Target-info mode (menu 12) asked for company, industry, and location as three
blank prompts. Once the company name is known the local Ollama model can often
supply the other two, so ask it and offer the answers as editable defaults.
- llm.research_target(): TargetResearchInput/TargetResearchOutput schemas plus a
research SystemPromptGenerator that tells the model to return empty strings
when it does not genuinely recognize the organization, so an unknown small
client yields blank prompts instead of a confident hallucination.
APITimeoutError is translated to LLMTimeoutError like generate_candidates.
- clean_research_field(): strips, collapses whitespace, caps at 80 chars, and
turns anything non-string into "" so model output cannot be pasted unbounded
into a prompt default.
- main.hcatOllamaResearchTarget(): runs the call inside the existing spinner and
degrades to blank suggestions on timeout or any other failure, so research can
never block the attack.
- attacks.ollama_attack(): shows suggestions as "Industry (freight rail): "
defaults, labelled explicitly as the model's GUESSES rather than OSINT.
- New ollamaAutoResearch config toggle (default true) in both config examples.
Research uses only the configured local Ollama server; the client name is never
sent to a search engine or third-party company-data API.
Co-Authored-By: Claude <noreply@anthropic.com>
Defect 1 — hcatOllama showed a plain print then blocked silently for
up to 300 s while waiting for the LLM. A new generic context manager
`hate_crack/progress.py::spinner()` now runs a daemon thread that
repaints a single line with a frame + elapsed-seconds counter every
~120 ms. TTY guard ensures non-TTY/test environments get a single
plain print instead of ANSI control characters. The line is erased
with \033[2K\r on exit (normal and exceptional).
Defect 2 — the wordlist path materialised every line in memory and
built a prompt that could massively overflow ollamaNumCtx on large
wordlists. The path now does a two-pass evenly-spaced sample: pass 1
counts usable lines, pass 2 stride-selects up to `ollamaMaxSampleLines`
(default 500, new config key) spread across the full file. When no
capping occurs the message reads "Loaded N passwords"; when capping
occurs it reads "Sampled N of M passwords".
Co-Authored-By: Claude <noreply@anthropic.com>
The Atomic Agents client was built with no timeout, so an Ollama server that
accepted the TCP connection but never replied (most commonly a large model still
loading into VRAM) left the CLI blocked in agent.run() forever. The caller's
`except Exception` handler only ever fired on connection-refused.
- llm.generate_candidates() takes a `timeout` parameter (default
DEFAULT_TIMEOUT_SECONDS = 300.0) and forwards it to the OpenAI client.
- openai.APITimeoutError is translated into a new domain-level
llm.LLMTimeoutError so main.py need not import openai itself, keeping the
atomic-agents/instructor dependency isolated to llm.py as documented.
- hcatOllama passes the new `ollamaTimeout` config value and prints
timeout-specific guidance (elapsed seconds, VRAM-loading hint, the setting to
raise) instead of the misleading "ensure Ollama is running" message.
- Documented `ollamaTimeout` in config.json.example and README.
Also closes two test gaps: the defensive `except ValueError` handler in
hcatOllama now has coverage, and the misleadingly-named `test_unknown_mode_raises`
is split into explicit unit tests of _build_request's mode validation. Ticked the
completed steps in the LLM Atomic Agents plan doc.
Co-Authored-By: Claude <noreply@anthropic.com>
Adds notify_* keys to both config.json.example files, threads
notification calls through hashcat invocations in main.py, and
exposes menu/attack hooks. Pushed for manual testing — verification
and PR still pending.
Refs #106
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Re-add hcatOptimizedWordlists config key (previously removed) with a
default of ./optimized_wordlists. Falls back to hcatWordlists if the
configured directory does not exist.
Update quick_crack to list files from and default to hcatOptimizedWordlists
instead of hcatWordlists when prompting for a wordlist or directory.
- Apply os.path.expanduser() to hcatPath so tilde (~) paths work,
consistent with every other config path (hcatWordlists, rulesDirectory,
hcatPotfilePath, hcatDebugLogPath)
- Improve error message when binary not found to show what hcatPath was
checked, making it easier to diagnose misconfiguration
- Update optimizedKernelAttacks in config.json.example to include all 21
attacks from DEFAULT_OPTIMIZED_ATTACKS (13 were added in v2.4.0 but
config.json.example was not updated, causing the runtime override to
exclude them)
- Merge combinator, combinator3, and combinatorX into one unified
combinator_crack function that routes by wordlist count:
2 (no sep) -> hcatCombination, 3 (no sep) -> hcatCombinator3,
4+ or any separator -> hcatCombinatorX
- Replace comma-separated wordlist input with one-at-a-time
tab-autocomplete prompts (blank line to finish)
- Add _prompt_wordlist_paths helper using existing readline infrastructure
- Add hcatCombinator3Wordlist and hcatCombinatorXWordlist config vars
with rockyou.txt defaults
- Print full hashcat command to stdout in --debug mode by calling
_debug_cmd at the end of _append_potfile_arg (covers all 27 invocations)
- Collapse combinator submenu from 6 options to 4; keep combinator3_crack,
combinatorX_crack, and combinator_3plus_crack as delegation shims
- Update tests to cover unified routing and new prompt interface
Extract _select_rules() helper from quick_crack/loopback_attack and wire
it into omen_attack so OMEN can run with rule chains. Extend hcatOmen()
to accept and apply an hcatChains argument including debug mode injection.
