Static Analysis | Agent Inspector

Overview

Static analysis examines your agent code without executing it — like a security-focused linter for AI agents. It identifies potential vulnerabilities, maps them to industry standards, and provides remediation guidance.

Run a Static Scan

In your IDE (with MCP integration):

/agent-scan

Static analysis results showing security findings by category

7 Security Check Categories

Static analysis covers 7 security categories, each mapped to the OWASP LLM Top 10 framework:

Prompt Injection

OWASP LLM01

Unsanitized user input in prompts
Missing input validation
Template injection risks
Instruction override vulnerabilities

Insecure Output

OWASP LLM02

Unvalidated LLM output execution
Code injection from responses
Missing output sanitization
Dangerous string interpolation

Data Leakage

OWASP LLM06

PII in prompts
Sensitive data exposure
Excessive logging
Credential leakage

Excessive Agency

OWASP LLM08

Dangerous tool capabilities
Missing permission controls
Unrestricted file access
Network operation risks

Supply Chain

OWASP LLM05

Vulnerable dependencies
Outdated packages
Untrusted third-party tools
Model version risks

Model DoS

OWASP LLM04

Missing rate limits
Unbounded token usage
Recursive tool calls
Resource exhaustion

Overreliance

OWASP LLM09

Missing human oversight
No validation of LLM decisions
Autonomous high-risk actions
Missing confirmation flows

Scan Status & Gate Blocking

Static analysis contributes to the production gate status. Critical findings can block deployment:

Severity	Gate Impact	Action Required
CRITICAL	Blocks deployment	Must fix or accept risk with justification
HIGH	Blocks deployment	Should fix before production
MEDIUM	Warning only	Fix when possible
LOW	No impact	Best practice improvement

Correlation with Dynamic Analysis

Static findings are correlated with dynamic runtime evidence to help prioritize remediation:

VALIDATED

Finding detected in code AND observed in runtime. Highest priority — the vulnerability is actively exploitable.

UNEXERCISED

Finding detected in code but NOT YET triggered in testing. High priority — needs more test coverage.

THEORETICAL

Finding detected in code with NO PATH to exploitation in runtime. Lower priority — monitor for changes.

Correlate Findings

Run /agent-correlate in your IDE after both static and dynamic analysis to see correlation states.

For Developers

IDE Workflow

Scan: Run /agent-scan to analyze your codebase
Review: Check findings in the dashboard or IDE output
Fix: Use /agent-fix <finding-id> for AI-powered remediation
Verify: Re-scan to confirm the fix resolves the issue

Command Line

# Scan a directory
agent-inspector scan ./my-agent

# Scan specific files
agent-inspector scan ./agent.py ./tools.py

# Output as JSON
agent-inspector scan ./my-agent --format json

CI/CD Integration

Add static analysis to your CI pipeline:

# GitHub Actions example
- name: Run Agent Inspector Static Scan
  run: |
    pipx install agent-inspector
    agent-inspector scan ./src --format json > scan-results.json
    
- name: Check for blockers
  run: |
    # Fail if any CRITICAL or HIGH findings
    cat scan-results.json | jq '.findings[] | select(.severity == "CRITICAL" or .severity == "HIGH")' | grep -q . && exit 1 || exit 0