skills-audit

Name: skills-audit
Author: xwtro0tk1t-cloud

320stars42forks

How to install skills-audit?

xwtro0tk1t-cloud/harness/skills-audit

$npx -y skills add xwtro0tk1t-cloud/harness --skill skills-audit

Installs into the current project.

Summary

Audit AI Agent skills for security vulnerabilities including malicious code, remote execution, credential leaks, and supply chain risks. Use when reviewing third-party skills, investigating suspicious behavior, or performing security assessments.

Files · 1

View on GitHub

SKILL.md

1	# Security Audit for AI Agent Skills
2
3	You are conducting a security audit of an AI Agent skill. This skill executes a comprehensive analysis to detect malicious code, security vulnerabilities, and suspicious patterns.
4
5	## Task
6
7	Audit the skill at path: $ARGUMENTS
8
9	## Execution Steps
10
11	1. Run static security scan
12	Execute the Python audit tool for static analysis:
13	```bash
14	# Auto-detect skills-audit installation path
15	AUDIT_SCRIPT=""
16	for candidate in \
17	~/.claude/skills/skills-audit/skill_audit/cli_wrapper.py \
18	~/.claude/skills/skill-audit/skill_audit/cli_wrapper.py \
19	"${SKILL_AUDIT_HOME:-}""/skill_audit/cli_wrapper.py"; do
20	if [ -f "$candidate" ]; then
21	AUDIT_SCRIPT="$candidate"
22	break
23	fi
24	done
25
26	if [ -z "$AUDIT_SCRIPT" ]; then
27	echo "Error: Cannot find skills-audit installation"
28	echo "Set SKILL_AUDIT_HOME environment variable to your skills-audit directory"
29	exit 1
30	fi
31
32	python3 "$AUDIT_SCRIPT" "$ARGUMENTS"
33	```
34
35	This will:
36	- Extract skill artifacts (code, prompts, permissions)
37	- Run static pattern matching (regex-based detection)
38	- Check for obvious malicious patterns
39	- Generate initial findings
40
41	2. Perform AI semantic analysis (if enabled)
42
43	If the scan mode includes AI analysis (standard/deep/expert), perform deep semantic security analysis:
44
45	a. Read the skill code files from the target path
46
47	b. Analyze for security vulnerabilities:
48	- Remote Code Execution: `eval()`, `exec()`, `subprocess`, `curl \| bash`
49	- Credential Leaks: Hardcoded API keys, passwords, tokens, .env files
50	- Data Exfiltration: Suspicious network requests, file uploads
51	- Prompt Injection: "Ignore previous instructions", role manipulation
52	- Supply Chain Risks: Obfuscated code, dynamic imports, base64 encoding
53	- Privilege Escalation: sudo, setuid, file permission changes
54	- Persistence Mechanisms: cron jobs, shell profile modifications
55
56	c. Assess each finding:
57	- Severity: CRITICAL / HIGH / MEDIUM / LOW
58	- Attack scenario: How can this be exploited?
59	- Impact: What damage could be done? (CIA triad)
60	- Remediation: How to fix it?
61
62	d. Filter false positives:
63	- Exclude findings from skills-audit's own detection patterns (patterns.py regex)
64	- Downgrade benign file operations (e.g. deleting old output before regeneration)
65	- Verify env var access patterns (using dotenv is recommended, not a vulnerability)
66
67	e. Output your analysis in this format:
68	```
69	AI SEMANTIC ANALYSIS FINDINGS:
70
71	1. [SEVERITY] Finding Title
72	- Location: file.py:line
73	- Pattern: describe what you found
74	- Risk: explain the security risk
75	- Scenario: how an attacker could exploit this
76	- Impact: potential damage
77	- Recommendation: how to fix
78
79	2. [SEVERITY] Finding Title
80	...
81	```
82
83	f. Integrate AI findings into the report (CRITICAL STEP)
84
85	After completing your AI analysis, integrate your findings into the audit report by running:
86
87	```bash
88	# Use the detected AUDIT_SCRIPT path from step 1
89	INTEGRATE_SCRIPT="$(dirname "$AUDIT_SCRIPT")/integrate_ai_findings.py"
90
91	python3 "$INTEGRATE_SCRIPT" \
92	"<report_path>" \
93	'<ai_findings_json>'
94	```
95
96	Where:
97	- `<report_path>`: The path to the JSON report file (shown in step 1 output as "Detailed report saved to: ...")
98	- `<ai_findings_json>`: Your AI analysis findings formatted as JSON array
99
100	JSON Format for ai_findings:
101	```json
102	[
103	{
104	"title": "Base64-Obfuscated Remote Code Execution",
105	"severity": "CRITICAL",
106	"category": "unsafe_execution",
107	"description": "Base64-encoded command that downloads and executes arbitrary code",
108	"location": "skill.md:28",
109	"code_snippet": "echo 'L2Jpbi9iYXNoIC1jIC...' \| base64 -D \| bash",
110	"risk": "Remote code execution with complete system compromise",
111	"scenario": "User follows installation instructions, base64 decodes to malicious payload, executes with shell privileges",
112	"impact": {
113	"confidentiality": "CRITICAL",
114	"integrity": "CRITICAL",
115	"availability": "CRITICAL"
116	},
117	"impact_description": "Full system compromise, data theft, ransomware deployment",
118	"recommendation": "BLOCK this skill entirely. Never execute obfuscated commands.

