The undocumented in-code default (1,000,000) let OMEN finish in ~2s
when config.json omitted the key, far below the 50,000,000 shipped in
config.json.example. Align both to 100,000,000 per issue #145.
ruff 0.16.0 broadened the default rule set and reports 317 findings
across the tree (65 auto-fixable) on code that 0.15.1 accepted
cleanly. Explicitly select the previous defaults ("E4", "E7", "E9",
"F") so this dep bump is behaviour-neutral; enabling the new rules
belongs in its own cleanup, not bundled into the version bump (per
the project's dep-bump convention).
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
ty 0.0.63 changed two things that broke CI on this bump:
- new rules (unsupported-operator, no-matching-overload,
invalid-assignment, unresolved-global) are error-severity by
default and fire on the same legacy module-globals pattern in
hate_crack/main.py the existing "warn" rules were pinned for;
downgrade them to match.
- default exit behaviour now fails on warning-level diagnostics;
pass --exit-zero-on-warning to ty check in CI and in the pre-push
prek hook to restore the prior fail-only-on-errors semantics.
No source changes; deferring the main.py globals refactor.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
The main-menu block still showed the pre-consolidation numbering (LLM at 15,
OMEN at 16) and was missing Notifications, PCFG, and PRINCE-LING entirely.
Verified the block now matches get_main_menu_items() exactly.
Also documents features that shipped without README coverage:
- N-gram, PCFG, and PRINCE-LING attack sections
- pcfgRuleset / pcfgMaxCandidates / pcfgPrinceLingMaxCandidates config keys
- pcfg_cracker in the submodule and asset lists
- llm, menu, noninteractive, notify, username_detect, formatting, and
progress modules in Project Structure
- LLM attack: structured output, target research pre-fill, spinner, timeout
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The repo had no secret-scanning gate of any kind: prek.toml carried no
detect-private-key/detect-secrets hook, CI had none, and bandit only
scans hate_crack/ so it never inspected config files, docs, or fixtures
for committed key material.
Adds the detect-private-key hook from the pre-commit-hooks remote repo
at the pre-commit stage, and records v2.14.3 in the changelog alongside
the removal of the published agent tooling.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
CLAUDE.md, the .claude/ directory, and the agent-generated planning docs
under docs/plans/ and docs/superpowers/ are local development aids, not
part of the shipped project. Untrack them and add them to .gitignore so
they stay local.
Drops the post-commit audit-docs hook from prek.toml, since the script it
invoked (.claude/audit-docs.sh) is no longer part of the repo.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
pipal() built one rigid regex requiring exactly pipal_count consecutive
base-word lines, so any cracked set with fewer unique base words than
pipal_count (default 10) matched nothing and returned []. Parse the
"Top N base words" section line by line instead, collecting up to
pipal_count words and stopping at the blank line that ends the section.
Also replace the shell=True string-formatted Popen with list-form
arguments so file paths containing shell metacharacters can't be
interpreted as commands.
Adds tests/test_pipal_e2e.py: hermetic end-to-end tests that drive the
real pipal() through a fake pipal executable (parsing, $HEX decode,
fewer-basewords regression, and injection safety).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The helper is shared by the OMEN attack and the LLM wordlist-mode attack,
so the OMEN-specific name was misleading.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The "p. Enter a custom path" branches of the OMEN and Markov training
pickers, the combipow wordlist prompt, and the rule cleanup/optimize
output-path prompts used a bare input() with no readline completer, so
TAB did nothing. Route them through select_file_with_autocomplete.
Also drop the path completer after each selection so later numeric-menu
and y/n prompts don't inherit stale file-path tab completion.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Adds the gates the hashview repo runs, adapted to hate_crack's uv toolchain:
- ruff format --check as a CI step + a ruff-format prek pre-push hook
(the package was reformatted in the preceding commit).
- Bandit SAST vs a committed baseline (.bandit-baseline.json, 114 reviewed
low-severity subprocess/shlex findings); [tool.bandit] config in
pyproject.toml; CI job + prek pre-push hook. Only NEW findings fail.
- pip-audit dependency-CVE gate (CI job). PYSEC-2026-2447 (diskcache 5.6.3,
transitive via instructor -> atomic-agents) is ignored — no upstream fix.
- Hygiene pre-commit hooks (trailing-whitespace, end-of-file-fixer, check-yaml,
check-merge-conflict, check-added-large-files) via the pre-commit-hooks repo.
- CLAUDE.md updated: lint/security commands, prek install (+pre-commit hook
type), and the active-hooks list.
Co-Authored-By: Claude <noreply@anthropic.com>
Normalizes formatting across the package so `ruff format --check` (added to
CI in a following commit) passes. No behavioral changes.
Co-Authored-By: Claude <noreply@anthropic.com>
The pre-dispatch 'check POT file' step in main() shells out to the real
hashcat binary, which is absent in CI, causing the four test_main_* tests
to fail with FileNotFoundError. Stub _run_hashcat_show in the shared
_run_main helper — these tests verify dispatch routing and prompt
suppression, not potfile recovery.
