Match your task to a row below and load the listed files into context. Loading the whole rules/ tree blindly wastes context; loading too little produces drift. This file is the decision tree.
If multiple rows apply, load the union of their files. Skip rows that don't apply.
| You are producing… | Load these first | Then these as needed |
|---|---|---|
| Magazine article or blog post (from scratch) | blog/blog-system-prompt.md, blog/blog-engagement-framework.md, CHECKLIST.md |
PROFILES.md for severities; individual engagement/, voice/, readability/ files when a checklist item flags |
| Magazine article (from a transcript) | transcript/transcript-cleanup-rules.md, blog/transform-from-transcript.md, blog/blog-system-prompt.md, blog/blog-engagement-framework.md |
CHECKLIST.md once a draft exists |
| Tutorial, reference, whitepaper, or book chapter | out-of-scope.md, then the canonical implementations in src/validators/profiles.ts and STYLEGUIDE.md |
PROFILES.md "Other profiles" row for the relaxed thresholds |
| Executive / detailed / bullets / chapters summary | out-of-scope.md (points at summarize-transcript.sh heredocs) |
None — these styles are not extracted |
| Anything else (notes, internal docs, README copy) | None of rules/ is binding. Optionally apply llm-artifacts/ if the prose feels machine-written |
— |
| You are doing… | Load |
|---|---|
| Self-verification of a finished draft | CHECKLIST.md (walks the seven stages in order), then drill into any failing rule's file |
| Stripping LLM artefacts only | All five files in llm-artifacts/: overused-words.md, filler-phrases.md, typographic-artifacts.md, conjunctive-adverb-openers.md, contractions.md |
| Fixing voice (passive, reader focus, topic sentences) | voice/DF-052-active-voice.md, voice/DF-053-reader-focus-you.md, voice/DF-055-frontload-topic-sentence.md, llm-artifacts/contractions.md |
| Fixing readability (sentence length, FK grade, headings) | readability/DF-050-flesch-kincaid.md, readability/DF-051-sentence-length.md, readability/DF-054-descriptive-headings.md |
| Fixing cognitive load (paragraphs, lists, section density) | The four files in cognitive-load/ |
| Fixing engagement (hooks, transitions, arc, examples) | The seven files in engagement/, plus blog/blog-engagement-framework.md for the orchestrating contract |
| Adding or auditing visuals | The three files in visuals/ |
| You are writing… | Load just this |
|---|---|
| The opening 200 words | engagement/DF-040-opening-hook.md |
| An H2 introduction | engagement/DF-041-question-before-answer.md, readability/DF-054-descriptive-headings.md |
| A conceptual section that needs grounding | engagement/DF-042-concrete-examples.md, engagement/DF-043-no-placeholder-names.md |
| Bridges between sections | engagement/DF-046-section-transitions.md |
| The closing | engagement/DF-045-next-steps.md |
| Any code block | engagement/DF-043-no-placeholder-names.md |
| Any image or diagram | visuals/DF-058-image-alt-text.md, visuals/DF-059-image-paths-resolve.md |
research-foundations.md consolidates RF-01..RF-19. Load it when a stakeholder asks "why is this rule like this?" or when you are deciding whether to break a rule deliberately. It is not needed for routine drafting.
blog/blog-system-prompt.md is the operating contract: fidelity to the source wins over stylistic rules. If a rule would force you to invent, soften, or distort a claim, leave the source phrasing and flag it under ## Agent Contributions → ### Unknowns (format in src/templates/agents.md).
See SKILL.md "Directory map" for the full tree. The shortest summary:
- Composites (load one, get many):
blog/,transcript/,CHECKLIST.md - Atomic rules (load one, fix one thing):
engagement/,voice/,readability/,cognitive-load/,visuals/,llm-artifacts/ - Reference (load to interpret):
PROFILES.md,research-foundations.md,out-of-scope.md