Releases
v1.2.0
Compare
Sorry, something went wrong.
No results found
Added
TestPyPI workflow for safe release testing before production
GitHub Actions workflow .github/workflows/publish-test.yml for TestPyPI publishing
Comprehensive TestPyPI testing documentation in RELEASING.md
Support for test release tags (v*-test*) to publish to TestPyPI
Manual workflow dispatch for testing workflow changes
Prompt injection detection as a new security layer in the hook flow
Heuristic-based pattern detection for common injection attacks (<1ms, local, privacy-preserving)
Configurable sensitivity levels (low, medium, high) for detection thresholds
Custom pattern support for organization-specific injection patterns
Allowlist patterns to handle false positives
Comprehensive test suite with 23 tests covering various attack patterns
Support for future ML-based detectors (Rebuff, LLM Guard)
Changed
Hook flow now includes prompt injection detection between directory check and secret scanning
Updated security architecture diagram in README.md
Security
CRITICAL : Added prompt injection detection to protect against manipulation attacks
Detects instruction override, system mode changes, prompt exfiltration, safety bypasses
Patterns include: "ignore previous instructions", "developer mode", "reveal prompt", etc.
Fail-open design maintains availability if detection encounters errors
Detection runs before AI receives prompts, providing proactive protection
You can’t perform that action at this time.