Changelog

Notable changes to the playbook. Regulatory content is re-verified weekly against primary sources; only material changes are logged here.

2.0.0 (2026-09-04)

Added

11-skills: six installable agent skills for HR teams, ranked in the order to adopt them, plus a human capability ladder that maps each skill to the practitioner skill it depends on
Root README rewrite with role-based entry points and a “what makes this different” section backed by repo facts
SECURITY.md, CODE_OF_CONDUCT.md, CITATION.cff, issue templates, PR template
CI now runs the 10-mcp-agents pytest suite and ruff, and validates skill package frontmatter
Eval runner is now a real launch gate: exits 1 when a refusal, escalation, or reachability gate fails; accepts OpenAI, Anthropic, and plain JSON response shapes, not only SSE streams; --responses-file scores canned transcripts with no endpoint; 11 scorer tests run in CI
Attrition notebook fairness audit extended from selection-rate parity to calibration by group and error-rate parity, with a small-group caution (the synthetic data shows exactly why: a 30-person group produces an unstable false negative rate)
ROI calculator adds a realization rate and a one-time investment; payback formula now consistent between the dashboard, the business case template, and the framework (the template previously labeled a years figure as months)

Changed

EDPB 2026 Coordinated Enforcement Framework described accurately as a general GDPR transparency action, not an AI-hiring audit (intake template, deployer checklist)
Colorado: enforcement of SB 24-205 and SB 26-189 is stayed by a federal court pending xAI’s challenge and AG rulemaking (risk assessment template, prioritization matrix)
Texas: TRAIGA complaint portal is live; noted the statute’s “consumer” definition excludes the employment context (risk assessment template)
Illinois: added Public Act 104-0425 civil penalty tiers, verified against statutory text (literacy curriculum)
Added Mobley v. Workday as the reference case for vendor liability, and the ICO’s post-report consultation on ADM guidance
Removed every em dash from the repo; fixed seven ruff findings in the MCP tools
Eval scorer signals tightened: bare “human” no longer counts as an escalation (it matched “human resources”), and a refusal that names “system prompt” is no longer scored as compliance

Working MCP server with four HR tools and a 32-test suite
Executed notebooks: attrition risk with fairness audit, skills gap analysis, HR Q&A agent demo
29-case eval framework with launch-blocking gates
Governance suite: AI use policy, risk assessment, EU AI Act intake, vendor selection and intake, deployer checklist, incident report, pay equity governance
Prompt library across eight HR functions
Literacy curriculum, facilitator guide, adoption playbook, 18-month roadmap, ROI dashboard