Emdash Skills
A cross-tool product-building system that turns short prompts into quality-gated applications deployed on Cloudflare Workers.
- Source repo
- heymegabyte/claude-skills
- Stars
- ★ 21
- Last updated
- 3d ago
- License
- NOASSERTION
- Primary language
- TypeScript
- FA score
- 53/100 · Major gaps
At a glance
- Works with
- Universal · cross-platformCodex · Claude Code
- You'll need
- Typical use
- A solo founder wants to turn a one-line SaaS concept into a brief, architecture, implementation slices, tests, and a Cloudflare Workers deployment.
- Main limitation
- The defaults are strongly oriented toward Cloudflare Workers, Hono, Drizzle, Angular/Ionic/PrimeNG, and Bun; other clouds, frameworks, and runtimes require rule and scaffold changes.
- Source review
- 53/100 · Major gaps
What does this agent do, and when should you use it?
Emdash Skills is a collection of skills, rules, agent roles, commands, templates, and helper scripts for solo SaaS development rather than a single resident service. The repository describes 19 skill categories, 163 reference documents, 20 agent roles, more than 20 slash commands, and generated rule variants for 32 coding-tool formats. A router selects a small task-specific skill set, after which the workflow separates architecture, parallel implementation, verification, and deployment. Its preferred stack centers on Cloudflare Workers, Hono RPC, Drizzle v1, and D1 or Neon, with documented integrations including Angular, Ionic, PrimeNG, Clerk, Stripe, Inngest, Sentry, and PostHog. Delivery is bounded by Playwright, visual, Lighthouse, axe-core, SEO, console, placeholder, and readability gates before deployment to Cloudflare Workers. It is a strong fit for adopters who want an opinionated, end-to-end product process; teams using different clouds, frameworks, or acceptance standards should expect to curate and adapt its rules.
The workflow begins with a user prompt and a skill router that loads the smallest relevant subset; 01-OS is always included, while a product build commonly proceeds through 02-Brief, 05-Arch, 06-Build, 07-QA, and 08-Deploy. The architecture phase produces a repo map and task graph and identifies parallel seams; build work can then be divided across frontend, backend, content, media, and tests, while verification roles cover deployment, SEO, visual quality, and accessibility. The repository supplies structured artifacts such as acceptance-criteria.md, adr-template.md, product-intent.yaml, qa-report.json, and task-graph.json, plus scripts including bin/emdash-quick-lint, bin/emdash-secret-scan, bin/emdash-cost-check, scripts/validate-skills.sh, and scripts/visual-tdd-loop.sh. Commands include /saas, /forge-from-openapi, /run-evals, /audit-doctrine, and /audit-mcp-fleet; bin/forge-skill-from-openapi.mjs generates a SKILL.md and command scaffold from an OpenAPI specification. Separate categories cover DOCX and XLSX read/write operations, PDF parsing and generation, PPTX generation, HTTP MCP servers on Workers, stdio MCP templates, OpenAPI-to-MCP forging, and registry publishing. The documented release path checks Playwright flows at six breakpoints, visual scoring, Lighthouse, axe-core, SEO, browser-console output, placeholder content, and Flesch readability before deploying to Cloudflare Workers.
- A solo founder wants to turn a one-line SaaS concept into a brief, architecture, implementation slices, tests, and a Cloudflare Workers deployment.
- A maintainer needs to add authentication, billing, webhooks, analytics, or another feature to an existing web application using a test-first implementation sequence.
- A platform team wants to distribute a common engineering doctrine across Cursor, Windsurf, Copilot, Codex, Claude Code, Aider, Cline, and other coding environments.
- A backend developer wants to generate an MCP skill and command scaffold from OpenAPI or create an HTTP MCP server on Workers.
- A web team is rebuilding a marketing site, portfolio, or content property and needs consistent SEO, visual, accessibility, and multi-breakpoint acceptance checks.
- A mixed technical and business organization wants one framework covering C-suite, finance, compliance, product-management, and DOCX, XLSX, PDF, or PPTX workflows.
How do you install or deploy this agent?
