Agent Skills

Skills are Markdown files that give an agent platform-specific instructions. For Claude Code, they live at ~/.claude/skills/<skill-name>/SKILL.md. You install a skill once; from then on, you reference it in a prompt and the agent follows it.

Earl ships five skills. Each one handles a distinct task — setup, template authoring, codebase migration, security hardening, and diagnosis.

Installing a skill

mkdir -p ~/.claude/skills/<skill-name>
curl -fsSL https://raw.githubusercontent.com/brwse/earl/main/skills/<skill-name>/SKILL.md \
  -o ~/.claude/skills/<skill-name>/SKILL.md

Replace <skill-name> with the skill you want. Then in a new agent session:

Use the <skill-name> skill to <goal>.

If you'd rather not install the file, paste this into your agent:

Fetch https://raw.githubusercontent.com/brwse/earl/main/skills/<skill-name>/SKILL.md
and follow the skill to <goal>.

The agent fetches and runs it inline. Nothing is written to disk.

Available skills

setup-earl

Installs Earl, writes MCP config for every detected agent platform, and sets up context breadcrumbs. The full walkthrough is at Agent-Assisted Setup.

Use the setup-earl skill to install and configure Earl for this project.

create-template

Guides the agent through writing a new Earl template for an API you specify. The agent works through five steps:

1. Identify which API endpoints to wrap
2. Determine the auth method (bearer, api_key, basic, or oauth)
3. Write the HCL template with correct parameter types
4. Test with earl call --yes --json
5. Store any required secrets in the keychain

The result is a committed .hcl file ready to use. The agent does not guess at auth — it asks you for the token or walks you through earl secrets set.

Use the create-template skill to build an Earl template for the Linear API.

migrate-to-earl

Points the agent at a codebase and converts existing API calls to Earl templates. The agent:

1. Scans for curl, gh, wget, and raw HTTP calls
2. Groups them by provider
3. Generates Earl templates for each provider
4. Rewrites the call sites to use earl call --yes --json equivalents
5. Writes a migration summary listing what changed and what was skipped

Useful when you have a repo full of inline curl commands and want them all going through Earl's security layer. The agent leaves the original calls commented out rather than deleted, so you can review the diff before committing.

Use the migrate-to-earl skill to convert all curl and gh calls in this repo to Earl templates.

secure-agent

Prevents the agent platform from bypassing Earl with raw CLI tools. It writes rules to CLAUDE.md, .cursorrules, or .windsurfrules (whichever applies) instructing the agent not to use curl, gh, wget, or raw HTTP directly — everything should go through earl call. If the platform supports allow/deny lists, the skill configures those too.

Run this after setup-earl when you want to actually enforce Earl as the only network path:

Use the secure-agent skill to lock down this project so the agent only uses Earl for API calls.

troubleshoot-earl

Diagnoses Earl problems step by step. The agent runs earl doctor, checks keychain access, validates template files, inspects MCP config, and suggests fixes for whatever it finds. Start here when something isn't working and you don't know where to look.

Use the troubleshoot-earl skill to figure out why earl call is failing.

The skill repository

All skills are at github.com/brwse/earl/tree/main/skills/. Third-party skills for specific integrations — memory systems, recall tools, specialized workflows — are under skills/3p/. The third-party skills aren't maintained by Earl but follow the same format and can be installed the same way.