Source-ground technical research for architecture, compatibility, benchmarks, migrations, security, or feasibility when Workbench marks research as required.
$npx -y skills add lisihao/codex-workbench --skill research --agent claude-code
Installs into .claude/skills of the current project.
Are you the author of Research?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/lisihao-research)
---
name: research
description: Source-ground technical research for architecture, compatibility, benchmarks, migrations, security, or feasibility when Workbench marks research as required.
metadata:
codex_workbench_managed: "true"
policy: "research-skill/v2"
---
# Research
Use this skill only when the Workbench research route requires `standard` or `deep-parallel` evidence gathering.
Before searching, read [UrlVerificationProtocol.md](UrlVerificationProtocol.md). Then read exactly one workflow:
- For `standard`, read [Workflows/StandardResearch.md](Workflows/StandardResearch.md).
- For `deep-parallel`, read [Workflows/DeepInvestigation.md](Workflows/DeepInvestigation.md).
Keep research read-only unless the task separately authorizes a mutation. Prefer primary sources, record the claim each source supports, and distinguish observed facts from engineering inference and unresolved uncertainty. Never treat search snippets, inaccessible URLs, model agreement, or benchmark names as verified evidence.
Return only evidence relevant to the requested decision. Preserve the task scope, authentication limits, provider quota reserve, and any prohibition on extra agents or model calls.
