SuperAPI AI Directory
Partner with us中文AI newsFavorites
All toolsAI agentsclaude-skills

claude-skills

14-category autonomous product-building OS bundling 18 specialized agents (architect, test-writer, deploy-verifier, security-reviewer, visual-qa, performance-profiler, accessibility-auditor, content-writer, completeness-checker, dependency-auditor, migration-agent, incident-responder, computer-use-o

Open website

About this tool

Purpose

Agent Skills is an "agent-neutral" autonomous product-building operating system, aimed at 32+ AI coding tools. Its goal is to turn a one-sentence prompt (or even just a domain name) into a product deployed on Cloudflare Workers and passing all quality gates. The repository is positioned as a "complete solo SaaS founder platform."

Main capabilities

  • Skill and agent system: Includes 23 skill categories, 28 agents, 149 reference documents, and 37 platform variants. Agents cover architects, test writers, deployment validators, security reviewers, visual QA, performance profilers, accessibility auditors, content writers, integrity checkers, dependency auditors, migration agents, incident responders, computer-use, and more.
  • Persistent AI fleet: GitHub schedules project loops on persistent Ubuntu hosts; OpenClaw routes official Codex/Claude configurations through Claw Router, and OpenCode uses DeepSeek directly. The control UI is at heymegabyte/agent.megabyte.space, and website publishing is currently excluded.
  • Skills 18-19 and the integration arc (2026-06): Skill 18 document processing (DOCX/XLSX reading and writing, PDF generation and parsing, PPTX generation); Skill 19 MCP authoring (forging HTTP MCP servers on Workers, stdio MCP templates, OpenAPI-to-MCP forger, one-click registry publishing).
  • 20+ slash commands: /saas, /forge-from-openapi, /forge-webhook-handler, /generate-prp, /execute-prp, /run-evals, /audit-doctrine, /audit-mcp-fleet, /dashboard-cockpit, /drift-check, etc.
  • Production forge script: bin/forge-skill-from-openapi.mjs can generate a complete SKILL.md and command scaffolding from any OpenAPI specification.
  • LLM evaluation framework: /run-evals + structured scoring criteria + schema validation results + regression tracking.
  • 15 new principles: state-is-the-enemy, fail-fast-build-fail-soft-prod, ttfr-north-star, sync-ui-async-backing, cost-per-request-accountability, inverted-abstraction-pyramid, one-way-two-way-doors, vendor-risk-tiering, data-residency-by-default, right-to-deletion, refund-automation, documentation-as-code, production-observability-default-on, hardware-aware-programming, webhook-receiver-architecture.
  • Other capabilities: DO agent scaffolding (using Durable Objects as stateful AI agents through Skill 19, connected to the MCP registry); RAG pipelines (Vectorize + D1 hybrid search, Skill 05); multi-tenant subdomain configuration (Cloudflare custom hostnames + per-tenant D1 namespaces, Skill 05); 5 new PreToolUse hooks (enforcing permission discipline before destructive tool calls); MCP server registry (/audit-mcp-fleet catalogs, health-checks, and version-pins each MCP).

How to use

  • Cross-platform support: Each push to master automatically generates 32 variants, with each format using its tool's native frontmatter.
  • Routing mechanism: The router loads the smallest useful subset by task, never all 149 documents; see _router.md for the complete routing table.
  • Prompt mode: Can be copied and pasted as-is, with each prompt routed through the skill graph to a deployed product that passes the gates. The shorter the prompt, the more reasoning; a single domain name is enough.
  • Typical workflow examples:
    • Build a SaaS from a domain name: Skill 01->02->03->05->06->07->08→13; agent architect(Opus)→5 parallel build agents→deploy-verifier + seo-auditor + visual-qa; the gate checks STRIPE_, CLERK_, INNGEST_*, CLOUDFLARE_API_KEY before scaffolding; the output is a live Worker on acmebilling.dev, including checkout, dashboard, webhook handler, Sentry + PostHog + GA4.
    • Rebuild an existing site: skills 01→02→09 (brand extraction)→15 (full corpus + media video extraction + grammar audit)→07→08; the gate requires every original URL to return 200 or 301, logo preserved, theme match, favicon set via the real favicongenerator.
    • One-sentence portfolio generation: skills 01→02 (founder inference)→09→10→11→12→14→07→08; the output is a dark-first single page, including hero + work + writing + contact.
    • Local business site (NAP + reviews + map): skills 01→02→09→10→13; the gate requires GOOGLE_MAPS_API_KEY + GOOGLE_PLACES_API_KEY; the output includes a full-width Maps embed, NAP schema.org LocalBusiness, OpenTable/Booksy embed, Twilio SMS booking.
    • Nonprofit donation flow: skills 01→02→09→13; the gate requires STRIPE_* keys; the output includes a navbar Donate CTA, recurring and one-time donations, Resend tax receipts, an impact counter with IO+rAF scroll-in.
    • Pre-build brainstorming: skill 14 (creative engine)→bounded web research→self-critique filtering; output 3 evidence-backed proposals, with an apa_citation attached to each claim, including virality coefficient and AI search visibility score, automatically implementing aligned ideas with confidence ≥ 0.85.
    • Add a feature to an existing project: skills 05→06→07; agent test-writer (write failing tests first)→implementation→deploy-verifier.
    • Debug a stuck pipeline: skills 08→07; load failed-pipeline-protocol.md (5 typical failure modes); the flow is detection via D1 query→diagnose root cause→fix→isolated verification→mint session→retrigger via direct worker URL→background monitoring.
    • Large-scale content-driven blog: skills 09 (pSEO 5 types + GEO + fact acquisition via APA citations)→06→07; output 50 unique routes, no templated copy, every quantitative claim cites an APA 7th edition source, Schema.org Article with citation:CreativeWork[].

Applicable Scenarios

Suitable for solo SaaS founders, developers who need to quickly build and deploy a product from a one-sentence prompt, and teams that want unified skill and agent orchestration for 32+ AI programming tools. Typical scenarios include SaaS building, site rebuilding, portfolio generation, local business sites, nonprofit donation flows, feature iteration, pipeline debugging, and large-scale content blogs.

Limitations

  • Website publishing is currently excluded.
  • All deployments must pass all gates, with no exceptions and no overrides.
  • Some workflows depend on specific API keys (such as STRIPE_, CLERK_, INNGEST_*, CLOUDFLARE_API_KEY, GOOGLE_MAPS_API_KEY, GOOGLE_PLACES_API_KEY).
  • The canonical tech stack table and version pinning are in CONVENTIONS.md § Stack.
  • The project uses The Rutgers License, by Brian Zalewski / Megabyte Labs.
  • More information on fees, pricing, platform support details, etc. is not currently provided in the materials.

Related tools

Browse category