Co-Authored-By: Claude <noreply@anthropic.com>
- Change quick subparser --rules to dest=rule_files to prevent collision with the top-level --rules=store_true Hashmob download flag
- Update run_noninteractive to read args.rule_files instead of args.rules
- Move ATTACK_COMMANDS above run_noninteractive (Fix 5)
- Fix has_attack_subcommand to scan all argv elements (supports leading global flags like --debug)
- Replace raw input() dedup prompt with _auto_input() so non-interactive runs don't block
- Update existing quick dispatch tests to use rule_files= namespace key
- Add test_main_quick_with_rules_dispatches: proves --rules routes to crack, not download
- Add test_main_debug_flag_before_subcommand: proves leading global flags don't break routing
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Appends `run_noninteractive(ctx, args)` to noninteractive.py, which
dispatches quick/dict/brute/topmask commands to the appropriate hcat*
function. Returns 0 on success, 1 on bad inputs, 2 on unknown command.
Includes 6 new unit tests (12 total in file), all passing.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
ollamaUrl was built as "http://" + OLLAMA_HOST, so any value carrying a scheme
- the form Ollama's own tooling accepts - was mangled:
OLLAMA_HOST=https://ollama.example.com
-> http://https://ollama.example.com
The LLM attack then could not connect, and reaching a remote Ollama over TLS
was impossible. llm.py derives its client base URL as f"{ollamaUrl}/v1", so the
malformation propagated into the OpenAI-compatible client.
Extract _normalize_ollama_url() so the logic is unit-testable - ollamaUrl is
assigned at module import, which is otherwise only reachable via reload - and
prepend the scheme only when absent. Also strip trailing slashes, since callers
append paths. The bare host:port default is unchanged.
Fixes#119
Co-Authored-By: Claude <noreply@anthropic.com>
The UI polish, cracked-password mode, and target research were all written
into the [2.12.0] section on the assumption they would ship in that release.
They did not: auto-tag cuts a release per merge that contains feat/fix
commits, so #129 became v2.12.0 and #132 became v2.13.0.
Move those entries into a new [2.13.0] section so each heading lists what its
tag actually contains. No entry is duplicated or dropped.
Verified against 'git log v2.11.4..v2.12.0' and 'git log v2.12.0..v2.13.0'.
Co-Authored-By: Claude <noreply@anthropic.com>
2.12.0 is not yet tagged (latest tag is v2.11.4), so this work folds into that
entry rather than cutting a new version.
Co-Authored-By: Claude <noreply@anthropic.com>
- Remove [*] prefix from all input() prompts (10 sites); [*] is a
status/output marker and must not appear in user-facing input prompts.
print() calls that use [*] are untouched.
- Normalize default-value hints to the bare-parentheses form (7 instances
already used it vs. 1 using "default N"); the lone outlier
"\nEnter n-gram group size (default 3): " is rewritten to "(3)".
- Add \n prefix to combipow "Add spaces between words?" prompt so it does
not run flush against the last output line of the wordlist path loop.
- All other prompts already satisfy: capital first letter, ends with ": ",
and correct leading-whitespace context (tab-indented where the surrounding
menu uses tabs, plain \n or none where it follows un-indented output).
- No input() calls added or removed (verified: attacks.py 42→42,
main.py 26→26).
Co-Authored-By: Claude <noreply@anthropic.com>
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>
The re-prompt loops repainted the whole multicolumn wordlist grid after every
typo, burying the error message under dozens of entries. Hoist the grid and the
p/q legend out of the loop so only the prompt repeats.
ollama_attack's loop also fell through to a silent redraw on an unrecognised
key, leaving no way to tell a rejected key from a repainted prompt.
Co-Authored-By: Claude <noreply@anthropic.com>
- ollama_attack: convert generation-mode menu to interactive_menu with
dynamic items list (option 3 only present when cracked file exists);
cancel via 99 or Escape; re-prompts in a while loop
- omen_attack: convert existing-model menu to interactive_menu; cancel
via 99 or Escape (replaces hard-coded "3. Cancel"); re-prompts in loop
- _omen_pick_training_wordlist: replace abort-on-invalid with re-prompt
loop; add explicit 'q' cancel key so users can exit without being
trapped; callers already handle None correctly
- _markov_pick_training_source: same re-prompt-loop + 'q' cancel fix;
callers already handle None correctly
- Update all affected tests to patch interactive_menu at
hate_crack.attacks.interactive_menu (module-level import path) and
drive follow-up prompts via builtins.input
- Add new tests: arrow-menu reach, Escape/99 cancel, re-prompt loop
coverage for both pickers, and conditional option-3 key presence
Co-Authored-By: Claude <noreply@anthropic.com>
Adds a third LLM Attack generation mode ("cracked") that samples the
plaintexts already recovered this session from <hashfile>.out and asks the
model to infer the target organization's password conventions and emit new
candidates in the same style.
- llm.py: new _CRACKED_PROMPT (offensive candidate generation, explicitly
not the denylist-oriented _WORDLIST_PROMPT), a _PROMPTS mode->prompt map,
and a "cracked" branch in _build_request that tells the model not to
repeat already-cracked passwords.
- main.py: extract the wordlist sampling logic into the shared module-level
helper _sample_plaintext_file(path, cap, source_label) and call it from
both the wordlist and cracked branches of hcatOllama; guard missing and
empty .out files.
- attacks.py: offer mode 3 only when <hashfile>.out exists and is non-empty
(matching _markov_pick_training_source), with a clear message otherwise.
- README: document the three modes and the widened ollamaMaxSampleLines
scope.
Co-Authored-By: Claude <noreply@anthropic.com>