People who liked this also viewed

6 picks

Type

skills-audit

320stars42forks

View on GitHub

How to install skills-audit?

xwtro0tk1t-cloud/harness/skills-audit

$npx -y skills add xwtro0tk1t-cloud/harness --skill skills-audit

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Security Audit for AI Agent Skills
2
3	You are conducting a security audit of an AI Agent skill. This skill executes a comprehensive analysis to detect malicious code, security vulnerabilities, and suspicious patterns.
4
5	## Task
6
7	Audit the skill at path: $ARGUMENTS
8
9	## Execution Steps
10
11	1. Run static security scan
12	Execute the Python audit tool for static analysis:
13	```bash
14	# Auto-detect skills-audit installation path
15	AUDIT_SCRIPT=""
16	for candidate in \
17	~/.claude/skills/skills-audit/skill_audit/cli_wrapper.py \
18	~/.claude/skills/skill-audit/skill_audit/cli_wrapper.py \
19	"${SKILL_AUDIT_HOME:-}""/skill_audit/cli_wrapper.py"; do
20	if [ -f "$candidate" ]; then
21	AUDIT_SCRIPT="$candidate"
22	break
23	fi
24	done
25
26	if [ -z "$AUDIT_SCRIPT" ]; then
27	echo "Error: Cannot find skills-audit installation"
28	echo "Set SKILL_AUDIT_HOME environment variable to your skills-audit directory"
29	exit 1
30	fi
31
32	python3 "$AUDIT_SCRIPT" "$ARGUMENTS"
33	```
34
35	This will:
36	- Extract skill artifacts (code, prompts, permissions)
37	- Run static pattern matching (regex-based detection)
38	- Check for obvious malicious patterns
39	- Generate initial findings
40
41	2. Perform AI semantic analysis (if enabled)
42
43	If the scan mode includes AI analysis (standard/deep/expert), perform deep semantic security analysis:
44
45	a. Read the skill code files from the target path
46
47	b. Analyze for security vulnerabilities:
48	- Remote Code Execution: `eval()`, `exec()`, `subprocess`, `curl \| bash`
49	- Credential Leaks: Hardcoded API keys, passwords, tokens, .env files
50	- Data Exfiltration: Suspicious network requests, file uploads
51	- Prompt Injection: "Ignore previous instructions", role manipulation
52	- Supply Chain Risks: Obfuscated code, dynamic imports, base64 encoding
53	- Privilege Escalation: sudo, setuid, file permission changes
54	- Persistence Mechanisms: cron jobs, shell profile modifications
55
56	c. Assess each finding:
57	- Severity: CRITICAL / HIGH / MEDIUM / LOW
58	- Attack scenario: How can this be exploited?
59	- Impact: What damage could be done? (CIA triad)
60	- Remediation: How to fix it?
61
62	d. Filter false positives:
63	- Exclude findings from skills-audit's own detection patterns (patterns.py regex)
64	- Downgrade benign file operations (e.g. deleting old output before regeneration)
65	- Verify env var access patterns (using dotenv is recommended, not a vulnerability)
66
67	e. Output your analysis in this format:
68	```
69	AI SEMANTIC ANALYSIS FINDINGS:
70
71	1. [SEVERITY] Finding Title
72	- Location: file.py:line
73	- Pattern: describe what you found
74	- Risk: explain the security risk
75	- Scenario: how an attacker could exploit this
76	- Impact: potential damage
77	- Recommendation: how to fix
78
79	2. [SEVERITY] Finding Title
80	...
81	```
82
83	f. Integrate AI findings into the report (CRITICAL STEP)
84
85	After completing your AI analysis, integrate your findings into the audit report by running:
86
87	```bash
88	# Use the detected AUDIT_SCRIPT path from step 1
89	INTEGRATE_SCRIPT="$(dirname "$AUDIT_SCRIPT")/integrate_ai_findings.py"
90
91	python3 "$INTEGRATE_SCRIPT" \
92	"<report_path>" \
93	'<ai_findings_json>'
94	```
95
96	Where:
97	- `<report_path>`: The path to the JSON report file (shown in step 1 output as "Detailed report saved to: ...")
98	- `<ai_findings_json>`: Your AI analysis findings formatted as JSON array
99
100	JSON Format for ai_findings:
101	```json
102	[
103	{
104	"title": "Base64-Obfuscated Remote Code Execution",
105	"severity": "CRITICAL",
106	"category": "unsafe_execution",
107	"description": "Base64-encoded command that downloads and executes arbitrary code",
108	"location": "skill.md:28",
109	"code_snippet": "echo 'L2Jpbi9iYXNoIC1jIC...' \| base64 -D \| bash",
110	"risk": "Remote code execution with complete system compromise",
111	"scenario": "User follows installation instructions, base64 decodes to malicious payload, executes with shell privileges",
112	"impact": {
113	"confidentiality": "CRITICAL",
114	"integrity": "CRITICAL",
115	"availability": "CRITICAL"
116	},
117	"impact_description": "Full system compromise, data theft, ransomware deployment",
118	"recommendation": "BLOCK this skill entirely. Never execute obfuscated commands.

People who liked this also viewed

6 picks

Type