Teach agents to blueprint
survey forms in
: what to measure, how to scope and sequence questions, and aesthetic intent (neutral credibility, scale rhythm, hierarchy)—without inventing study hypotheses, sampling plans, or consent language the user did not supply.
This skill produces survey forms as
packages. Follow the
k2f skill for authoring: unpack or init a workspace, edit the semantic tree, pack/verify, and export. K2F keeps agents on
meaning—sections, prompts, response types—not coordinates.
Before you build, install the core skill:
npx skills add suzheng/k2f --skill k2f
.
Single blueprint for survey content and design intent. Use stable labeled blocks (header, intro, sections, questions, scales, thank-you) the k2f skill can map later.
-
Survey frame (triage) — Freeze what is measured, who responds, why now, and how results will be used. Set mode (customer, product/UX, employee pulse, event eval, academic). Note anonymity, channel, time budget, and tone (neutral research, warm brand, formal).
-
Measurement contract — List
must-learn constructs; mark
out of scope topics and deferred waves. Unknown targets or legal copy →
/
; do not fabricate IRB, GDPR, or policy text.
-
Information architecture (include / cut / merge) — Default stack:
- Header → survey title, sponsor/org, optional version or wave label.
- Introduction (3–5 lines): purpose, estimated time, confidentiality/anonymity as supplied, how to skip optional blocks.
- Screener (only if needed) → eligibility or experience gates; short yes/no rows.
- Core blocks → one theme per section; state the section’s measurement goal in one line before questions (claim-then-items, not mystery headings).
- Open feedback → at most one or two capped prompts; place after scaled blocks unless the user requires early qualitative probes.
- Demographics → usually last (unless screener requires them); collect only fields tied to stated analysis—default cut is “nice to know.”
- Thank-you / close → appreciation and optional contact line if identified follow-up is allowed.
Cut duplicate constructs, double-barreled prompts, leading or loaded wording, essay sprawl, and every question that does not serve the frozen what/why.
-
Question design pass — One construct per item; split double-barreled prompts. Keep response types consistent within a block; define shared scale anchors once and reuse—do not mix agreement and frequency in one matrix. Note branching intent in prose; state limits on open text.
-
Burden & order pass — Easy, non-sensitive items first; sensitive or identifying items late. Group matrix-style items by shared stem where it aids scanning. If the frame risks fatigue, drop optional sections before shortening scales in the blueprint.
-
Design intent — Tone: neutral stems; warmth only in intro/thank-you. Hierarchy: title, sponsor, section goals, then items. Density: section breaks; aligned stems/options; no underline-only grids. Rhythm: compact scale rows; separate open-comment bands. Brand: restrained, one accent, readable contrast.