Add TogetherLink usage and configure skills

[jkail-together]

Summary

Adds two skills for TogetherLink, Together AI's CLI for running Claude Code, Codex CLI, Claude Desktop/Cowork, and ChatGPT Desktop on Together AI models.

• together-togetherlink: using TogetherLink.
  • Launching sessions (togetherlink claude, tclaude, tcodex, desktop apps).
  • Pinning a model with --main. It must come before the tool name; --model after it is refused.
  • The auto router and the Claude tier mapping (Opus→Kimi K3, Fable→GLM 5.3, Sonnet→GLM 5.3 Flash, Haiku→DeepSeek V4.1 Flash).
  • Headless runs (< /dev/null), spend via togetherlink usage, and togetherlink image.
  • Troubleshooting. References: cli-reference.md, models.md.
• together-togetherlink-configure: installing and configuring it.
  • Install script, togetherlink configure, TOGETHER_API_KEY / --api-key, and the optional Anthropic key.
  • Claude Desktop and ChatGPT Desktop on / off / reset.
  • togetherlink update, and restoring the normal setup.

Each skill has agents/openai.yaml, a trigger-eval set with near-miss negatives, and a verbatim hand-off sentence to the other. They also hand off to together-chat-completions / together-images for building on the API directly. AGENTS.md and the README table were regenerated with ./scripts/publish.sh.

Content notes

• Sources: the docs page and the TogetherLink v0.9.53 CLI (--help, models, image --help).
• Live-verified: the headless and --main behaviour, --model being refused, and settings.json model not applying under TogetherLink.
• Cost caveat: Claude Code's own total_cost_usd in JSON output uses Anthropic prices; togetherlink usage is the billed figure. This is why the skill steers agents to togetherlink usage.
• Model table: no prices; it points agents to togetherlink models for live values, since TogetherLink is beta.
• Platform: the desktop integrations are documented as macOS (beta), following the docs page. The v0.9.53 CLI --help says "macOS/Linux". Happy to change it if Linux is supported.

Verification

• python3 scripts/quick_validate.py skills/together-*: 16 skills, 0 errors
• ./scripts/publish.sh --check: all artifacts up to date
• quality_check.py: no warnings for the new skills. The existing missing eval for together-dedicated-model-inference fails on main too.
• Routing smoke test: a headless Claude Code session was asked "how do I pin GLM 5.3 … and run it headless", with the plugin installed. It loaded together-togetherlink and answered with --main + < /dev/null.

Notes

• unify skills into one #74 ("unify skills into one") would restructure skills/. If it lands first, I'll fold these into the new layout.
• Follow-up stacked PR adds a togetherlink marketplace plugin that ships only these two skills.

🤖 Generated with Claude Code

[broly-code-security-scanner]

Broly Security Scan

Note

✅ Clean scan
No vulnerabilities detected in this PR.

Note

Re-scan this PR anytime with /broly scan — useful after /broly undismiss, or to refresh findings without a new push.

Broly — SAST (DeepSeek-V4.1-Flash) · Secrets · SCA · IaC · GH Actions · Base Images · Supply Chain Threats · Attack Hypotheses · Adversarial Verification

We're continuously improving Broly's accuracy and finding quality — your feedback is valuable. False positives, missed findings, bugs, and feature requests all welcome.