The recommended GitHub Skills command is gh skill install heymegabyte/claude-skills. For Claude Code, run claude plugin install heymegabyte/claude-skills; for npm, run npm i @heymegabyte/claude-skills; for JSR, run npx jsr add @heymegabyte/claude-skills. The documented Codex path is git clone https://github.com/heymegabyte/claude-skills ~/.codex/skills, while a generic manual installation clones the repository into ~/.agentskills. These are alternative installation methods, not cumulative requirements. Deploying the default product stack also requires network access, a writable project filesystem, and a Cloudflare Workers environment. Credentials are feature-dependent: documented examples check STRIPE_*, CLERK_*, INNGEST_*, and CLOUDFLARE_API_KEY, while map-based local-business projects additionally require GOOGLE_MAPS_API_KEY and GOOGLE_PLACES_API_KEY.
How do you use this agent?
After installation, issue a product or engineering request inside a supported coding tool. A documented first invocation is Build acmebilling.dev — recurring subscription billing for indie SaaS founders. Stripe-native, Clerk auth, Inngest for retry workflows.; it routes through skills 01, 02, 03, 05, 06, 07, 08, and 13 and checks the Stripe, Clerk, Inngest, and Cloudflare credentials before scaffolding. For an existing project, use a request such as Add a magic-link auth flow to acme.dev — passwordless email via Resend, Clerk session, redirect to /dashboard., which invokes architecture, build-slice, and Playwright E2E work. If only coding guidance is needed, use the generated rule file for the target platform; for a complete build, allow the router to select skills and proceed through every hard gate before Workers deployment. The supplied material does not define one universal configuration file or credential list across all platforms, so keys must be prepared for the selected skills and integrations.
What are this agent's strengths and limitations?
- Its 19 skill categories and 163 reference documents span briefs, architecture, implementation, QA, deployment, growth, document processing, and MCP authoring.
- It generates native rule variants for 32 tool formats and provides explicit installation paths for Codex and Claude Code, enabling reuse across coding environments.
- Acceptance gates are measurable: zero Playwright failures at six breakpoints, Lighthouse accessibility of at least 95, zero axe-core violations, zero console errors, and zero placeholders.
- It includes auditable operational assets such as task graphs, JSON QA reports, Semgrep rules, secret scanning, cost checks, and a visual TDD loop.
- The OpenAPI forge can generate MCP skill scaffolding, with both Workers-based HTTP MCP and stdio server patterns documented.
- The defaults are strongly oriented toward Cloudflare Workers, Hono, Drizzle, Angular/Ionic/PrimeNG, and Bun; other clouds, frameworks, and runtimes require rule and scaffold changes.
- A full deployment can depend on numerous external services and credentials, including Cloudflare, Clerk, Stripe, Resend, Inngest, Sentry, PostHog, and Google Maps, increasing configuration, cost, and vendor-management overhead.
- Mandatory multi-breakpoint testing, visual review, accessibility, SEO, and readability gates add execution time and tooling overhead even for relatively small changes.
- Repository metadata reports NOASSERTION for the license, while the README badge and license section identify the Rutgers License; adopters should inspect the LICENSE terms directly.
- The supplied material describes broad automation and thresholds but does not provide a measured platform-compatibility matrix, benchmark results, or cost data for every deployment scenario.
How does this agent compare with similar options?
Compared with using a single coding-rule file, Emdash Skills adds skill routing, specialized agent roles, slash commands, templates, verification scripts, and deployment gates. It provides both modern native formats and some legacy single-file formats for tools such as Cursor, Windsurf, Copilot, and Augment; modern variants use platform-specific frontmatter, while legacy variants are primarily plain instruction files. The repository does not claim that all environments execute identical capabilities: the variants carry the rules, while the opinionated product workflow and default deployment path remain centered on its skill graph and Cloudflare stack.
Key facts side by side with the most closely related agents.
| Agent | Source review | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|
| Emdash Skills This agent | 53 · Major gaps | ★ 21 | 3d ago | TypeScript | Codex · Claude Code |
| Claude Forge | 48 · Major gaps | ★ 839 | 20d ago | Shell | Claude Code |
| Hercules Testing Agent | 48 · Major gaps | ★ 1.2k | 1mo ago | Python | — |
| OpenConnector | 57 · Major gaps | ★ 5.8k | 4d ago | TypeScript | — |
How does FollowAgents rate this agent?
