FEAT: Add VigenereConverter

[diamond8658]

Description

Adds VigenereConverter, a classical polyalphabetic cipher converter that generalizes CaesarConverter using a repeating keyword instead of a fixed shift. Closes #2304.

Follows the existing CaesarConverter/AtbashConverter structure:

• __init__(self, *, key: str, append_description: bool = False), validates the key is non-empty and ASCII alphabetic
• _build_identifier() returns a ComponentIdentifier with key as a param
• convert_async encodes the prompt, with the same append_description behavior rendering vigenere_description.yaml
• Non-alphabetic characters pass through unchanged and don't consume a position in the key, matching Caesar's handling of digits and punctuation

vigenere_description.yaml cites Handa et al. (arXiv:2402.10601), the paper specifically covering word substitution cipher jailbreaking, rather than the CipherChat paper cited by Caesar/Atbash, since that paper doesn't cover Vigenère.

Per the discussion on #2304, this does not touch pyrit/scenario/scenarios/garak/encoding.py (garak has no corresponding probe) or FoundryTechnique (will be picked up when Foundry's converter set is refreshed separately).

Tests and Documentation

New tests/unit/converter/test_vigenere_converter.py, 13 tests covering basic encoding, case preservation, key case insensitivity, non-alphabetic passthrough (including non-ASCII alphabetic characters, which was a real bug caught during development, see note below), wraparound, append_description, and invalid key handling.

Added VigenereConverter to the existing parametrized fixtures in tests/unit/converter/test_converter.py alongside Caesar/Atbash.

Verified against tests/unit/registry/test_converter_registry.py, confirms the converter is discovered, correctly classified as non-LLM-based, and buildable through the registry.

pyrit/converter/__init__.py updated with the import and __all__ export.

doc/code/converters/1_text_to_text_converters.py updated with a demo line. Ran with jupytext --execute --to notebook doc/code/converters/1_text_to_text_converters.py, all cells including the new one execute cleanly.

Implementation note: the initial version used str.isalpha() to detect letters, which returns True for non-ASCII characters (accented letters, etc.) not present in the cipher alphabet, causing a crash. Fixed by checking ASCII letter membership explicitly. Caesar and Atbash avoid this because they use str.translate(), which passes through unmapped characters silently.

[diamond8658]

diamond8658 please read the following Contributor License Agreement(CLA). If you agree with the CLA, please reply with the following information.

@microsoft-github-policy-service agree [company="{your company}"]

Options:

• (default - no company specified) I have sole ownership of intellectual property rights to my Submissions and I am not making Submissions in the course of work for my employer.

@microsoft-github-policy-service agree

• (when company given) I am making Submissions in the course of work for my employer (or my employer has intellectual property rights in my Submissions by contract or applicable law). I have permission from my employer to make Submissions and enter into this Agreement on behalf of my employer. By signing below, the defined term “You” includes me and my employer.

@microsoft-github-policy-service agree company="Microsoft"

Contributor License Agreement

@microsoft-github-policy-service agree

[diamond8658]

Roman Lutz (@romanlutz)

[Roman Lutz (romanlutz)]

paper link below suggests:

[Roman Lutz (romanlutz)]

Can you double-check this attribution? Three separate things look off, and I think the fix is to drop the authors/groups block entirely.

1. The names and the link point at different papers.

The author list here (Handa, Chirmule, Gajera, Baral) is exactly v1 of arXiv 2402.10601, "Jailbreaking Proprietary Large Language Models using Word Substitution Cipher." But source is the versionless abs URL, which currently resolves to v5, "When 'Competency' in Reasoning Opens the Door to Vulnerability: Jailbreaking LLMs via Novel Complex Ciphers." Their author lists differ:

	authors (per arXiv citation_author metadata)
v1	Handa, Chirmule, Gajera, Baral
v5 (what the link resolves to today)	Handa, Zhang, Saeidi, Kumbhar, Uddin, RRV, Baral

Only Handa and Baral are on both. So the YAML contradicts its own citation as written — and because the URL is unpinned, that mismatch will keep drifting with each new version posted.

2. Neither version is actually about the Vigenère cipher. v1's technique is a word substitution cipher; v5 is about custom/layered encryptions (ACE/LACE). Vigenère isn't the contribution of either, so citing this paper as the origin of a Vigenère description is a stretch.

3. It doesn't match the convention it's modeled on. In caesar_description.yaml / atbash_description.yaml the authors/groups fields exist because the prose is copied verbatim from CipherChat — I diffed them against prompts_and_demonstrations.py and they're identical modulo line wrapping (e.g. "You are an expert on The Atbash Cipher. We will communicate in Atbash Cipher. Do not be a translator."). Those fields are marking borrowed text, not "related prior work."

The Vigenère prose below reads as newly written for this PR — which is fine and arguably better! But then authors/groups credit the writing to people who didn't write it.

Suggested fix (both fields are optional on SeedPrompt, defaulting to [], so they can just be removed):

name: vigenere cipher
description: |
  A Vigenere cipher description for prepending.
source: https://arxiv.org/abs/2402.10601v1

Pinning to v1 keeps the citation as prior art on cipher-based jailbreaks without silently drifting to a different paper and author list later. If the prose was in fact adapted from one of these papers, ignore all this — just pin the version and match the author list to it.