jclipa-toolkit

Claude Skills Plugin

Project-agnostic skill set for Claude Code. Skills define behavior; per-project configuration resolves at runtime via a layered search path. Talk to your PM tool, code platform, and docs platform from inside Claude Code with the skills configured to match how each org actually works.

Status: v0.3.2

Quickstart

/plugin marketplace add jclipa-lightit/claude-code-plugin
/plugin install jclipa-toolkit

Optional — enable auto-update so you get new skills and fixes without running /plugin marketplace update manually. Add to your project or user .claude/settings.json:

{
  "extraKnownMarketplaces": {
    "jclipa": {
      "source": { "source": "github", "repo": "jclipa-lightit/claude-code-plugin" },
      "autoUpdate": true
    }
  }
}

With autoUpdate: true, Claude Code refreshes the marketplace at startup and pulls installed plugins to their latest published version. Third-party marketplaces default to false, so this is opt-in.

Then, from inside Claude Code:

/plugin-setup

This walks you through everything: config dir location, user profile, your first org and repo, capability satisfaction per skill. Re-entrant — you can stop and resume anytime. Run /plugin-doctor later to see what's missing.

Two skills you'll use first

Skill	What it does
/plugin-setup	Interactive wizard. Bootstraps your config, asks only what's missing, writes the right files.
/plugin-doctor	Read-only. Shows your config paths, configured orgs/repos, env vars, per-skill capability status. Run this whenever something doesn't work or before reporting a bug.

What's in the box

30 skills, all project-agnostic:

• Git — git-branch, git-commit, git-cr, git-cr-view, git-cr-review, git-cr-comment, git-cr-feedback, git-diff, git-worktree, git-worktree-setup
• Tickets (PM tool) — ticket-create, ticket-comment, ticket-analyze, ticket-analyze-tech, ticket-business-value, ticket-estimate, ticket-subtasks, ticket-sync, ticket-test-plan
• Code — code-landscape, code-plan, code-refactor
• Session — session-checkpoint
• Docs — doc-generate
• Skill lifecycle — skill-manage, skill-export, skill-stats, skill-scan
• Plugin — plugin-setup, plugin-doctor

Each one is project-agnostic: same SKILL.md works against any client/team once the per-org config exists.

Configuration model

Five resolution layers, low → high precedence:

Layer	What it holds	Where
Skill default	Behavior + safe defaults	skills/<bucket>/<skill>/SKILL.md (shipped)
User	Your identity, paths	~/.claude/configs/jclipa-toolkit/user/profile.md
Org	Client / team tools	~/.claude/configs/jclipa-toolkit/orgs/<org>/<skill>.md
Repo	Per-codebase overrides	~/.claude/configs/jclipa-toolkit/repos/<repo>/<skill>.md
Branch	Per-task overrides (rare)	~/.claude/configs/jclipa-toolkit/branches/<branch>/notes.md

Repo > Org > User > Skill default. Branch overrides all when present.

The hook walks the search path on every skill invocation, merges the layers, caches the resolved config per session, and injects only the keys each skill asks for. Zero LLM-side resolution work.

Three transports per skill (mcp / api / cli)

Skills that talk to external tools support three modes:

Transport	When to use
mcp (default)	You already have an MCP server for that tool
api	Direct REST via bin/shared/api_call.py (env-var auth)
cli	gh / glab for GitHub/GitLab — token in OS keychain

Resolution order is configurable per user (transport_order in profile, default [mcp, api, cli]). Per-org override pins a specific transport. The wizard asks you which one and writes the right config.

Capability tiers

Every skill that needs config declares its requirements as capabilities (skills/<skill>/capabilities.json). Three tiers:

• hard — skill aborts if not satisfied (e.g. no PM access)
• soft — falls back to a sensible default + warns once
• no — purely optional

The hook checks capabilities BEFORE invoking a skill. If a hard capability is missing, you get a clear error listing what to set, not a confusing mid-flight failure. See docs/CAPABILITIES.md.

Repo layout