Capgo Capacitor Agent Skills
Give Claude Code, Codex, Cursor and other AI coding agents ready-made skills for building, shipping and maintaining Capacitor apps.
- Source repo
- Cap-go/capgo-skills
- Stars
- ★ 74
- Last updated
- 12d ago
- License
- MIT
- Primary language
- JavaScript
- FA score
- 43/100 · Major gaps
At a glance
- How it runs
- Works with
- Portable with changesCodex · Claude CodeChatGPT · Claude.ai (Partial support)
- Cost
- Free tier plus a paid hosted plan
- Setup effort
- Low · running in minutes
- You'll need
- Typical use
- Developers building a Capacitor app in Claude Code or Codex who want the assistant to follow official best practices
- Not a fit if
- Teams not building Capacitor apps — most skills won't apply
- Environments that must stay fully offline without MCP or cloud services
- Developers working in non-Capacitor mobile stacks expecting ready-made guidance
- Source review
- 43/100 · Major gaps
What does this agent do, and when should you use it?
capgo-skills is a MIT-licensed skill library maintained by Capgo (formerly @capgo/capacitor-skills). Each skill lives in a skills/ folder as SKILL.md plus metadata.json, and together they cover the Capacitor lifecycle: building apps, plugin work, migrations, version upgrades, debugging, testing, security, store publishing and OTA live updates. The collection installs through npx skills add, and is also published as a Claude Code plugin marketplace, a Gemini CLI extension, and Cursor/Codex plugin marketplaces, grouped into 11 focused plugins such as capgo-cloud, capacitor-core and capacitor-quality. Skills can be paired with the hosted Capgo MCP server or a local MCP instance so the agent can operate apps, channels, bundles and hosted builds. It is a content repository rather than runtime code: value comes from how well the host agent follows the packaged instructions.
The repository ships prompts, not executables. Each skill directory contains SKILL.md (agent instructions) and metadata.json (metadata); agents load the matching skill when they detect a relevant task. On the install side, npx skills add Cap-go/capgo-skills drops skills into the agent's skill directory, claude plugin marketplace add Cap-go/capgo-skills followed by claude plugin install capgo-cloud@capgo-skills installs a focused plugin group, and gemini extensions install https://github.com/Cap-go/capgo-skills handles Gemini CLI. At runtime the skills guide the agent to call Capgo CLI commands such as npx @capgo/cli@latest init, login and bundle upload, run Capsec scans with npx @capgo/capgo-sec@latest scan, and use the Capgo MCP server (hosted at https://api.capgo.app/mcp or local via npx @capgo/cli@latest mcp) to manage apps, channels, bundles and native builds. For maintainers, bun run sync-skills and bun run lint-skills enforce byte-for-byte parity between plugins/ copies and skills/, and ENABLE_SKILLGRADE=1 bun run lint-skills-skillgrade runs skill evals.
- Developers building a Capacitor app in Claude Code or Codex who want the assistant to follow official best practices
- Teams converting an existing web app, PWA or Next.js/React project into a store-ready Capacitor app
- Projects migrating from Cordova, Ionic Appflow, Capawesome live updates or the Ionic Enterprise SDK
- Claude Code users who want to install only selected plugin groups, such as capacitor-deployment or capacitor-app-upgrades
- Plugin authors adding Swift Package Manager support or moving a Capacitor iOS app off CocoaPods
- Operators who use Capgo hosted builds and OTA updates and want an agent to manage channels and bundles
How do you install or deploy this agent?
Install every skill through skills.sh:
npx skills add Cap-go/capgo-skillsClaude Code users can add the repo as a plugin marketplace and install groups:
claude plugin marketplace add Cap-go/capgo-skills
claude plugin install capgo-cloud@capgo-skillsGemini CLI users:
gemini extensions install https://github.com/Cap-go/capgo-skillsCursor and Codex read the bundled .cursor-plugin/marketplace.json and .agents/plugins/marketplace.json marketplaces. To let the agent act on Capgo cloud resources, also connect the Capgo MCP server: the hosted endpoint is https://api.capgo.app/mcp, or run it locally:
npx @capgo/cli@latest mcpThe README does not document the exact marketplace UI steps for each tool, nor any runtime prerequisites beyond Node/npx.
How do you use this agent?
Once installed, skills fire automatically when the agent recognizes a matching task — saying "Add push notifications" loads capacitor-push-notifications, "Set up GitHub Actions" loads capacitor-ci-cd, and "Upgrade a Capacitor app from v7 to v8" loads capacitor-app-upgrade-v7-to-v8. For Capgo cloud workflows, log in and initialize the project first:
npm install -g @capgo/cli
npx @capgo/cli@latest login
npx @capgo/cli@latest initTo add or edit skills, create a folder under skills/ with SKILL.md and metadata.json, then sync and validate the plugin copies:
bun run sync-skills
bun run lint-skillsRunning the skillgrade eval on the authoring workflow requires an API key:
ENABLE_SKILLGRADE=1 bun run lint-skills-skillgradeWhat are this agent's strengths and limitations?
