Research Loop

Claude Code compatibility mirror for the Codex-native 10000 Mentors Research Workflow. Use only when running from Claude Code and the user wants the same sou...

This skill appears coherent for running a repo-focused research loop. Before installing or running it: (1) confirm the runtime has python3, git, and the GitHub CLI installed and authenticated if you expect publishing; GH/git will use whatever credentials are already configured, so be mindful that pushes or gh actions could occur under those identities. (2) Verify the `autonomous_research_workflow` CLI/module the skill invokes is present in the runtime or in the target repo to avoid the agent attempting to download or install code at runtime. (3) Because the skill writes changes into a `source_changes/` mirror and may publish via GitHub, consider trying it first in a sandbox or on a fork to observe behavior. (4) If you do not want any remote publishing, ensure the environment has no active gh auth or run with a local-only dry-run. Overall the skill is internally consistent, but watch for implicit use of existing git/gh credentials and ensure you trust the runtime environment.

Type: OpenClaw Skill Name: research-loop Version: 1.0.0 The research-loop skill bundle defines a highly structured, 15-phase autonomous research workflow focused on scientific rigor, reproducibility, and 'protocol hygiene.' While it requests broad permissions including Bash, WebSearch, and the GitHub CLI (gh), these tools are consistent with its stated purpose of automating research, running experiments, and publishing results. The instructions (SKILL.md and references/) are designed to prevent common AI agent failures, such as making misleading claims or getting stuck in 'asset-polishing' loops, rather than to exfiltrate data or establish persistence. No evidence of malicious intent, obfuscation, or unauthorized data access was found.