Add agchk — AI agent architecture audit tool

[huangrichao2020]

Project

• PyPI: https://pypi.org/project/agchk/
• GitHub: https://github.com/huangrichao2020/agchk
• Python-first: 100% Python, zero dependencies (stdlib only)

Description

agchk audits any AI agent system or LLM-integrated project for 7 categories of architectural failures:

Scanner	Severity	What It Catches
Hardcoded Secrets	critical	API keys, tokens, credentials in source code
Tool Enforcement Gap	high	"Must use tool X" in prompt but no code validation
Hidden LLM Calls	high	Secret second-pass LLM calls in fallback/repair loops
Unrestricted Code Execution	critical	exec(), eval(), subprocess(shell=True) without sandbox
Memory Pattern Issues	medium	Unbounded context growth, missing TTL, no retention policy
Output Pipeline Mutation	medium	Response transformation corrupting correct answers
Missing Observability	medium	No tracing, logging, or cost tracking

Usage

pip install agchk
agchk /path/to/your/agent/project

from agchk import run_audit, generate_report
results = run_audit("/path/to/your/agent/project")
print(generate_report(results))

Why It Belongs

As AI agent frameworks grow rapidly, developers need tools to audit their architectures for security holes, reliability issues, and hidden failures. agchk fills this gap.

Integrated into: Langflow, GenericAgent, superpowers, Everything Claude Code, OpenCode, Anthropic Skills.