metaskillsloopsroutingrecommendationactive-useworkflowssecond-brainUpdated 2026-07-14

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: 246 executable registry entries, 247 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; four description-only imports remain intentionally discoverable but not promoted as active routes (copilotkit, illo, zero-tech-debt, 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 156/156 (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, 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.

The loops to run live (the spine)

Loop When Where
Agent iteration loop every feature / fix Browser Testing Skills
Loopy craft or audit a bounded feedback loop Loopy
Agent self-eval loop benchmark an agent run and harden the evaluator Agent Self-Improvement Eval Library + agent-eval-library
Borrowed intelligence audit -> bank plans now, execute cheap later Borrowed Intelligence (Drain It into Plans) + Frontend and Design Skills (/improve) -> plans/
Beta DX walk before exposing a flow to users Browser Testing Skills
Eval loop shipping AI output / content The Eval Loop (Slop Is an Output Problem)
Brain loop answer from compiled knowledge first The Brain-Agent Loop (qmd)
Graphify sidecar answer repo topology/path/affected-node questions cheaply Graphify Sidecar Workflow + graphify-sidecar
Loop-me interview discover recurring work to delegate Frontend and Design Skills -> workflow spec -> skill/automation
Hermes operator loop persistent personal agent OS / intake / software factory Hermes Mac Mini Agent OS + hermes-operator-stack

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 -> Frontend and Design Skills -> Agent Engineering Skills -> Google Workspace Skills -> Security and Review Skills for Kevin's source-derived build flow · Frontend and Design Skills for test-first vertical slices · Security and Review Skills for repro-first debugging · Frontend and Design Skills + Code Taste for deep modules, seams, and readable abstractions · Frontend and Design Skills for architecture-deepening reports · Browser Testing 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 / Frontend and Design Skills (general review + QA) · beta-dx-walk (first-run UX) · agent-browser (live verify) · 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) · Cloud, Data, and Service 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/component polish first; upstream emil-design-eng) · 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 before hand-rolling) · Motion Library Ranking / Motion.dev for Motion.dev, Framer Motion, Anime.js, GSAP, and copy-paste effect routing · Web3D Library Ranking for 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 · 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 · Frontend and Design 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 Security and Review Skills and Service CLI Coverage. 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 bookmark source/media/capsule review · Brain Capsules as the public brain interface · graphify-sidecar for repo topology/path/affected-node questions without replacing qmd · agent-eval-library (npm run agent-self-eval) for receipt-based self-improvement proof · 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 · Security and Review 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 · 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: Security and Review 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 · 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. Reference/watch-only: OpenEnv for open-source agentic RL environment standards. Track via Signal Radar Workflow and Source Compile Workflow, but do not make it 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

Related Pages

Agent Operations HubSkill ResolverSlash Command IndexCapability Routing MapCapture Ingest ProtocolBrain CapsulesX Bookmark Artifact AuditOpenEnvBorrowed Intelligence (Drain It into Plans)Browser Testing SkillsLoopyAgent Self-Improvement Eval LibraryRed Queen Gödel MachineNo One-Off WorkCompounding SystemsComponent Library SourcesStyleXDESIGN.mdModern Web VPS StackTanStack EcosystemCloudflare EcosystemCoolifyAwesome EvalsMinerUAWS Agent ToolkitConductor Programming Interview PrepGraphifyCross-Harness Model Routing

45 pages link here

Agent Capability RegistryConceptsAgent Company OSArchitectureAgent ExpectationsUSERAgent Operating SystemArchitectureAgent Operations HubMetaAgent Skill BundlesConceptsAgent SoulSOULBuilderIO SkillsToolsCapability Harvest PatternMetaCapability Routing MapMetaCapture Ingest ProtocolMetaCapture Ingest WorkflowWorkflowsCareer ProfileCareerConductor Programming Interview PrepInterview PrepCross-Harness Model RoutingWorkflowsCurrent WorkUSERCursor Rules OverviewAgent DocsGitHub Starred Repos Harvest - 2026-06-26ResearchGoogle Workspace CLIToolsHermes Mac Mini Agent OSArchitectureHow to Update the WikiMetaIs This Tech Dead?ToolsLinkedIn Article: How I Actually Work (Agent Setup)CareerLoop LibraryToolsLoopyToolsModern Web VPS StackConceptsMulti-Agent TopologiesConceptsQuality GatesMetaSession Startup ProtocolMetaSignal Radar CadenceHEARTBEATSkill ResolverSkill ResolverSkill Resolver FilingFiling Decision TreeSkill Sync WorkflowWorkflowsSkills FirstSOULStrengthsUSERTechnical StackUSERTool Hierarchy CompactAgent DocsTool Routing DisciplineSOULTools FilingFiling Decision TreeUser Model MapUSERUser ProfileUSERUser Profile HubUSERWeekly CadenceHEARTBEATWiki Capsules And Agent Doc PacksMetaWiki MaintenanceUSER