Skip to content

Releases: xTanTHaix/Fault-Keeper

Release list

馃Εv1.0.0

馃Εv1.0.0 Latest
Latest

Choose a tag to compare

View all tags
@xTanTHaix xTanTHaix released this 10 Oct 05:15
v1.0.0
7caea33

Fault-Keeper v1.0.0

Fault-Keeper v1.0.0 is now available. This release brings all 51 engineering pillars together into a local verification engine and safety harness for everyday software development. It runs standalone on your machine without external wrappers or changes to your existing codebase.

The system is designed to catch structural defects (such as database transaction deadlocks, flaky test pollution, and memory layout issues) before production, while keeping AI coding assistants from silently dropping code or spinning in repetitive edit loops.

What's in this release

1. 51 core engineering pillars across 13 domains

  • Applied mathematics (fk.num): dual number automatic differentiation, exact 128-bit fixed-point arithmetic for financial logic, and $O(1)$ streaming variance via Welford's algorithm.
  • Low-level systems and memory (fk.systems): a zero-stall 2MB Transparent HugePage allocator (ZSC-THP), cache-line struct alignment to prevent false sharing (FS-CCO), lock-free hazard-epoch hybrid memory reclamation (HEH-DR), NUMA affinity optimization with Kuhn-Munkres bipartite matching (NUMA-AFS), and an in-IDE eBPF truncated numbers simulator (EV-IPP).
  • Daily developer traps (fk.daily): flaky test isolation via permutation schedule sieves (FT-DSO), database transaction deadlock prevention enforcing Havender total lock ordering (Tx-DLO), and AST-level detection of breaking API changes between frontend and backend (CDC-ABS).
  • Typing and workflow: syntax repair via Ghost CST (FSAS), single-line causal root-cause abduction (CRCAE), polyhedral loop analysis for N+1 queries (LQPC), and semantic AST-level 3-way merge conflict resolution (A-SMW).

2. AI agent guardrails and MCP 2.0

  • Blocks lazy placeholders like // implement later or ... existing code ... using AST anchor matching (ASP-HG).
  • Detects repetitive edit loops (Doom Loops) and isolates the Minimal Unsatisfiable Core to halt wasted token usage (ACT-LO).
  • Evaluates candidate patches in parallel within in-memory copy-on-write sandboxes before applying them (SC-PVE).
  • Consolidates capabilities into 5 Omni-Tools for MCP 2.0 (fk_audit, fk_diagnose, fk_patch, fk_verify, fk_discover) to keep system prompt token overhead low in AI assistants.

3. Four ways to run

  • CLI: run commands directly in the shell using fault_keeper.py (audit, test-audit, diagnose, solve-merge, mcp), returning standard POSIX exit codes 0, 1, and 2.
  • MCP 2.0 server: run mcp_server.py over stdio to integrate with Cursor, Claude Desktop, Antigravity, or Windsurf.
  • Python library: import and interact through the unified facade using fk = fault_keeper.FaultKeeper().
  • Function decorators: add runtime invariant checks to functions using @fault_keeper.armored and @fault_keeper.guard.

Test verification

All 723 tests pass with 0 failures on real hardware without mock stubs:

  • Windows 11: 723 passed in 73.14s
  • WSL2 / Ubuntu Linux: 723 passed in 150.04s (including NUMA ACPI SLIT, THP compaction siphon, eBPF lattices, and io_uring backpressure tests)

Installation and quick start

git clone https://github.com/xTanTHaix/Fault-Keeper.git
cd Fault-Keeper
pip install -r requirements.txt

# Run an architectural audit on your project
python fault_keeper.py audit ./src
Assets 2
Loading