Facet Skills

SDD-oriented agent skills for design stress-testing and staged learning. Each skill is intentionally scoped: it either pressure-tests a design before implementation, or helps turn knowledge into something usable without falling into perfectionist deep-dives.

Skills

Skill	Description
facet	Design stress-test engine. Grills your draft against edge cases, code constraints, and failure paths. Outputs a Facet Brief.
fable-fuse	Meta-learning protocol. Uses fable-based simplification, bounded regress, Feynman checks, and cognitive debt to help users learn in stages and turn knowledge into something usable.

---

Facet

The last gate before the spec gets written. Facet stress-tests your draft design — one sharp question at a time — then hands you a Facet Brief (or Quick Note) that feeds directly into openspec or speckit. No chat-scrolling. No "let me summarize."

Install

# All skills from this repo
npx skills@latest add hondrytravis/skills

# Just Facet
npx skills add https://github.com/hondrytravis/skills --skill facet

# Just Fable-Fuse
npx skills add https://github.com/hondrytravis/skills --skill fable-fuse

Fable-Fuse

Fable-Fuse is a meta-learning protocol for staged understanding. It does not try to explain everything at once; it helps the user get one usable slice of knowledge, validate it with a Feynman-style check, record the skipped boundaries as cognitive debt, and come back to repay that debt when practice actually touches the edge.

What makes it different

Most design-review skills stop at "shared understanding." Facet goes further with structured artifacts — the Facet Brief and Facet Quick Note — that map onto spec tooling without a translation step.

Capability	grill-me	grill-with-docs	brainstorming	Facet
Grill an existing draft	✅	✅	❌	✅
Structured SDD output	❌	❌	❌	✅ Brief + Quick Note
Silent-failure probing	❌	❌	❌	✅
Verify claims against code	partial	partial	❌	✅
Early exit, resume later	❌	❌	❌	✅ Quick Note
Update project docs	❌	✅	❌	❌
Generate design from nothing	❌	❌	✅	❌

Facet is not a design generator. You bring the draft — even three sentences. Facet takes it from 0.7 to 0.95 before the spec crystallizes.

How it works with openspec

rough draft  →  facet  →  Facet Brief / Quick Note  →  openspec  →  code

Walkthrough

You walk in with a draft:

"Notification system. Users get notified on @mentions. Email + in-app. Done."

Facet grills it, one question at a time:

"If the email provider is down for 30 seconds, does the in-app notification still fire — or does one failure block the other?"

"Three comment types exist in the codebase: inline reviews, issue comments, PR threads. Which ones trigger a notification?"

"Write succeeds but the notification queue write fails silently — does the comment still appear? Who discovers the gap? How soon?"

After 8–12 rounds, Facet hands you a Brief. Drop it into openspec:

openspec/
├── changes/
│   └── notification-system/
│       ├── proposal.md    ← Facet "Conclusion" + "Evidence"
│       ├── design.md      ← Facet "Weak points" + "Rejected approaches" + "Risks"
│       └── tasks.md       ← Facet "Next steps"

No synthesis. No re-typing. Brief → spec artifacts in one pass.

Structure

skills/<name>/SKILL.md          ← English skill (shipped)
reference/<name>/SKILL.zh.md    ← Chinese original (reference only)
reference/<name>/RFC_zh.md      ← Chinese RFC (design doc, where applicable)

Every skill needs a reference in this README and an entry in .claude-plugin/plugin.json.