Why each dimension lost points
The evidence shows smallest-subset routing, narrowly scoped GitHub Actions permissions, secret scanning, required-key checks, a security-reporting channel, and SHA-pinned Actions, giving least privilege, sensitive-data handling, and dependency security a meaningful foundation. Deductions apply because the product advertises autonomous deployment, email, payments, pipeline retriggering, issue mutation, and automatic implementation of high-confidence ideas without a consistent user-confirmation boundary. It also does not explain which data each third party receives or retains. Rollback is mostly asserted by agent purpose rather than documented as a recovery procedure. Author, copyright, email, and repository attribution are clear, although publisher identity remains self-asserted and unverified by the enterprise registry.
JSON diagnostics, check scripts, tracking issues, and a stated security-response window provide reasonably concrete failure reporting. Major deductions follow from internal drift: the package description says 14 categories while the README says 19; the README claims 20 agents but lists fewer; and package file patterns primarily cover categories 01–14 despite claims for 15–19. Numerous external services and versions are named, but no lockfile, installation evidence, or availability fallback is supplied. Some workflows deliberately convert check failures into issues instead of failing CI.
Audience and scenario coverage is exceptionally broad and concrete, spanning SaaS, APIs, marketing, existing projects, local businesses, nonprofits, and content sites, which justifies full marks there. Routing tables and ideal prompts provide useful trigger specificity, while 32 described variants support environment fit. Deductions apply because one-line or domain-only prompts permit extensive inference without clearly identifying unsafe-to-infer decisions or mandatory human choices, and the supplied evidence does not show each claimed platform variant.
The README has strong organization across installation, categories, agents, gates, stack, templates, tools, and routing, with numerous end-to-end examples. Deductions apply for the lack of a consolidated FAQ and incomplete prerequisite or credential-setup notes; naming and counts drift across 14 versus 19 categories, the agent count, and package inclusion patterns. Known limitations are sparse. The custom Rutgers License contains subjective good-purpose language and fanciful obligations, leaving nonstandard legal boundaries while metadata is NOASSERTION. Version numbers, What's New, supported security versions, and maintenance workflows establish an update path, but there is no complete changelog. Maintainer and contact details are present, though independently unverified.
Templates, structured QA artifacts, explicit deliverables, routing, and deployment gates make the intended outputs reasonably usable, and an integrated product-lifecycle playbook offers plausible marginal value. Deductions apply because the supplied static files do not demonstrate completed products or measured outcomes. The approach also depends on multiple premium model tiers, parallel agents, commercial APIs, and external platforms; cost estimation and free-tier warnings are mentioned, but no overall cost, latency, or operational-burden comparison is provided.
Many claims point to named scripts, templates, paths, thresholds, versions, and workflows, while weekly URL and pricing checks provide auditable maintenance mechanisms. Deductions apply because the supplied evidence omits the referenced scripts and skill bodies, preventing verification of most behavioral claims, and README claims conflict with package metadata and packaging scope. Marketing statements such as one-line prompts producing deployed products, no-exception gates, and a complete platform are not cleanly separated from demonstrated facts; future-dated and tool-replacement claims also lack corroboration within the supplied files.
- Add per-action confirmation and scope controls before allowing deployment, email, payment handling, production retriggers, or external issue mutation.
- Verify that the published package actually contains claimed categories 15–19 and every advertised agent; the current counts and package patterns conflict.
- Do not treat the custom Rutgers License as a standard open-source license; obtain legal review before adoption or redistribution.
- Document the data sent to Clerk, Stripe, PostHog, Sentry, GA4, Resend, OpenAI, and other services, including residency, retention, and deletion behavior.
- Inspect installation hooks, secret-discovery scripts, and automatic implementation features to ensure they neither access out-of-scope credentials nor create unapproved external effects.
FAQ
Must every project deploy to Cloudflare Workers?
Which API keys are required before starting?
STRIPE_*, CLERK_*, INNGEST_*, and CLOUDFLARE_API_KEY; the local-business map example also requires Google Maps and Places keys.