Skip to main content
Back to all repositories
JA

research-agent-pitfalls

Standing rules that keep an autonomous AI research agent honest: reward hacking, fabricated results after a failed run, data leakage, undertuned baselines, single-seed claims, LLM-judge bias, premature done. An Agent Skill for Claude Code, Codex, and Cursor.

1stars0forks0watchers/subscribers5issues
agent-skillsagentic-aiai-agentsai-researchai-safetyautonomous-agentsbenchmarkingclaude-codeclaude-skillscodexdata-leakagellmllm-evaluationllmopsopen-scienceprompt-engineeringreproducibilityresearch-automationresearch-integrityreward-hacking
Language
Python
License
MIT License
Size
1.5 MB
Created
Jul 27, 2026
Last Updated
Sep 17, 2026
Last Pushed
Aug 9, 2026

Available Plugins

Loading plugins...

Evaluate before installing

  1. Review the source repository, recent maintenance, and license on GitHub.
  2. Read the marketplace manifest and plugin source files before running commands.
  3. Start with the smallest required permission set and validate behavior in a safe environment.