Skip to content

xskill-agent_xskill.md

Latest commit

 

History

History
44 lines (33 loc) · 1.99 KB

xskill-agent_xskill.md

File metadata and controls

44 lines (33 loc) · 1.99 KB
Raw
repo XSkill-Agent/XSkill
url https://github.com/XSkill-Agent/XSkill
content_timestamp 2026-05-01
time_slice 2026-05
timestamp_source web_observed_public_github_page_2026_06_05_1543
collected_at 2026-06-05 15:43:00 +0800
source github

GitHub - XSkill-Agent/XSkill: XSkill is a multimodal continual-learning paper implementation that extracts task-level skills and action-level experiences from agent trajectories, stores them in a memory bank, and re-injects them during inference across benchmark suites.

Source: https://github.com/XSkill-Agent/XSkill

Raw public page capture notes

This raw-style public GitHub page capture was refreshed by the hourly public metadata update. Shell GitHub API access remained blocked in this workspace, so freshness is web-observed rather than API-verified.

Visible repository metadata

  • Repository: XSkill-Agent/XSkill
  • URL: https://github.com/XSkill-Agent/XSkill
  • Stars: 216
  • Forks: 27
  • Commits: 16
  • Issues: 1
  • Pull requests: 0
  • License: MIT
  • Primary language / stack signal: Python/Multimodal Agent/Memory Bank/Benchmark Eval
  • Latest visible dated signal: 2026-05-01
  • Collection timestamp: 2026-06-05T15:43:00+08:00

Visible README / page excerpts

  • The public GitHub page showed 216 stars, 27 forks, 1 issue, 16 commits, no releases, and explicit ICML 2026 acceptance news dated 2026-05-01.
  • Visible repository structure includes benchmark, eval, exskill, memory_bank, output, and logs folders rather than only a paper PDF pointer.
  • The README describes a two-phase loop: accumulation of skills and experiences from trajectories, then retrieval and injection during inference.
  • The overview names five benchmark suites and claims stronger zero-shot cross-task transferability over baselines.

Raw capture boundary

No benchmark was run, no source clone was modified, and no private or authenticated metadata was used. This file preserves public page evidence for downstream classification, model-card analysis, public reports, and the site index.