Changelog¶
All notable changes to HydraSight are documented here. The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.
The canonical changelog is
CHANGELOG.md
in the repository root.
4.1.1¶
Fixed¶
helpdocumentsauthorize,roe,resume, and theexit / quitsynonym — newAUTHORIZATIONsection leads the reference.
Changed¶
- Bridge setup for current Kali packaging:
sudo apt install mcp-kali-server, start withkali-server-mcp.
4.1.0 — OBSIDIAN¶
Security & accountability¶
- Mandatory authorization attestation before any scan (
authorize <cidr>→ typeI AUTHORIZE, or a pre-signedhydrasight.authorization.jsonfor CI). Enforced at theDispatcherchokepoint and across shell execution paths; deny by default. - Tamper-evident audit trail — append-only, SHA-256 hash-chained JSONL with
AuditLogger.verify()integrity checking and automatic secret redaction. - ROE ∩ authorization scope enforcement — a target must satisfy both the Rules-of-Engagement envelope and the operator attestation; ROE kill-switch blocks dispatch.
Packaging & onboarding¶
- Single-sourced version and full PyPI metadata;
python -m buildproduces sdist + wheel. - Non-root Docker image + docker-compose (Kali MCP backend) + Makefile.
- MkDocs Material documentation site (builds
--strict); CHANGELOG; CODEOWNERS. - CI: ruff + mypy (strict function contracts) + pylint + pytest with a 75%
branch-coverage floor and a strict docs job.
release.ymlpublishes to PyPI via trusted publisher onv*tags and publishes a GitHub Release. - One-line install:
pip install hydrasight— the PyPI project was auto-created by the first trusted-publishing run (no manual registration).
Reliability & architecture¶
- Broad
except Exceptionat network/IO boundaries narrowed to specific types. post_access.pydecomposed into a focused package (import path preserved); hash-crack logic extracted tocore/hash_crack.py; engagement-outcome classification extracted toreporting/outcome.py.- Fixed:
check_targetfalse-positive on "100% packet loss";save_jsonleakingValueError.
4.0.0¶
- Initial professional release: AI-orchestrated engagement engine, command sanitizer, Rules of Engagement, NL mode separation, JSON/PDF reporting.