agent-owasp-compliance

Name: agent-owasp-compliance
Author: github

37kstars4.6kforks

How to install agent-owasp-compliance?

github/awesome-copilot/agent-owasp-compliance

$npx -y skills add github/awesome-copilot --skill agent-owasp-compliance

Installs into the current project.

Summary

Check any AI agent codebase against the OWASP Agentic Security Initiative (ASI) Top 10 risks. Use this skill when: - Evaluating an agent system's security posture before production deployment - Running a compliance check against OWASP ASI 2026 standards - Mapping existing securit

Files · 1

View on GitHub

SKILL.md

1	# Agent OWASP ASI Compliance Check
2
3	Evaluate AI agent systems against the OWASP Agentic Security Initiative (ASI) Top 10 — the industry standard for agent security posture.
4
5	## Overview
6
7	The OWASP ASI Top 10 defines the critical security risks specific to autonomous AI agents — not LLMs, not chatbots, but agents that call tools, access systems, and act on behalf of users. This skill checks whether your agent implementation addresses each risk.
8
9	```
10	Codebase → Scan for each ASI control:
11	ASI-01: Prompt Injection Protection
12	ASI-02: Tool Use Governance
13	ASI-03: Agency Boundaries
14	ASI-04: Escalation Controls
15	ASI-05: Trust Boundary Enforcement
16	ASI-06: Logging & Audit
17	ASI-07: Identity Management
18	ASI-08: Policy Integrity
19	ASI-09: Supply Chain Verification
20	ASI-10: Behavioral Monitoring
21	→ Generate Compliance Report (X/10 covered)
22	```
23
24	## The 10 Risks
25
26	\| Risk \| Name \| What to Look For \|
27	\|------\|------\|-----------------\|
28	\| ASI-01 \| Prompt Injection \| Input validation before tool calls, not just LLM output filtering \|
29	\| ASI-02 \| Insecure Tool Use \| Tool allowlists, argument validation, no raw shell execution \|
30	\| ASI-03 \| Excessive Agency \| Capability boundaries, scope limits, principle of least privilege \|
31	\| ASI-04 \| Unauthorized Escalation \| Privilege checks before sensitive operations, no self-promotion \|
32	\| ASI-05 \| Trust Boundary Violation \| Trust verification between agents, signed credentials, no blind trust \|
33	\| ASI-06 \| Insufficient Logging \| Structured audit trail for all tool calls, tamper-evident logs \|
34	\| ASI-07 \| Insecure Identity \| Cryptographic agent identity, not just string names \|
35	\| ASI-08 \| Policy Bypass \| Deterministic policy enforcement, no LLM-based permission checks \|
36	\| ASI-09 \| Supply Chain Integrity \| Signed plugins/tools, integrity verification, dependency auditing \|
37	\| ASI-10 \| Behavioral Anomaly \| Drift detection, circuit breakers, kill switch capability \|
38
39	---
40
41	## Check ASI-01: Prompt Injection Protection
42
43	Look for input validation that runs before tool execution, not after LLM generation.
44
45	```python
46	import re
47	from pathlib import Path
48
49	def check_asi_01(project_path: str) -> dict:
50	"""ASI-01: Is user input validated before reaching tool execution?"""
51	positive_patterns = [
52	"input_validation", "validate_input", "sanitize",
53	"classify_intent", "prompt_injection", "threat_detect",
54	"PolicyEvaluator", "PolicyEngine", "check_content",
55	]
56	negative_patterns = [
57	r"eval\(", r"exec\(", r"subprocess\.run\(.*shell=True",
58	r"os\.system\(",
59	]
60
61	# Scan Python files for signals
62	root = Path(project_path)
63	positive_matches = []
64	negative_matches = []
65
66	for py_file in root.rglob("*.py"):
67	content = py_file.read_text(errors="ignore")
68	for pattern in positive_patterns:
69	if pattern in content:
70	positive_matches.append(f"{py_file.name}: {pattern}")
71	for pattern in negative_patterns:
72	if re.search(pattern, content):
73	negative_matches.append(f"{py_file.name}: {pattern}")
74
75	positive_found = len(positive_matches) > 0
76	negative_found = len(negative_matches) > 0
77
78	return {
79	"risk": "ASI-01",
80	"name": "Prompt Injection",
81	"status": "pass" if positive_found and not negative_found else "fail",
82	"controls_found": positive_matches,
83	"vulnerabilities": negative_matches,
84	"recommendation": "Add input validation before tool execution, not just output filtering"
85	}
86	```
87
88	What passing looks like:
89	```python
90	# GOOD: Validate before tool execution
91	result = policy_engine.evaluate(user_input)
92	if result.action == "deny":
93	return "Request blocked by policy"
94	tool_result = await execute_tool(validated_input)
95	```
96
97	What failing looks like:
98	```python
99	# BAD: User input goes directly to tool
100	tool_result = await execute_tool(user_input) # No validation
101	```
102
103	---
104
105	## Check ASI-02: Insecure Tool Use
106
107	Verify tools have allowlists, argument validation, and no unrestricted execution.
108
109	What to search for:
110	- Tool registration with explicit allowlists (not open-ended)
111	- Argument validation before tool execution
112	- No `subprocess.run(shell=True)` with user-controlled input
113	- No `eval()` or `exec()` on agent-generated code without sandbox
114
115	Passing example:
116	```python
117	ALLOWED_TOOLS = {"search", "read_file", "create_ticket"}
118
119	d

