Critical Compass v0.1.0-beta.11

Critical Compass v0.1.0-beta.11

Beta.11 is a course-reliability release on top of beta.10. It adds a standardized classroom audit prompt/resource, routes course users toward standard audit_text, keeps advanced panel review optional, and strengthens score-delta interpretation so class discussions focus on findings, severity, and revision decisions rather than small point changes.

Download Guide

• critical-compass-plugin.zip: Claude Cowork and Claude Code plugin upload.
• critical-compass-0.1.0.mcpb: Claude Desktop extension install through Settings > Extensions.
• critical-compass.skill and skill.zip: skill / prompt-only alternative.
• critical-compass-source.zip: source files for inspection, modification, or rebuilds.
• critical-compass-example-report.html and critical-compass-example-report.pdf: redacted public example audit.
• checksums.txt: SHA-256 checksums.

What's New

• classroom_standard_audit_prompt scaffold for Week 2 / classroom bias-audit assignments.
• get_started and critical_compass_overview now surface the classroom standard-audit path.
• Public landing page includes a Classroom Bias Audit prompt card.
• Standalone skill guidance includes the classroom prompt for prompt-only use.
• compare_audit_results warns that score movement is diagnostic, not improvement proof by itself.
• validate_report_payload dry-runs report payloads before rendering and can return schema docs with schema_only=true.
• generate_audit_report(preflight_check=true) runs the same dry-run checks without writing files.
• Advanced synthesis returns a report_ready_payload so host AIs have less brittle payload assembly work.
• Advanced synthesis confirms accepted silent mode with human_loop_confirmation.
• Markdown is written before PDF rendering; renderer or QA timeout failures preserve markdown_path, markdown_hash, and render_error_log_path.
• PDF/HTML reports suppress placeholder/debug strings, raw enum values, and reader-facing internal notes.
• CIS bands and action routing are explicit; low-score reports cannot render Trust as the primary action.
• Human-loop guidance now foregrounds PAUSE_REQUIRED and Claude's native Ask_User_Input_V0 style question UI.
• Runtime bundles exclude planning-registry material and do not rely on external MCP orchestration.

Claude's unverified developer/access warning is expected for locally shared plugins and extensions. A Claude Artifact mirror is available at https://claude.ai/public/artifacts/9e50e78b-3877-40c0-855b-07031eb15c95.