Add tokrepo-discovery skill — agent-native registry discovery for Codex

[henu-wang]

What this PR adds

A new Codex skill tokrepo-discovery/ (under Meta & Utilities) that teaches Codex to call TokRepo's MCP tokrepo_discover at the plan→implementation boundary, so common reusable artifacts (skills, prompts, MCP configs, scripts) get pulled from https://tokrepo.com instead of re-generated per project.

Skill folder layout

tokrepo-discovery/
├── SKILL.md          # name + description + 4 MANDATORY rules + setup
├── scripts/
│   └── install.sh    # one-line bootstrap: `npx tokrepo init-agent --target codex`
└── references/
    └── tools.md      # full 12-tool MCP catalog (loaded only on request)

Why this fits Meta & Utilities

Pattern matches existing skill-installer/ and template-skill/ — meta utilities that help Codex discover and install other skills. TokRepo is broader (covers Cursor/Cline/Copilot/Windsurf/Roo/OpenHands/Aider/Gemini/Claude in addition to Codex), but for this list the skill is scoped to its Codex behavior: when Codex sees a reusable-artifact task, it calls tokrepo_discover first.

What TokRepo is

• MCP server (tokrepo-mcp-server on npm; hosted at https://tokrepo.com/mcp) — 12 tools for discover / verify / install_plan / install / lifecycle / handoff / push.
• Public agent funnel at https://tokrepo.com/agent-stats — anonymous aggregate of real agent events on a 14-day window (currently mcp_discover=83, install_apply=53, push=83 on a sample 14-day window).
• Official MCP Registry entry: io.github.henu-wang/tokrepo-mcp-server (v2.12.1).

Happy to move the entry to a different category, shorten the SKILL.md, or split scripts/references if the maintainer prefers a tighter scope.