Advent Prompt Pwn

Spec sheet

Purpose

Advent Prompt Pwn is a typed Python toolkit for authorized adversarial testing of language models and AI applications.Read 5 Oct 2026 · adventcybersecurity.com

Platforms

Installation instructions specify Python 3.10 or newer and show use from PowerShell, macOS, or Linux shells.Read 5 Oct 2026 · adventcybersecurity.com

Integrations

Targets include custom HTTP JSON APIs, OpenAI-compatible APIs, Ollama, Python callbacks, and preview adapters for OpenAI, Azure OpenAI, Anthropic, and Gemini.Read 5 Oct 2026 · adventcybersecurity.com

License

The project’s GitHub page identifies the source as Apache License 2.0.Read 5 Oct 2026 · github.com

Maker

The product is maintained by Advent Cybersecurity LLC; the maker page identifies Jashid Sany as its founder.Read 5 Oct 2026 · github.com

Assessment workflow

It connects attack case generation to explicit scope, bounded execution, deterministic oracles, resumable checkpoints, reporting, and regression comparison.Read 5 Oct 2026 · adventcybersecurity.com

Attack coverage

Strategies cover direct, instruction override, delimiter, encoding, role confusion, indirect injection, and multi-turn attacks, with mutation and RAG poisoning listed as 1.1 preview features.Read 5 Oct 2026 · adventcybersecurity.com

Attack features

Documented strategies include direct, instruction override, delimiter, encoding, role confusion, indirect injection, and multi-turn testing.Read 5 Oct 2026 · adventcybersecurity.com

Availability

Installation requires Python 3.10 or newer and is documented for PowerShell, macOS, or Linux shells.Read 5 Oct 2026 · adventcybersecurity.com

Evidence and reports

It produces JSON, JSONL, Markdown, HTML, JUnit, and SARIF reports, plus evidence bundles with checksums and per-attempt SHA-256 hashes.Read 5 Oct 2026 · adventcybersecurity.com

Evidence handling

Evidence bundles include a checksum manifest and per-attempt SHA-256 hashes; the documentation says to treat model output as confidential assessment material.Read 5 Oct 2026 · adventcybersecurity.com

Known limitations

Oracle matches require practitioner review, redaction may miss sensitive values, and the static tool sandbox is not a production agent runtime.Read 5 Oct 2026 · adventcybersecurity.com

Known limits

Oracle matches require practitioner review, redaction may miss sensitive values, and the static tool sandbox is not a production agent runtime.Read 5 Oct 2026 · adventcybersecurity.com

Maturity

Version 1.0.0 is the stable public release; the 1.1 source line remains beta pending independent practitioner review and field assessment.Read 5 Oct 2026 · adventcybersecurity.com

Outcome checks

Built-in oracles can check strings, canary leaks, regular expressions, JSON keys and paths, and tool calls, including combined conditions.Read 5 Oct 2026 · adventcybersecurity.com

Python API

The toolkit provides a Python API for configuring cases, targets, scope, runs, and reports.Read 5 Oct 2026 · adventcybersecurity.com

Reports

Supported report formats are JSON, JSONL, Markdown, HTML, JUnit, and SARIF.Read 5 Oct 2026 · adventcybersecurity.com

Scope controls

Remote assessments support exact host and port allowlists, authorization windows, DNS controls, and request, rate, concurrency, retry, timeout, and evidence limits.Read 5 Oct 2026 · adventcybersecurity.com

Security controls

Remote assessments can be bounded by exact host and port grants, DNS checks, authorization windows, request limits, and workload ceilings.Read 5 Oct 2026 · adventcybersecurity.com

Security handling

Credentials are referenced by environment-variable names, and built-in HTTP clients disable redirects, ignore ambient proxy variables, and bound response bytes.Read 5 Oct 2026 · adventcybersecurity.com

Support contact

The documentation lists [email protected].Read 5 Oct 2026 · adventcybersecurity.com

Targets and integrations

Targets include custom HTTP JSON, OpenAI-compatible APIs, Ollama, and preview integrations for OpenAI, Azure OpenAI, Anthropic, and Gemini.Read 5 Oct 2026 · adventcybersecurity.com

Use restriction

The maker says to use the toolkit only on owned systems, intentionally vulnerable labs, public research benchmarks, or targets with explicit written authorization.Read 5 Oct 2026 · adventcybersecurity.com

Where Specifiction places it

• AI Red Teaming Tools#6 of 278.0