local Mac parallel workspaces, harness auth, MCP, run scripts, files-to-copy, diff/check review, and PR flow. Updated live skill inventory counts. Source: User request, 2026-06-27; Conductor docs

• 2026-06-26 | Integrated GitHub stars harvest into the active stack: DESIGN.md for agent-readable visual identity, Awesome Evals for eval research maps, MinerU for hard document extraction, AWS Agent Toolkit for AWS-supported agent plugins/MCP/skills, and Hand-Drawn Diagrams as an Excalidraw-style diagram candidate. Source: GitHub Starred Repos Harvest - 2026-06-26

• 2026-06-25 | Added Matt Pocock's no-op pruning thread to the active skill-quality route: skill-creator now owns tightening bloated skills and deleting lines whose removal would not change behavior. Source: X/@mattpocockuk, 2026-06-24

• 2026-06-25 | Promoted Matt Pocock's long-form writing trio into the active content route: Writing and Content Skills for raw material, Writing and Content Skills for choose-your-path article beats, and Writing and Content Skills for paragraph/block shaping. Source: X/@mattpocockuk, 2026-06-25; Source: mattpocock/skills commit 5d78bd0903420f97c791f834201e550c765699f8

• 2026-06-25 | Added Wiki Quality Audit to the second-brain active stack for corpus-wide read-first/router/source/tweet-artifact/skill-load checks. Source: User request, 2026-06-25; scripts/wiki-quality-audit.ts

• 2026-06-25 | Added Agent Operations Skills to the active loop/second-brain stack as the route for discovering day-to-day work worth delegating to agents, then promoting specs to skills or automations. Source: X bookmark reingest, 2026-06-25

• 2026-06-25 | Added Code Taste to the active build flow and ai-test-kit to the Vercel AI SDK route after the latest X bookmark ingest. Source: X/@shadcn thread, 2026-06-25; X/@zirkelc_, 2026-06-24; X/@aisdk, 2026-06-25

• 2026-06-25 | Promoted the core Matt Pocock engineering skills into the active build flow: Agent Engineering Skills, Agent Engineering Skills, Agent Engineering Skills, Agent Engineering Skills, Agent Engineering Skills, Agent Engineering Skills, Agent Engineering Skills, and Agent Engineering Skills, while keeping overlapping review/teaching/handoff/writing utilities on existing local routes. Source: github.com/mattpocock/skills commit 8370e760d0251a3738e006aeacec6d1cb31dd208

• 2026-06-25 | Expanded Capability Harvest Pattern in the active second-brain loop from registry refresh to exhaustive skills/MCP/CLI/pattern/source discovery, including templates, design techniques, eval/review patterns, browser workflows, runtime substrates, and automation candidates. Source: User request, 2026-06-25

• 2026-06-25 | Promoted Media Generation Skills as the default official $imagegen Codex pet route, with Media Generation Skills kept as the RunComfy-backed single-reference route. Source: autoreview closeout, 2026-06-25

• 2026-06-26 | Expanded active frontend/design component-source routing to Radix/Base primitives, shadcn, Blocks.so, Chanh Dai, Evil Charts, Skiper UI, Cult UI, Magic UI, React Bits, Fancy Components, Origin UI, and Shaders.com. Source: User request, 2026-06-26; Component Library Sources

• 2026-06-25 | Promoted Component Library Sources and animated-component-libraries into the active frontend/design stack, with Skiper UI and Cult UI as default modern component sources alongside Magic UI, React Bits, and shadcn/ui. Added Google Workspace CLI as the upstream reference for gws-* service work. Source: User request, 2026-06-25

• 2026-06-25 | Added first-class tool discovery enforcement: pages with routing_summary frontmatter must have a ## Routing Summary section and be mentioned from central routers or natural category routers such as tools/README and design/README. npm run routing-doctor now runs check-tool-discovery. Source: User request, 2026-06-25

• 2026-06-24 | Promoted 39 official/reputable skills from the latest capability harvest into the active stack: specialized Azure, current Firebase, Google Workspace write/triage, Vercel AI SDK/CLI, Better Auth, Stitch, and Excalidraw skills. Drifted registry slugs were skipped or mapped to current upstream names. Source: User request, 2026-06-24; skills.sh; upstream skill repos

• 2026-06-24 | Updated active routing priority so code review defaults to no-sus-code-doctor then Security and Review Skills before gstack-review, and design/frontend work defaults to design-engineering-polish, frontend-design, and web-design-engineer before gstack-design-review. Source: User request, 2026-06-24; route renamed 2026-06-30

• 2026-06-27 | Skill catalog health now uses Skill Registry plus family synthesis pages, and scripts/sync-skills.sh --check blocks stale registry rows or retired individual skill articles. Source: User request, 2026-06-27

• 2026-06-24 | Skill catalog health became fail-closed under the old 1:1 executable skill/catalog pairing model; superseded by the generated registry model on 2026-06-27. Source: User request, 2026-06-24

• 2026-06-23 | Added no-sus-code-doctor to the active building stack as the strict PR/diff quality gate for "no sus code", "nothing Windsor/Anny would flag", and score-to-100 reviews. Source: User request, 2026-06-23

• 2026-06-22 | Added Capability Harvest Pattern as the active route for broad skills.sh/MCP/CLI registry refreshes without mass-installing harvested capabilities. Source: User request, 2026-06-22

• 2026-06-22 | Added Cloud, Data, and Service Skills / Service CLI Coverage as the active horizontal route for SaaS/infra CLIs not covered by service-specific skills. Source: User request, 2026-06-22

• 2026-06-22 | Promoted a first batch of official service skills from the capability harvest: Cloudflare, Supabase, Clerk, Next.js, Firebase, Azure, Google Workspace, and browser-use as opt-in. Source: User request, 2026-06-22; skills.sh

• 2026-06-20 | Added OpenEnv as a reference/watch-only research item rather than an active default, with monitoring routed through signal-radar and source-compile. Source: User request, 2026-06-20

• 2026-06-18 | Updated inventory counts (144 personal, 17 Claude, 43 Dedalus skills; 21 automations) and added explicit agent usage rules for the active recommendation surface. Source: User request, 2026-06-18; filesystem inventory

• 2026-06-13 | Created as the active-recommendation surface; first audit of skills + loops into the active subset; Activation gate added to the capture protocol and proactive surfacing to session startup. Source: User, 2026-06-13

• 2026-06-17 | Refreshed routing health after section audit: zero dark/unrouted executable skills; noted the remaining orphan catalog warnings as reference-only pages. Source: routing-doctor, 2026-06-17

• 2026-06-17 | Added Capability Routing Map as the active surface for overlapping tool families; clarified that skill/tool duplicates are valid layer mirrors when routed. Source: User, 2026-06-17