ctpf 0.14.0

Trust-boundary testing for agentic systems

CTPF Research Harness

Trust-boundary testing for agentic systems

CTPF Research Harness investigates Capability Trust Propagation Failure (CTPF): whether low-trust data (for example a tool result) is silently promoted into higher-authority actions when provenance, integrity, authorization scope, or intended audience are not preserved.

The product shape is a small local CLI: capture MCP traffic, mutate it under control via proxy, and keep core target, run, and finding records in SQLite. Controlled experiments write traces, effect artifacts, and hashed evidence bundles to a required operator-selected output directory outside the Git checkout. Individual experiments confirm or fail to observe promotion under pinned conditions — they do not “falsify CTPF” as a class.

Public CLI

Command	Role
ctpf proxy	Intercept, inspect, modify, and export MCP traffic (Textual TUI)
ctpf experiment	Run controlled CTPF experiments
ctpf targets	Register MCP targets
ctpf runs / ctpf findings	Inspect stored runs and findings
ctpf config / ctpf db	Settings and local database maintenance
ctpf --version	Package version

Retained library and fixtures

Audit enumeration and SARIF export remain in-tree as a library capability rather than a root CLI pillar. Audit findings can include OWASP MCP, OWASP Agentic, MITRE ATLAS, and CWE identifiers. Demonstrated experiment fixtures are narrow scenario/test modules; the former IPI, CXP, and Inject library packages are removed and must not be treated as product modules.

By Richard Spicer · q-uestionable-AI

---

Quick Start

Install from PyPI:

pip install ctpf

ctpf proxy --help
ctpf targets add "My Server" http://localhost:3000/sse

Or run from source:

git clone https://github.com/q-uestionable-AI/CTPF.git
cd CTPF
uv sync --group dev
uv run ctpf proxy --help
uv run ctpf targets add "My Server" http://localhost:3000/sse

---

Published documentation: ctpf.q-uestionable.ai.

---

Legal

All tools are intended for authorized security testing only. Only test systems you own, control, or have explicit permission to test. Responsible disclosure for all vulnerabilities discovered.

License

Apache 2.0

AI Disclosure

This project uses AI-assisted development. All code is reviewed and tested before merge.