my-skills

my-claude-plugins

Personal Claude Code plugin marketplace.

Skills

The main bundled skills are shipped under a single plugin (my-skills).

Skill	Description
team-lead-orchestration	Runtime-neutral principles for the team-lead role: lead/lane split, bounded execution lanes, and acceptance gates
opencode-companion	OpenCode serve/session/job/review runtime via direct companion scripts
daytona-companion	Daytona sandbox lifecycle, global project-scoped state, and artifact workflows
paseo-companion	Paseo CLI runtime: agents, terminals, schedules, worktrees, host/port targeting
google-workspace	Google Docs/Drive/Sheets via the gws CLI
mcp-skill	On-demand MCP server invocation via MCPorter

Optional standalone plugins

Some skills are also available as separate opt-in plugins and are not included in the my-skills bundle.

Plugin	Skill	Description
skill-creator	skill-creator	Vendored fork for creating, evaluating, packaging, and improving Claude Code skills with configurable CLI runners

Upgrade notes

The my-skills bundle now keeps runtime adapters and generic orchestration only. Legacy public skills that mixed workflow policy with local implementation details were removed from the bundle and source tree:

• opencode-orchestrator: generic orchestration guidance moved to team-lead-orchestration; OpenCode runtime commands remain in opencode-companion.
• task-iteration: plan execution should be handled by the active agent workflow using team-lead-orchestration plus the selected runtime adapter.
• agentic-orchestration: renamed to team-lead-orchestration — same skill, clearer trigger surface around the team-lead framing.
• debug-workflow: browser, documentation, repository, and web-search MCP access should go through mcp-skill when needed; runtime-specific execution stays in the companion skills.

Companion skills expose direct script entrypoints:

node "${CLAUDE_PLUGIN_ROOT}/skills/opencode-companion/scripts/opencode-companion.mjs" session new --directory "$WORK_DIR" -- "<prompt>"
node "${CLAUDE_PLUGIN_ROOT}/skills/daytona-companion/scripts/daytona-manager.mjs" status --directory "$WORK_DIR"

Installation

Local (recommended for development)

{
  "extraKnownMarketplaces": {
    "my-claude-plugins": {
      "source": {
        "source": "directory",
        "path": "/path/to/my-claude-plugins"
      }
    }
  }
}

GitHub

{
  "extraKnownMarketplaces": {
    "my-claude-plugins": {
      "source": {
        "source": "github",
        "repo": "atomlink-ye/my-claude-plugins"
      }
    }
  }
}

Then enable the bundled plugin:

{
  "enabledPlugins": {
    "my-skills@my-claude-plugins": true
  }
}

To enable the standalone optional skill creator plugin instead of (or in addition to) the bundled plugin:

{
  "enabledPlugins": {
    "skill-creator@my-claude-plugins": true
  }
}

Development

pnpm install
pnpm test

Evaluating skills

Per-skill triggers, outcomes, smoke and unit tests live under eval/<skill>/. The repo follows the skill-creator eval schema.

# Trigger eval for one skill (default runner: `claude`)
eval/_shared/run-trigger-eval.sh opencode-companion

# Smoke tests (offline-safe)
eval/opencode-companion/tests/smoke.sh
eval/daytona-companion/tests/smoke.sh
eval/paseo-companion/tests/smoke.sh

If you want eval runs to use a different Anthropic-compatible endpoint (e.g. an overflow-quota wrapper), point EVAL_RUNNER at your own wrapper executable. See eval/_shared/README.md for the layout, the runner contract, and the coverage plan across skills.