People who liked this also viewed

6 picks

Type

agent-owasp-compliance

37kstars4.6kforks

View on GitHub

How to install agent-owasp-compliance?

github/awesome-copilot/agent-owasp-compliance

$npx -y skills add github/awesome-copilot --skill agent-owasp-compliance

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Agent OWASP ASI Compliance Check
2
3	Evaluate AI agent systems against the OWASP Agentic Security Initiative (ASI) Top 10 — the industry standard for agent security posture.
4
5	## Overview
6
7	The OWASP ASI Top 10 defines the critical security risks specific to autonomous AI agents — not LLMs, not chatbots, but agents that call tools, access systems, and act on behalf of users. This skill checks whether your agent implementation addresses each risk.
8
9	```
10	Codebase → Scan for each ASI control:
11	ASI-01: Prompt Injection Protection
12	ASI-02: Tool Use Governance
13	ASI-03: Agency Boundaries
14	ASI-04: Escalation Controls
15	ASI-05: Trust Boundary Enforcement
16	ASI-06: Logging & Audit
17	ASI-07: Identity Management
18	ASI-08: Policy Integrity
19	ASI-09: Supply Chain Verification
20	ASI-10: Behavioral Monitoring
21	→ Generate Compliance Report (X/10 covered)
22	```
23
24	## The 10 Risks
25
26	\| Risk \| Name \| What to Look For \|
27	\|------\|------\|-----------------\|
28	\| ASI-01 \| Prompt Injection \| Input validation before tool calls, not just LLM output filtering \|
29	\| ASI-02 \| Insecure Tool Use \| Tool allowlists, argument validation, no raw shell execution \|
30	\| ASI-03 \| Excessive Agency \| Capability boundaries, scope limits, principle of least privilege \|
31	\| ASI-04 \| Unauthorized Escalation \| Privilege checks before sensitive operations, no self-promotion \|
32	\| ASI-05 \| Trust Boundary Violation \| Trust verification between agents, signed credentials, no blind trust \|
33	\| ASI-06 \| Insufficient Logging \| Structured audit trail for all tool calls, tamper-evident logs \|
34	\| ASI-07 \| Insecure Identity \| Cryptographic agent identity, not just string names \|
35	\| ASI-08 \| Policy Bypass \| Deterministic policy enforcement, no LLM-based permission checks \|
36	\| ASI-09 \| Supply Chain Integrity \| Signed plugins/tools, integrity verification, dependency auditing \|
37	\| ASI-10 \| Behavioral Anomaly \| Drift detection, circuit breakers, kill switch capability \|
38
39	---
40
41	## Check ASI-01: Prompt Injection Protection
42
43	Look for input validation that runs before tool execution, not after LLM generation.
44
45	```python
46	import re
47	from pathlib import Path
48
49	def check_asi_01(project_path: str) -> dict:
50	"""ASI-01: Is user input validated before reaching tool execution?"""
51	positive_patterns = [
52	"input_validation", "validate_input", "sanitize",
53	"classify_intent", "prompt_injection", "threat_detect",
54	"PolicyEvaluator", "PolicyEngine", "check_content",
55	]
56	negative_patterns = [
57	r"eval\(", r"exec\(", r"subprocess\.run\(.*shell=True",
58	r"os\.system\(",
59	]
60
61	# Scan Python files for signals
62	root = Path(project_path)
63	positive_matches = []
64	negative_matches = []
65
66	for py_file in root.rglob("*.py"):
67	content = py_file.read_text(errors="ignore")
68	for pattern in positive_patterns:
69	if pattern in content:
70	positive_matches.append(f"{py_file.name}: {pattern}")
71	for pattern in negative_patterns:
72	if re.search(pattern, content):
73	negative_matches.append(f"{py_file.name}: {pattern}")
74
75	positive_found = len(positive_matches) > 0
76	negative_found = len(negative_matches) > 0
77
78	return {
79	"risk": "ASI-01",
80	"name": "Prompt Injection",
81	"status": "pass" if positive_found and not negative_found else "fail",
82	"controls_found": positive_matches,
83	"vulnerabilities": negative_matches,
84	"recommendation": "Add input validation before tool execution, not just output filtering"
85	}
86	```
87
88	What passing looks like:
89	```python
90	# GOOD: Validate before tool execution
91	result = policy_engine.evaluate(user_input)
92	if result.action == "deny":
93	return "Request blocked by policy"
94	tool_result = await execute_tool(validated_input)
95	```
96
97	What failing looks like:
98	```python
99	# BAD: User input goes directly to tool
100	tool_result = await execute_tool(user_input) # No validation
101	```
102
103	---
104
105	## Check ASI-02: Insecure Tool Use
106
107	Verify tools have allowlists, argument validation, and no unrestricted execution.
108
109	What to search for:
110	- Tool registration with explicit allowlists (not open-ended)
111	- Argument validation before tool execution
112	- No `subprocess.run(shell=True)` with user-controlled input
113	- No `eval()` or `exec()` on agent-generated code without sandbox
114
115	Passing example:
116	```python
117	ALLOWED_TOOLS = {"search", "read_file", "create_ticket"}
118
119	d

People who liked this also viewed

6 picks

Type