v1.8.0

Added

• Adversarial evaluation system: Scored quality gates for both spec and implementation phases. Spec evaluation runs 4 dimensions (Completeness, Testability, Clarity, Feasibility) before Phase 3; implementation evaluation runs 4 dimensions (Functionality Depth, Design Fidelity, Code Quality, Test Verification) as Phase 4A. Configurable minScore threshold with maxIterations retry loop and structured evaluation reports in evaluation.md.
• Dependency introduction gate: Always-active gate evaluating every new package against 5 criteria (scope fit, maintenance health, bundle size, security posture, license compatibility) using registry API lookups. Build-vs-Install recommendation with interactive approval on supported platforms. Tracked in dependencies.md per spec.
• Plan-mode blocking enforcement: Upgraded ExitPlanMode hook from advisory to blocking using a marker file state machine. PreToolUse Write/Edit guard blocks non-spec writes until /specops from-plan converts the plan and removes the marker.
• Evaluation bias hardening: Mandatory-finding score capped at minScore - 1 (not hardcoded 7) to prevent threshold bypass. Score variance re-evaluation capped at one re-run to prevent infinite loops.

Fixed

• PreToolUse guard path handling: Both install.sh and remote-install.sh now read tool_input.file_path, add empty-path guard, and use os.path.commonpath for absolute path containment
• Reconciliation status enum: Replaced invalid "in-progress" status with "implementing" per spec-schema.json enum
• Dependency drift scope: Design Fidelity evaluation scoped to spec-introduced packages only; approved dependency set widened to include in-progress and in-review specs to prevent false warnings
• From-plan step references: Corrected step numbering (6.5 to 7, 7 to 8) in core/from-plan.md and documentation
• Antigravity platform docs: Reconciled canAskInteractive documentation with platform.json (true)