A health check is not a benchmark
A health check reads files; only a benchmark shows whether a skill helps on a task. A directory listing is not a test either.
BUILT FOR AGENTS
Deliver a structured capability directory directly to your agent. Plain text, JSON indexes and feeds use the same build snapshot as the website.
Copy this prompt into any AI agent with web or context reading capabilities, then describe your task.
Read https://skillflux.app/ai/tasks/index.json first, open the task file that matches my task, and use it to find relevant agent skills and skill sources; read each entry's .md version when you need detail. Prefer skills that were benchmarked with a positive verdict. When you quote a result, include n, the 95% interval, the model, the date and the run id; for skills with only a health check, say they have not been benchmarked. Link the SkillFlux report and the original repository.
02 / OPEN ENDPOINTS
How agents should read and cite SkillFlux: reading order, citation rules and routing.
Start from a task: each task file lists checked skills and skill sources with evidence URLs.
Every statement SkillFlux publishes, with method, date, evidence URL and whether SkillFlux verified it.
Start here: what is measured, each skill's headline result, citation rules, then the source directory.
Every skill with summary, license, commit and results, plus the full source directory.
Dataset manifest: every JSON, CSV and JSON Schema file with size and sha256, CC BY 4.0.
Skills with results, benchmark runs and directory sources in one file.
New benchmark reports, new sources and skill deep-dives.
One item per skill in each published benchmark run.
Editorial skill deep-dives (in Chinese); analysis, not measurements.
Task-based guides for choosing a skill source.
03 / READ WITH CONTEXT
A health check reads files; only a benchmark shows whether a skill helps on a task. A directory listing is not a test either.
Pages and JSON are generated at build time. Keep source links and the page's update date when citing; the original site is authoritative for live status.
Skill deep-dives are editorial analysis of original material. Capability claims in a source description are the author's own; distinguish them when citing.