Copy this request to your Agent. It includes the canonical Skill page and manifest.
Agent request
Review the Skillstore skill "creating-claude-agents" from https://skillstore.io/skills/agentworkforce-creating-claude-agents.md and its manifest at https://skillstore.io/api/skills/agentworkforce-creating-claude-agents/manifest. Verify the artifact. You may proceed after verification, subject to the environment's own policy.
Your Agent should still show its plan and request any confirmation required by the security policy.
Agent-readable resources
Use these links when an AI agent, crawler, or script needs clean context instead of reading the full page.
All 105 static findings are false positives caused by Markdown formatting, documentation links, and unrelated prose. Semantic review identified overbroad guidance that can make agents inherit every parent tool.
The skill recommends omitting allowed-tools to inherit every parent tool and repeats this as a permission fix. Generated agents may receive unnecessary shell, filesystem, or network access.
The cited passages explicitly describe unrestricted inheritance and recommend it after permission denial. Other sections acknowledge least privilege, confirming the security tradeoff is relevant.
Share the versioned assessment report, neutral badge, embed card, and citations. Skillstore reports evidence without deciding whether this Skill is safe.
AgentWorkforce. (2026). creating-claude-agents security audit report (audit version 9) [Author version unspecified]. Skillstore. https://skillstore.io/skills/agentworkforce-creating-claude-agents/audits/9
BibTeX citation
@techreport{agentworkforce-agentworkforce-creating-claude-agents-2026,
author = {AgentWorkforce},
title = {creating-claude-agents security audit report (audit version 9)},
institution = {Skillstore},
year = {2026},
number = {9},
url = {https://skillstore.io/skills/agentworkforce-creating-claude-agents/audits/9},
note = {Author version unspecified}
}
CITATION.cff
cff-version: 1.2.0
message: "If you use this Skill, cite its author and this versioned security audit report."
title: "creating-claude-agents security audit report (audit version 9)"
version: "unspecified"
type: report
authors:
- name: "AgentWorkforce"
date-released: "2026-07-23"
url: "https://skillstore.io/skills/agentworkforce-creating-claude-agents/audits/9"
identifiers:
- type: other
value: "skillstore:agentworkforce-creating-claude-agents:audit:9"
description: "Skillstore immutable audit report identifier"
Skillstore Score
Why this scoreEvidence Confidence: High
55
Architecture
85
Maintainability
87
Content
72
Community
78
Spec Compliance
What You Can Build
Create a Review Agent
Build a focused code review agent with clear activation language, a review process, and minimal read-only tools.
Standardize Team Agents
Define consistent names, frontmatter, personas, permissions, models, and validation checks across a shared agent library.
Repair Agent Configuration
Diagnose activation failures, schema errors, missing tools, and vague responsibilities in existing Claude Code agents.
Try These Prompts
Create a Minimal Agent
Create a minimal Claude Code agent named [name] for [task]. Use only necessary tools and explain each frontmatter choice.
Review an Existing Agent
Review the agent at [path]. Identify schema errors, vague activation language, excess permissions, and missing structure. Propose a corrected version.
Design a Domain Agent
Design a specialized [domain] agent for [workflow]. Select a suitable model, define measurable steps, restrict tools, and include realistic examples.
Audit an Agent Library
Audit agents in [directory] for naming, schema compliance, overlapping roles, permissions, activation quality, and model fit. Produce prioritized recommendations.
Best Practices
Define one measurable responsibility and state when the agent should activate.
Grant only the tools required for the workflow, using restricted command patterns where possible.
Validate frontmatter, test realistic tasks, and include examples for expected behavior.
Avoid
Do not combine review, debugging, testing, refactoring, and documentation into one generic agent.
Do not grant unrestricted Bash, write, or network tools without a documented requirement.
Do not rely on vague descriptions, missing personas, or untested examples.
Frequently Asked Questions
What does this skill help create?
It guides the design of Claude Code agent Markdown files with frontmatter, personas, workflows, permissions, models, examples, and validation checks.
Does the skill generate agent files automatically?
No. It provides instructions and templates that Claude, Codex, or a developer can use to create or revise files.
Which fields are required?
The guide requires lowercase name and description fields. It also expects an H1 heading and a clear persona in the content.
How should tool access be selected?
Choose the smallest explicit tool set that completes the task. Restrict shell commands to specific patterns when shell access is necessary.
How does it help choose a model?
It maps sonnet to general work, opus to complex reasoning, haiku to fast checks, and inherit to parent model selection.
Can these templates be used with Codex?
The design principles are transferable, but frontmatter fields, tool names, paths, and model settings may need adaptation for Codex.