K2F Policy Document — AI-Native Aesthetic Design

Purpose

Teach agents to blueprint policy documents before layout: scannable rule architecture, plain-language layers for employees, and formal hierarchy—without inventing benefits, legal obligations, jurisdictions, or approval paths the user or HR/legal did not supply.

Built on K2F

This skill produces policy documents as .K2F packages. Follow the k2f skill for authoring: unpack or init a workspace, edit the semantic tree, run pack/verify, then export. K2F keeps agents on meaning (sections, emphasis, tables)—not pixel coordinates—while the engine compiles layout, typography, and theme into consistent export-ready output.

Before you build, install the core skill: npx skills add suzheng/k2f --skill k2f.

When to Use

• New or refreshed single-policy pages (PTO, remote work, expenses, conduct, travel, security, leave types).
• Handbook sections that must read consistently with house style but stay findable on first scan.
• Reformatting draft or pasted policy language into a publishable structure for HR or legal review.
• Not for chat-only policy Q&A, AgentRC scoring policies, or counsel-grade contracts (use k2f-contract).

Workflow

Step 1. Generate the content_and_design.md

One blueprint pairing information architecture with design intent. Label blocks the k2f skill can map to sections, callouts, and emphasis later.

Sub-steps

1. Document frame — Policy title, document ID if user provided, effective date, version, policy owner (team/role), and applicability (entities, locations, employment types). State whether this is new, amendment, or supersedes prior text; link prior version only from user inputs.

2. Purpose & scope — Two short blocks: why the policy exists (one paragraph max) and who/when it applies. Cut narrative that repeats the title; flag HR VERIFY when scope crosses regions the user did not specify.

3. At-a-glance layer — Quick summary (3–5 bullets or one short paragraph): what employees must know before reading detail. This is the scan anchor—not a second copy of every rule.

4. Definitions — Only terms that change compliance meaning in this policy; omit dictionary glossaries.

5. Policy statements — Numbered rules: clear subject, modal verb (must / may not / requires approval), and outcome. One obligation per bullet; no mixed “policy + how-to” in the same bullet.

6. Procedures — Separate numbered flows: how to request, report, escalate, or document compliance (forms, systems, timelines from user only). Procedures reference rules; they do not restate them in prose.

7. Roles & responsibilities — Compact matrix or bullets: employee, manager, HR, IT, finance, legal—who does what at each step.

8. Exceptions & approvals — Who may grant exceptions, required documentation, and where records live. If unknown, APPROVAL PATH — USER TO CONFIRM.

9. Related policies & references — Cross-links and external regs only when user named them; no fabricated citations.

10. Enforcement & questions — Consequences or progressive discipline only if user supplied; otherwise LEGAL/HR TO CONFIRM. Who to contact for edge cases and a single questions channel.

11. Revision history — Table or list: version, date, summary of change, approver—placeholders marked TBD until user fills.

12. Design intent — Clarity over decoration:

    • Tone: neutral, authoritative, plain language in summary; formal precision in numbered rules; no marketing or apology voice.
    • Hierarchy: title and applicability first; at-a-glance and policy statements strongest; procedures and roles secondary; revision history tertiary.
    • Density: moderate—short paragraphs, bullets for rules and procedures; tables for roles or thresholds when user provided numbers; avoid wall-of-text essays.
    • Palette: restrained neutrals, high body contrast, minimal accent on must/may not, deadlines, and approval paths; WCAG AA; no stock imagery that competes with rules.

Rules & constraints

• Source gate: mandatory or jurisdiction-specific language stays verbatim from user; mark all other drafts FOR HR/LEGAL REVIEW.
• Honesty gate: no invented PTO days, dollar limits, legal citations, systems, or approver names.
• Separation gate: rules in policy statements; steps in procedures; do not hide the rule inside a procedure paragraph.
• Applicability gate: every obligation names who it binds; ambiguous geography → flag, do not assume US-default silently.
• Single-topic gate: one policy subject per blueprint; split unrelated topics (e.g., travel vs. expenses) unless user explicitly wants a combined handbook chapter with named parts.

Common failure modes

• Essay introduction with no at-a-glance → employees never find the rule.
• Rules buried in FAQ narrative → extract imperatives into numbered policy statements.
• Procedures before rules → reorder so “what” precedes “how.”
• Duplicate quick summary and full section → summary stays short; detail lives below.
• Generic AI handbook voice → replace with user terms, systems, and approvers or mark TBD.
• Missing who to contact → edge cases bounce to informal channels.
• Amendment without what changed in revision history → compliance tracking breaks.
• Mixing employee-facing plain text with privileged legal analysis → keep external-facing tone only.

Quality checklist

• Frame includes applicability, effective date, and new/amend/supersedes status.
• At-a-glance layer stands alone; policy statements are numbered and imperative.
• Procedures separated; roles and exception path present or marked TBD.
• User-mandated text verbatim; no invented limits, citations, or contacts.
• Related policies listed only from user inputs.
• Design intent specifies tone, hierarchy, density, and emphasis for rules vs. procedures.

Step 2. Use the content_and_design.md and k2f skill to generate a K2F file and export to an appropriate file format.

Follow k2f to build the policy from the blueprint, apply design-intent roles, verify, and export for HR/legal review (typically PDF or DOCX).