- Covers the full Capacitor lifecycle — Build, Plugins, Migrate, Upgrade, Debug, Test, Security, Publish, Live updates — which is broader than most skill packs
- One skill set ships to skills.sh, Claude Code, Gemini CLI, Cursor and Codex marketplaces, so the same content works across several coding agents
- sync-skills and lint-skills enforce byte-for-byte parity between plugin copies and canonical skills, reducing distribution drift
- Versioning tracks Capacitor majors (plugin v8 for Capacitor 8) with explicit upgrade skills for each step
- MIT licensed with plain SKILL.md files you can read, fork and edit without a black box
- It is a prompt and metadata collection; effectiveness depends entirely on the host agent, and no benchmark results are published
- Cloud builds, live updates and organization management skills are tightly coupled to the Capgo platform, so non-Capgo users get only part of the value
- Only the latest major version is actively maintained (v6 and older are marked unmaintained), leaving older Capacitor projects to verify compatibility themselves
- Contributing requires Bun for the sync and lint scripts
- The skillgrade eval needs an API key, so external users cannot easily reproduce the author's quality checks
How does this agent compare with similar options?
The README notes the repo was formerly @capgo/capacitor-skills and its migration skills explicitly address moving off Cordova, Ionic Appflow, Capawesome live updates, the Ionic Enterprise SDK and CocoaPods. It does not benchmark itself against other AI agent skill packs.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Capgo Capacitor Agent Skills This agent | 43 · Major gaps | Agent plugin / skillFreemium | ★ 74 | 12d ago | JavaScript | Codex · Claude Code |
| Elephant-Goldfish Model (EGM) | 75 · Good | Agent plugin / skillFree + model costs | ★ 5.7k | 1d ago | Python | Codex · Claude Code |
| Expo Skills – Official AI Agent Skills for Expo | 49 · Major gaps | Agent plugin / skillFreemium | ★ 3.8k | 1d ago | Shell | Codex · Claude Code |
| Modern Go Guidelines | 56 · Major gaps | Agent plugin / skillFree | ★ 3.9k | 1mo ago | Go | Codex · Claude Code |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence is limited to README, LICENSE, package.json and one lint workflow; no SKILL.md bodies are included, so actual permissions granted, user confirmation prompts, data flows and sensitive-data handling cannot be verified. README mentions Capgo MCP and CLI login but not credential storage or least privilege, so those score 1. source_attribution scores 2 because README names Capgo as maintainer, documents the rename and LICENSE carries MIT copyright.
The skill list in README broadly matches the package.json skills array and CI runs lint-skills, so self_consistency scores 2; however only yaml is declared as a dependency while scripts rely on bun and skillgrade without pinned versions, so dependency_availability scores 1; no failure messages or error handling are documented, scoring 1.
README targets Claude Code, Codex, Cursor, Copilot and Gemini CLI with lifecycle scenarios and explicit trigger phrases, so audience_and_scenarios and trigger_precision score 2, and environment_fit scores 2 for multi-tool install paths; capability boundaries inside skills cannot be verified from the available files, scoring 1.
Information architecture is clear (lifecycle table, plugin groups, skill catalog), install notes cover multiple tools, naming_stability scores 2 for documented rename redirects, and license scores 2; but FAQ, known limitations and changelog are absent and maintenance responsibility is only a generic Capgo mention, so those score 1.
README shows a catalog and usage triggers but no skill bodies or output samples, so output usability and marginal value cannot be judged beyond the catalog, scoring 1; cost_benefit scores 1 because reliance on Capgo cloud and MCP adds setup cost that cannot be weighed from the files.
Most README claims (63+ rules, 55 skills, compatibility table) lack supporting source files; only the skill count is partly corroborated by package.json, so all three score 1. No fabricated tests or provenance were found, but fact/inference separation is not explicitly documented.
- This review covers only README, LICENSE, package.json and one CI workflow; no SKILL.md bodies were provided, so actual skill instructions, permissions and security behavior are unverified.
- README directs users to log in to the Capgo CLI and connect an MCP server without documenting credential storage, least privilege or data egress scope; review before use.
- Dependencies and scripts (bun, skillgrade, @capgo/cli) are not version-pinned, so supply-chain risk must be assessed independently.
- Publisher identity is not verified by the FollowAgents curated registry; treat as unknown rather than suspicious, and do not infer trust from it.