Skip to content

Releases

v1.1.0 - fast lint + engine orchestrator + commit gate

Choose a tag to compare

@cipherc1024 cipherc1024 released this 07 Sep 05:47

Secure-Vibe v1.1.0 - repositioning release

Repositioned from "secure-by-generation guarantee" to an honest fast lint + engine orchestrator + commit gate for AI coding agents.

What's new

  • Lexical stripping (core/strip.py): python tokenize + js lexer - comments/docstrings/strings never trigger rules (0% false positives on SecurityEval prompts)
  • SAST orchestrator (core/sast.py + cli.py sast): builtin engine + semgrep + pip-audit/govulncheck/npm audit in one pass; reports which engines ran and which were skipped, and why
  • Commit gate: hooks/pre-commit + .pre-commit-config.yaml + cli.py precommit --install-hook; validates staged files before every commit, skips cleanly when no usable interpreter
  • CI gate: GitHub Actions sast job (semgrep auto-installed in CI)
  • Post-repair regression verification (core/regression.py): broken fixes auto-reverted, pytest/npm/go/cargo/make detected per ecosystem
  • Human review tickets: full markdown (context + alternatives + round history) when the repair loop fails to converge
  • 56-sample positive/negative self-test suite (core/selftest_suite.py)
  • Startup self-check: machine-specific fix commands instead of bare tracebacks; interpreter cache in config.yaml
  • Line/file-level suppression: # secure-vibe: ignore / # secure-vibe: ignore-file

Authoritative evaluation (SecurityEval, MSR4P&S'22)

Full official dataset.jsonl (121 CWE-annotated insecure samples + 121 benign prompts):

Metric Value
detection_rate 38.0%
false_positive_rate 0.0%
avg_latency_ms 1.1

Honest ceiling of a fast single-pass linter; dataflow-depth findings are delegated to the sast orchestrator. See docs/evaluation.md.

Quality

  • 255 tests passing (Python 3.9-3.14)
  • CI green: test matrix (3.10/3.11/3.12) + sast job

v1.0.0 - initial release

Choose a tag to compare

@cipherc1024 cipherc1024 released this 07 Sep 13:02

Initial release (13-language rule engine, deterministic auto-repair, JSONL audit trail, cross-agent installers). See CHANGELOG.md.