Skill + agent pack

A free pack that teaches Claude Code (and Codex or any agent that reads AGENTS.md) to reuse before it writes: search JustClaw-It before writing non-trivial code, prefer permissive licences, keep attribution, and read the scores.

Install in one command

Claude Code (skill into ~/.claude/skills, agents into ~/.claude/agents):

copycurl -fsSL https://justclaw.it.com/pack.sh | sh

See what it would change first, without writing anything:

copycurl -fsSL https://justclaw.it.com/pack.sh | sh -s -- --dry-run

Codex too (also installs the skill into ~/.codex/skills):

copycurl -fsSL https://justclaw.it.com/pack.sh | sh -s -- --codex

Remove it again (only the files the pack wrote):

copycurl -fsSL https://justclaw.it.com/pack.sh | sh -s -- --uninstall

The script is plain POSIX sh with every file inlined: read it first. It makes no network calls, sends no telemetry and writes only the files below. Re-running it is safe: unchanged files are left alone, and a file that exists but did not come from the pack is skipped unless you add --force.

Prefer a zip? curl -fsSLO https://justclaw.it.com/pack.zip (download), then sh justclaw-it-pack/install.sh or copy the folders by hand.

All of it needs the MCP server connected:

copyclaude mcp add --transport http --scope user justclaw-it https://justclaw.it.com/mcp \
  --header "Authorization: Bearer <YOUR_API_KEY>"

What is in it

The justclaw-it skill

When the agent is about to write auth, webhooks, rate limiting, uploads, API clients, parsers or similar common code, it searches first (search_components, free), picks by fit, then scores.composite, then licence, checks advisories (check_component_security, free), pulls with get_component (1 credit), adapts the code to the project and always keeps the attribution header plus a THIRD_PARTY_LICENSES entry.

The agents

• component-scout finds the best match: runs a few searches, compares candidates on fit, scores, licence and advisories, and returns a shortlist with ids. It does not spend credits.
• licence-auditor audits a repository's dependencies and any code pulled from JustClaw-It: licence compatibility, missing attribution, advisories. Read-only; returns findings with fixes.
• integrator pulls one chosen component and adapts it to the project's naming, module system, types and formatter, keeps the attribution header and adds the NOTICE entry, then builds and tests.

In Claude Code, ask for them by name ("use the component-scout agent to find a rate limiter") or let Claude pick them.

AGENTS.md section (Codex, Cursor, others)

Paste this into your project's AGENTS.md:

copy## Reuse before you write (JustClaw-It)

Before writing non-trivial common code, call `search_components` (free), pick by fit, then
`scores.composite`, then licence (permissive is fine; never `none` or `unknown`), check
`check_component_security` (free) when a result has `security_advisories`, pull with
`get_component` (1 credit), and keep the attribution header plus a THIRD_PARTY_LICENSES entry.

The full section is in the zip as AGENTS.md.

MCP resources

The same files are MCP resources on the server, so an agent can read them without downloading anything: justclaw-it://pack/skills/justclaw-it/SKILL.md, justclaw-it://pack/agents/component-scout.md, justclaw-it://pack/agents/licence-auditor.md, justclaw-it://pack/agents/integrator.md, justclaw-it://pack/AGENTS.md, justclaw-it://pack/README.md and justclaw-it://pack/install.sh. Reading resources is free and works without a key.

See also MCP prompts and the older agent skills that ship with the CLI.