Keep OpenRouter model configs in sync with the OpenClaw leaderboard
A skill that extracts model IDs from the OpenRouter app page, verifies them against the API catalog, and adds missing entries to your local OpenClaw config with correct specs and aliases.
- LinkLoot access
- Free
- Provider costs
- Unknown
What you get from it
> AI-generated: This Loot was created and published automatically by LinkLoot and was not substantively reviewed by a human editor. ## What it does
The add-top-openrouter-models skill automates a tedious maintenance task: keeping your local OpenClaw installation's model configuration aligned with the models actually used by the OpenClaw app on OpenRouter. It fetches the app leaderboard, extracts model IDs matching the // pattern, filters out navigation links, then runs a Python sync script that validates each ID against the OpenRouter /api/v1/models catalog before writing anything.
Verified models get converted into OpenClaw format — context window, pricing, reasoning flags, modalities — and added to ~/.openclaw/agents//agent/models.json and ~/.openclaw/openclaw.json. Timestamped backups are created before every write. Aliases come from a maintained references/aliases.json file or auto-derived defaults.
Who should use it
OpenClaw operators who want their agent configs to reflect current OpenRouter model availability without manually editing JSON files. Particularly useful when new models appear on the leaderboard and you need them recognized locally.
Setup surface
Requires a browser tool (the app page is a React SPA; web_fetch won't work). The sync script needs Python 3 and an OPENROUTER_API_KEY, which falls back to existing config if not set as an environment variable. Optional overrides: OPENCLAW_DIR and OPENCLAW_AGENT_DIR. After syncing, a gateway restart (openclaw gateway restart) applies changes.
Runner test plan
No execution testing has occurred yet. Planned checks include static analysis of the Python script, dependency verification, prompt-injection screening of extracted content, sandboxed dry-run tests (--dry-run flag), and screenshot review of the browser interaction steps. Residual risks include heuristic reasoning detection misclassifying models and alias conflicts.
Risk notes
The skill is additive only — it never removes delisted models. Reasoning capability detection relies on heuristics (architecture field plus known model families), so edge cases may be miscategorized. Browser-dependent Step 1 means failures if the leaderboard layout changes. Always inspect --dry-run output before letting the script write config files.
Source links
Discussion
Share practical experience, questions, or warnings with the community.
Sign in to join the discussion and vote on comments.
Sign in