Fix resolve_path() to honour HATE_CRACK_ORIG_CWD (set by the install shim)
so relative hash/wordlist paths resolve against the caller's working directory
instead of the repo root. Increase default omenMaxCandidates to 50M.
- Add hashcat as git submodule; compile with make, skip if already in PATH
- Convert princeprocessor from tracked files to git submodule
- Change .DEFAULT_GOAL to install so plain `make` does a full install
- Install uv, Xcode CLT (macOS), build-essential (Debian) if missing
- vendor-assets falls back to system hashcat if submodule not compiled
- Remove hcatOptimizedWordlists; all attacks now use hcatWordlists
- Default hcatWordlists to ./wordlists, rules_directory to ./hashcat/rules
- Default hcatTuning to empty string (no --force --remove)
- Backfill missing config.json keys from config.json.example at startup
- Wrap hcatBin/hcatTuning/hcatWordlists loading in try/except with defaults
- Fall back to vendored hashcat binary at hate_path/hashcat/hashcat
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Add ability to fine-tune PassGPT models on custom password wordlists.
Models save locally to ~/.hate_crack/passgpt/ with no data uploaded to
HuggingFace (push_to_hub=False, HF_HUB_DISABLE_TELEMETRY=1). The
PassGPT menu now shows available models (default + local fine-tuned)
and a training option. Adds datasets to [ml] deps and passgptTrainingList
config key.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add PassGPT as attack mode 17, using a GPT-2 model trained on leaked
password datasets to generate candidate passwords. The generator pipes
candidates to hashcat via stdin, matching the existing OMEN pipe pattern.
- Add standalone generator module (python -m hate_crack.passgpt_generate)
- Add [ml] optional dependency group (torch, transformers)
- Add config keys: passgptModel, passgptMaxCandidates, passgptBatchSize
- Wire up menu entries in main.py, attacks.py, and hate_crack.py
- Auto-detect GPU (CUDA/MPS) with CPU fallback
- Add unit tests for pipe construction, handler, and ML deps check
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add OMEN (Ordered Markov ENumerator) as a probability-ordered password
candidate generator. Trains n-gram models on leaked passwords via
createNG, then pipes candidates from enumNG into hashcat.
Also fix a pre-existing bug where ensure_binary() used quit(1) instead
of sys.exit(1) - quit() closes stdin before raising SystemExit, which
caused "ValueError: I/O operation on closed file" when any optional
binary check failed and the program continued to use input().
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Remove ollamaWordlist config key and all references. Wordlist mode now
requires the cracked hashes .out file to exist and extracts passwords
by splitting on the first colon. Detect Ollama refusal responses and
abort gracefully. Update tests accordingly.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Remove 500-line wordlist cap and send the entire file to Ollama.
Add ollamaNumCtx config key (default 32768) to control the context
window size. Invert wordlist prompt to default-yes, remove unused
ollamaCandidateCount config.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Rename markov_attack → ollama_attack and hcatMarkov → hcatOllama across
menu, attacks, and tests. Remove candidate count prompts and cracked-output
default wordlist logic. Rename config keys (markov* → ollama*) and drop
ollamaUrl. Fix Dockerfile.test to use granular build steps.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add a new attack mode that uses a local LLM via Ollama to generate
password candidates, converts them into hashcat .hcstat2 Markov
statistics via hcstat2gen, and runs a Markov-enhanced mask attack.
Two generation sub-modes:
- Wordlist-based: feeds sample from an existing wordlist to the LLM
as pattern context (config-selectable default with Y/N override)
- Target-based: prompts for company name, industry, and location
for contextual password generation
Pipeline: Ollama API -> candidate file -> hcstat2gen -> LZMA compress
-> hashcat -a 3 --markov-hcstat2
Config additions: ollamaUrl, ollamaModel, markovCandidateCount,
markovWordlist. No new pip dependencies (uses stdlib urllib/lzma).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
When hcatBin is a relative name (e.g. "hashcat"), construct the full
path by joining hcatPath and hcatBin so the correct hashcat binary is
used instead of relying on PATH resolution.
