Skip to content

The agent that grows with you

chore(catalog): review disclosure for model-leaderboard

Tekniumdf1dded57e6b
diff --git a/plugin-catalog/model-leaderboard.yaml b/plugin-catalog/model-leaderboard.yaml
index f5bd65ebd92..ddd39a8b989 100644
--- a/plugin-catalog/model-leaderboard.yaml
+++ b/plugin-catalog/model-leaderboard.yaml
@@ -1,24 +1,11 @@
name: model-leaderboard
repo: https://github.com/e-is/hermes-model-leaderboard
sha: 516e2f8d4dd0790f2c486d1fe0b444de5c2b3c10
-description: 'Model leaderboard for Hermes Desktop: a full /model-leaderboard page that ranks any OpenRouter
- model for a chosen Hermes profile — context window, prices, Artificial Analysis indices and curated
- SWE-bench Verified / LiveCodeBench — with per-profile criteria, six score views (cards, table, score
- bars, histograms, heatmap, quality-vs-cost Pareto frontier), an OpenRouter news feed (new models, price
- changes, promos) and a documented REST surface under /api/plugins/model-leaderboard/. Disclosure — outbound
- HTTPS to OpenRouter''s public model API (catalogue and pricing) on a 60 s poll and a forced refresh
- every 10 minutes while the page is visible, and to Hugging Face''s public model API (safetensors metadata,
- for parameter counts) on demand and cached; reads the installed Hermes profiles'' own config.yaml files
- once, to seed the tracked list from the models those profiles declare; keeps its state in <hermes home>/plugin-data/model-leaderboard/
- (seeded from the JSON shipped with the plugin), which LLM_DASH_DATA / LLM_DASH_CACHE_TTL can move or
- tune; the optional "Auto-fill criteria" action runs on the host LLM facade (ctx.llm.complete_structured),
- so it spends the user''s configured model tokens; ships one read-only skill (model-leaderboard:refresh-benchmarks)
- whose only write is a POST to its own /benchmarks route; no API keys of its own, no writes outside its
- data directory.'
+description: 'Model leaderboard for Hermes Desktop: a full /model-leaderboard page that ranks any OpenRouter model for a chosen Hermes profile — context window, prices, Artificial Analysis indices and curated SWE-bench Verified / LiveCodeBench — with per-profile criteria, six score views (cards, table, score bars, histograms, heatmap, quality-vs-cost Pareto frontier), an OpenRouter news feed (new models, price changes, promos) and a documented REST surface under /api/plugins/model-leaderboard/. Disclosure — outbound HTTPS to OpenRouter''s public model API (catalogue and pricing) on a 60 s poll and a forced refresh every 10 minutes while the page is visible, and to Hugging Face''s public model API (safetensors metadata, for parameter counts) on demand and cached; reads the installed Hermes profiles'' own config.yaml files once, to seed the tracked list from the models those profiles declare; keeps its state in <hermes home>/plugin-data/model-leaderboard/ (seeded from the JSON shipped with the plugin), which LLM_DASH_DATA / LLM_DASH_CACHE_TTL can move or tune; the optional "Auto-fill criteria" action runs on the host LLM facade (ctx.llm.complete_structured), so it spends the user''s configured model tokens; ships one read-only skill (model-leaderboard:refresh-benchmarks) whose only write is a POST to its own /benchmarks route; no API keys of its own, no writes outside its data directory.'
maintainer: e-is
tier: community
category: models
-requires_hermes: '>=0.21'
+requires_hermes: ">=0.21.6"
docs_url: https://github.com/e-is/hermes-model-leaderboard#readme
version: "1.2.0"
image: https://raw.githubusercontent.com/e-is/hermes-model-leaderboard/516e2f8d4dd0790f2c486d1fe0b444de5c2b3c10/assets/card-podium.png