Guide · Agent skill
How to use the Agent OWASP ASI Compliance Check agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The agent-owasp-compliance skill lets an AI coding agent review an agent project's code and documentation against the ten risks in the OWASP Agentic Security Initiative (ASI) Top 10. It is aimed at developers preparing an agent system for production and at security reviewers who want a first-pass checklist. By the end of this guide you will know what the skill does, how to install it, how to ask for a check in plain language, and what its results can and cannot tell you.
Agent OWASP ASI Compliance Checkgithub/awesome-copilot · MIT Download the skill Quick start
- 1Download the ZIP. Get the agent-owasp-compliance ZIP from this page.
- 2Unzip into a skills folder. Extract it into ~/.claude/skills/ for personal use or into the project's .claude/skills/ folder.
- 3Or upload in the Claude apps. Go to Settings → Capabilities → Skills and upload the skill there instead.
- 4Confirm Python is available. The included check examples are Python scripts that scan .py files.
- 5Open the agent project. Make sure the agent's working directory contains the project's source code and documentation.
- 6Ask for the check. Request an ASI compliance check or agentic security audit in plain language.
- 7Review the report. Read each risk's status, matched controls and flagged patterns, then verify the findings yourself.
What is an agent skill and what does this one do?
An agent skill is a folder containing a SKILL.md file that gives an AI agent instructions for a specific task. When the skill is installed, the agent can load these instructions when a request matches the skill's description.
Once installed, this skill lets the agent review an AI agent project's code and documentation against the ten OWASP ASI risks, such as prompt injection, unsafe tool use, excessive permissions and missing audit logs. For each risk it looks for evidence of a protective control and reports whether the project appears to cover it, then produces a summary such as how many of the ten are covered.
- ASI-01 Prompt Injection: input validation before tool calls, not just LLM output filtering
- ASI-02 Insecure Tool Use: tool allowlists, argument validation, no raw shell execution
- ASI-03 Excessive Agency: capability boundaries and least privilege
- ASI-04 Unauthorized Escalation: privilege checks before sensitive operations
- ASI-05 Trust Boundary Violation: verification between agents, signed credentials
- ASI-06 Insufficient Logging: structured audit trail for tool calls
- ASI-07 Insecure Identity: cryptographic agent identity
- ASI-08 Policy Bypass: deterministic policy enforcement, not LLM-based permission checks
- ASI-09 Supply Chain Integrity: signed plugins, integrity checks, dependency auditing
- ASI-10 Behavioral Anomaly: drift detection, circuit breakers, kill switch
How do you install the skill?
Download the ZIP for this skill from this page and unzip it. You can install it in one of two places. For a personal setup, unzip it into ~/.claude/skills/. For a single project, unzip it into the project's .claude/skills/ folder.
In the Claude apps you can instead upload the skill under Settings → Capabilities → Skills.
- Personal install: ~/.claude/skills/agent-owasp-compliance/
- Project install: .claude/skills/agent-owasp-compliance/ inside your project
- Claude apps: Settings → Capabilities → Skills
What do you need before you run a check?
The skill's example checks are Python scripts that scan .py files, so you need Python available. You also need the agent project's source code in the working directory the agent is using, because the skill searches that code and its documentation.
The skill runs scripts and does not use the network, according to its listing. Read the safety notes below before running it on sensitive code.
- Python installed (the included check examples are Python)
- The agent project's source code in the working directory
- Scripts: yes. Network access: no
How do you trigger the skill with a natural request?
You do not need a special command. Describe what you want in plain language and the agent will match your request to the skill's description. The skill is meant to be triggered by requests about checking an agent system's security posture, mapping controls to the risks, generating a compliance report, or running an agentic security audit.
Name the project or folder if the agent is working across several. Asking for the ASI list by number helps the agent focus on specific risks.
- "Is my agent OWASP compliant?"
- "Check ASI compliance for this repo before we deploy."
- "Run an agentic security audit and give me a report I can send to the security team."
- "Map our existing controls to the 10 agentic risks."
- "Check ASI-01 and ASI-02 only in the src folder."
What does a check look like in practice?
The skill's method is to scan the codebase for signals for each ASI control. For example, the ASI-01 check searches Python files for input validation and policy names such as validate_input, sanitize and PolicyEngine, and also flags patterns such as eval(, exec( and os.system(. A risk is marked as passing only when a positive signal is found and no negative pattern is found.
The result for each risk is a status such as pass or fail, with the matched controls and any flagged patterns, plus a recommendation. The final report counts how many of the ten risks are covered.
- Controls found: the file names and patterns that matched
- Vulnerabilities: risky patterns that matched
- Recommendation: a suggested fix for the risk
tool_result = await execute_tool(validated_input)What should you watch out for?
The checks are keyword and pattern searches. A project can have a real control under a name the skill does not look for, which gives a false fail. A project can also contain a matching word in a comment or unrelated code, which gives a false pass. Treat every verdict as a heuristic starting point.
The skill is not a replacement for a professional penetration test or a certified audit. It is also not useful for plain chatbots or LLM apps without tool access, because the risks it checks do not apply to them.
- Not a substitute for a penetration test or certified audit
- Keyword matching can give both false passes and false failures
- Checks for projects with tools, system access or delegated actions; not plain chatbots
What people use it for
- Pre-production security review of an agent system before launch
- First-pass checklist for a formal security audit
- Mapping an agent framework's security features to the ten OWASP ASI risks
- Finding places where tool calls run without input validation or where subprocess calls use shell=True
- Producing a summary of how many of the ten risks a project appears to cover
Frequently asked questions
How do I install the agent-owasp-compliance skill?
Download the ZIP from this page and unzip it into ~/.claude/skills/ for personal use or into a project's .claude/skills/ folder. In the Claude apps, you can instead upload it under Settings → Capabilities → Skills.
What do I need to run the OWASP ASI compliance check?
You need Python, because the included check examples are Python scripts that scan .py files. You also need the agent project's source code in the working directory. The skill runs scripts and does not use the network.
Is the agent-owasp-compliance skill free?
The source material for this page lists the skill under the MIT licence on the github/awesome-copilot repository. It does not state any price, so check the repository for licence terms.
Does passing this check mean my agent is secure?
No. The skill's verdicts are heuristic keyword and pattern matches, not a full security assessment. A pass can be a false pass and a fail can be a false fail, so it should be a first-pass checklist, not proof of security.
What are the alternatives to this skill?
The source does not name alternatives. For a formal assessment, the skill's own description points to a professional penetration test or certified audit. You can also compare the ten ASI risks yourself against your project using the checklist in SKILL.md.
Why did the skill flag a risk my project handles?
The check looks for specific names and patterns, so it may not recognise a control that uses different naming. Read the matched files and patterns in the report, confirm the control manually, and note that the verdict is heuristic.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
This skill suits developers preparing an agent system for production and security reviewers who want a structured first-pass checklist against the OWASP ASI Top 10. It is not suited to replacing a penetration test or certified audit, or to projects without tool access, such as plain chatbots. Use its results as a starting point for human review.
Download the skill