Skip to content

Releases

v0.1.1: renamed to blue-harness

Choose a tag to compare

@seifheakal seifheakal released this 15 Nov 16:31

The project is now called blue-harness everywhere (it was developed under the working name Carapace).

Changed

  • CLI: skill/carapace.py is now skill/blue_harness.py (make targets are unchanged).
  • Hardening modules: harness/harden/carapace_* are now harness/harden/bh_*.
  • C symbols and macros use the bh_ / BH_ prefix (for example bh_guard_coredump()). Rebuild anything linked against the old carapace_* guards.
  • Decoy and honeytoken section names changed to match (.bh_honey, .bh_exh_*, Mach-O segment __BH_EXH).
  • README demo GIF re-recorded from docs/assets/demo.tape. Same result: plain build 84/100, guarded and hardened build 40/100, still runs.

All 350 verify checks pass.

v0.1.0 — first public cut

Choose a tag to compare

@seifheakal seifheakal released this 15 Nov 15:39

First tagged release. At this tag the project still used its working name, Carapace (CLI: skill/carapace.py). It was renamed to blue-harness in v0.1.1.

A build-time audit, harden and verify harness that measures and raises a native or mobile app's resistance to reverse engineering, with a focus on the new wave of AI-assisted, one-command RE.

What's in 0.1.0

  • Exposure audits for ELF, PE and Mach-O (read-only, scored, honest tiers).
  • Post-build harden paths for all three native formats, each with a named bypass.
  • ELF build-time guards: core-dump suppression, anti-ptrace, string encryption, runtime attestation, anti-Frida.
  • Mobile reach: APK/AAB/IPA unpacked, native payloads audited and hardened, then repacked (re-signing left to the operator, and said so).
  • A Claude skill that drives audit -> recommend -> apply -> harden -> report.
  • A one-command empirical before/after run (make slice-k-run) with a sealed CrackMe and a scorer.
  • CI enforces make verify on every push.

Honest limits

  • PE and Mach-O runtime behaviour is verified structurally on Linux, not by running on Windows/macOS.
  • Anti-RE raises cost; it never makes software unbreakable. Every countermeasure names a bypass.
  • The honest #1 move is still: secrets you cannot afford to leak belong on a server, not the client.

See CHANGELOG.md for the full list.