Operator-directed PRIO-0. Graduates Prong 1 of Discussion #14347 (the guide-quality immune system — Euclid's divergence pass + author convergence are done). Refs #14310, #14347.
Why this is PRIO-0. The guide bar currently lives in #14310's epic body + the #14347 ideation thread — but a maintainer heads-down on a sub loads neither; they load the skill. With no guide-authoring skill, every guide is a review lottery: the operator is rejecting 75%+, and both maintainers (Grace + Euclid) have rubber-stamped guides that don't clear the bar. The skill is the only per-sub-loaded enforcement vehicle. Until it exists we keep redoing the same work 4–5× — negative ROI.
Build — elegant, Progressive Disclosure (per .agents/skills/create-skill):
.agents/skills/guide-authoring/SKILL.md — lightweight router (triggers: authoring OR reviewing a learn/ guide).references/guide-authoring-bar.md — the payload:
- Grounding discipline (before the first sentence): memory-mine the topic (
/memory-mining) → USE the subsystem's own tools + read real artifacts → V-B-A vs current code. (The root-cause fix for inference-writing — the MemoryCore 6/10.)
- Content bar: narrative arc (problem → why → how → what's-in-it;
v13.0.0.md depth; clean+accurate ≈ 5/10, narrative = 9/10); industry-friction + benefits-driven; personal / lived maintainer voice + real testimony; ≥1 render-verified TD Mermaid (no-mermaid = reject; no LR-squish / self-loops); conceptual ≠ reference (Diátaxis; extract/generate to tooling/); current-paradigm; dual-audience; never "framework"; never hand-edit generated files (sitemap.xml / llms.txt / SEO — regenerate via buildScripts/docs/seo).
- Reviewer no-rubber-stamp gate: re-grade the WHOLE guide against the bar (lived-voice + industry-friction + significant-improvement), NOT the narrow required-action delta. Approving on the delta is the #1 review failure (this happened on #14346 + #14334).
Acceptance Criteria
Files: .agents/skills/guide-authoring/** (+ manifest registration).
Operator-directed PRIO-0. Graduates Prong 1 of Discussion #14347 (the guide-quality immune system — Euclid's divergence pass + author convergence are done). Refs #14310, #14347.
Why this is PRIO-0. The guide bar currently lives in #14310's epic body + the #14347 ideation thread — but a maintainer heads-down on a sub loads neither; they load the skill. With no guide-authoring skill, every guide is a review lottery: the operator is rejecting 75%+, and both maintainers (Grace + Euclid) have rubber-stamped guides that don't clear the bar. The skill is the only per-sub-loaded enforcement vehicle. Until it exists we keep redoing the same work 4–5× — negative ROI.
Build — elegant, Progressive Disclosure (per
.agents/skills/create-skill):.agents/skills/guide-authoring/SKILL.md— lightweight router (triggers: authoring OR reviewing alearn/guide).references/guide-authoring-bar.md— the payload:/memory-mining) → USE the subsystem's own tools + read real artifacts → V-B-A vs current code. (The root-cause fix for inference-writing — the MemoryCore 6/10.)v13.0.0.mddepth; clean+accurate ≈ 5/10, narrative = 9/10); industry-friction + benefits-driven; personal / lived maintainer voice + real testimony; ≥1 render-verified TD Mermaid (no-mermaid = reject; no LR-squish / self-loops); conceptual ≠ reference (Diátaxis; extract/generate totooling/); current-paradigm; dual-audience; never "framework"; never hand-edit generated files (sitemap.xml/llms.txt/ SEO — regenerate viabuildScripts/docs/seo).Acceptance Criteria
SKILL.md+ the bar payload authored (Progressive Disclosure;turn-memory-pre-flightperformed for the loaded substrate).npm run ai:lint-skill-manifestgreen.Files:
.agents/skills/guide-authoring/**(+ manifest registration).