More screenshots
Works with the AI Agents you already use
AI-Generated Code Verification Agent
Audit AI-generated code for hallucinations, security flaws, and architectural drift before you merge.
$7.99
AI-Generated Code Verification Agent
Example session with this skill installed
Review this AI-generated PR that adds a new auth provider. Check for hallucinated library calls, workspace isolation issues, and ensure the tests actually verify the token exchange flow.
- Read your context and instructions
- Compiled the ai-generated code verification
AI-Generated Code Verification Audit
Decision: Changes Required
Finding: Hallucinated API call to auth.provider.verifyOnce() in auth_service.ts. The installed version only supports verify().
Finding: Missing workspace authorization. The endpoint checks identity but not resource ownership.
Test Review: Tests use excessive mocking; they pass even if the exchange logic is deleted.
Remediation: Update to verify(), add workspace ownership middleware, and implement integration tests.
Connects securely to your tools. The creator never sees your data.
About this skill
The problem
AI assistants often produce code that looks correct but contains hallucinated APIs, security vulnerabilities, and subtle logical flaws. Blindly merging these changes introduces technical debt, breaking changes, and production risks.
What it does
- Detects hallucinated methods, imports, and configuration keys by cross-referencing repository evidence.
- Performs security audits focused on authentication bypass, injection risks, and sensitive data exposure.
- Evaluates architectural consistency to ensure AI-generated code follows existing patterns and service layers.
- Analyzes edge cases like concurrent updates, race conditions, and empty states that AI often overlooks.
- Verifies test quality to ensure test suites confirm behavior rather than just asserting mocks.
Frameworks & tools
Language-agnostic. Compatible with Cursor, Claude Code, GitHub Copilot, and standard CI/CD environments. Supports REST/GraphQL APIs, database migrations, and infrastructure-as-code.
Why this beats prompting it yourself
Generic prompts often miss environment-specific constraints and fail to verify if a library method actually exists in your lockfile. This skill follows a rigorous 4-phase verification framework that treats AI output as untrusted until proven valid against your specific codebase.
Use cases
- Reviewing complex pull requests generated by AI coding agents.
- Auditing AI-suggested refactors for hidden breaking changes in public APIs.
- Verifying database migrations for lock risks and nullability constraints.
- Checking AI-generated frontend components for accessibility and state management.
Known limitations
Cannot independently guarantee production safety without repository access, dependency info, or executed validation evidence.
Requires repository context and dependency versions to verify API claims. Findings are recommendations and require human approval for safety-critical systems.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 3 months ago
Permissions
File Scopes
This skill uses file access to inspect AI-generated source code, repository files, pull-request diffs, patches, tests, package manifests, dependency lock files, framework configuration, build files, database schemas, migrations, API definitions, CI workflows, deployment manifests, infrastructure definitions, architecture documentation, CODEOWNERS files, contribution guides, security requirements, accessibility requirements, performance reports, logs, stack traces, screenshots, issue descriptions, product specifications, AI conversation transcripts, generated implementation plans, and generated test summaries. It uses write access to create verification audits, requirement-traceability matrices, hallucinated-API reports, logical-defect findings, security findings, architecture findings, edge-case matrices, test-quality assessments, unsupported-claim reports, remediation plans, verification plans, reviewer guidance, approval decisions, structured JSON reports, and SKILL.md files. Browser access is useful when the host environment permits authorized repository access, authoritative framework-documentation review, dependency-version research, package-documentation validation, API-reference verification, security-advisory research, standards review, licensing validation, or compatibility research. The skill should clearly distinguish: Verified Requirement: Explicitly supported by the original request, acceptance criteria, approved specification, maintainer decision, or repository behavior. Repository Evidence: Confirmed through inspected source code, tests, configuration, schemas, documentation, dependency manifests, or repository conventions. Observed Defect: A problem directly supported by the generated implementation or verification evidence. Inferred Risk: A plausible problem that requires additional evidence or execution. Hallucinated Reference: A symbol, method, import, endpoint, option, command, configuration key, or behavior unsupported by the actual repository, dependency version, or authoritative documentation. Unsupported Claim: An AI-generated statement that lacks sufficient evidence. Recommended: A proposed correction, test, verification step, architecture alignment, or reviewer action. Not Run: A command, test, build, migration, scan, or verification step that has not been executed. Blocked: A conclusion or check that cannot be completed because repository access, dependency information, runtime evidence, environment access, or approval is unavailable. The agent must never invent test results, build results, command output, repository files, API behavior, dependency methods, security properties, performance results, accessibility results, migration safety, or deployment outcomes. The agent must never expose or hard-code passwords, access tokens, private keys, signing keys, connection strings, production records, customer information, payment data, or other protected information. The agent should audit only the requested change and the minimum surrounding context required to evaluate correctness, security, compatibility, maintainability, and testability. Generated code should not be approved merely because it compiles, appears polished, or includes tests. Public examples, fixtures, screenshots, and reports should use fictional, anonymized, masked, or approved data.