Mode d’emploi
- Copiez la commande et examinez le paquet avant de l’installer dans votre environnement agent.
- Consultez le document original pour les outils, identifiants et environnements requis.
- Décrivez les entrées et le livrable souhaité ; comparez la sortie aux sources.
npx skills add https://github.com/jeffabailey/skills-writing --skill write-strategyAvant de commencer
- L’installation n’inclut ni abonnement au modèle, ni quota API, ni accès aux fichiers.
- Étoiles et installations reflètent l’activité, pas un test indépendant de qualité ou de sécurité.
Document Skill original
Rendu depuis le dépôt source en conservant titres, exemples, code, tableaux, liens et images.
Writing Strategy
Also invoked as /write:competitor-analysis (competitor mode) and /write:mckinsey-consultant (growth mode).
Pick the mode
| Request sounds like | Mode | Prompt |
|---|---|---|
| "content strategy", "what should I focus on publishing", editorial plan, content calendar for a blog, newsletter, or docs | content | references/content-strategy.md |
"competitor analysis", "competitive landscape", "who else does this", "where are the gaps", /write:competitor-analysis | competitor | references/competitor-analysis.md |
"growth audit", "McKinsey", "grow X to Y", "where am I stuck", scorecard or roadmap toward a number, /write:mckinsey-consultant | growth | references/mckinsey-consultant.md |
State which mode you picked and why in the report's first line under the title. If a request spans two (a content strategy that needs a competitor view), run the primary mode and fold in at most a short competitor table. Nearby skills: topic lists go to write-ideate; keyword research and site architecture go to claude-seo:seo-plan.
Workflow
- Load the prompt for the mode. Each prompt opens with an Inputs table: the
{{var}}names, what they mean, and the default. - Fill inputs without interrogating the user. Pre-fill from the request, then from the repo or site (About page, README,
hugo list all). Use the table's default where one exists. Ask once, in a single message, only for inputs marked required that you cannot infer. "unknown" and "n/a" are valid answers; carry them through as "unknown", never as an estimate. Show the filled inputs as a short table at the top of the report. - Gather evidence within budget. Measure local facts with tools: for a Hugo site,
python3 scripts/content_inventory.py <hugo-dir>(or--csv <saved hugo list all output>; needshugo, and on "tool missing" printbrew install hugoand fall back to the sitemap or RSS). In competitor and growth modes, use it for a short subject baseline when the subject is a Hugo site. Web research is capped by size: solo blog, side project, or OSS: at most 6 web calls (8 in competitor mode, one landing page per competitor plus a gap check); small business: 10; funded or established company: 15. Search summaries rarely say which URL holds which figure: cite a figure to a URL only if you fetched that page or the result showed it, otherwise tag it [Inference]. Prefer fetching the primary page over trusting a search snippet. Each search or fetch is one call; loading tools is not. When the cap is hit, stop and list what you did not check. - Apply the prompt and write the report in its output format, under the evidence contract below. Keep the prompt's
##headings in order; add###subsections or an Inputs table freely. - Self-check:
python3 scripts/check_strategy.py <report.md> --mode content|competitor|growth. Fix every FAIL. Review each WARN line and tag or reword it. - Deliver: save to the path the user named; otherwise
strategy/<YYYY-MM-DD>-<mode>-<subject-slug>.mdunder the current directory, and never under a Hugocontent/folder (Hugo would publish it). Tell the user the path; do not commit it. In chat, give the top three actions and the "Data the user should supply" list. Offer write-article for drafting or write-ideate for topics.
Evidence contract
Strategy built on invented numbers is worse than none. Every report follows these rules:
- Tag claims inline, with these exact openings (a note after a colon is fine, e.g.
[Assumption: 2% churn]).[Evidence: <url, file, command, or "user input">]for sourced or measured facts,[Inference]for reasoning from evidence,[Assumption]for a value you chose so a calculation can proceed. Arithmetic keeps its inputs' tag; if any input is an assumption, the result is [Assumption]. Numbers in a table can share one tag line under the table. - "unknown", not estimates. Missing facts about the present (traffic, revenue, churn, prices, dates) stay "unknown". Projections may still run on a stated [Assumption] range, listed under Assumptions, so the plan is sized. Never invent revenue, users, installs, testimonials, URLs, or dates.
- Conflicting figures become ranges, citing both sources.
- Cite only what you saw. A source is a page you fetched, a search result you read, a file, or a command you ran.
- Close with "Sources and Assumptions": numbered sources, the assumptions list, and "Data the user should supply" (what would most change the conclusions).
Impact and Feasibility rubric (all modes)
| Score | Impact on the goal metric | Feasibility for this team and budget |
|---|---|---|
| 5 | Changes the trajectory of the goal metric | Doable this week with current skills and tools, no spend |
| 4 | Clear, measurable movement on the goal metric | Within a month, small spend or one new tool |
| 3 | Noticeable on a secondary metric or one segment | One to three months, or needs a new skill |
| 2 | Small or local effect | More than three months, or meaningful spend or help |
| 1 | Negligible or unmeasurable | Needs resources the user does not have |
Priority Score = Impact x Feasibility. Sort descending; break ties by Feasibility. A data-gathering action (connect analytics, measure churn) scores Impact by the decisions it unlocks. When the goal metric is unmeasured, scores are judgment: say so once above the table. Growth mode's "ease" is this Feasibility scale; its 0-10 Scorecard is separate.
Output style
Plain Markdown, short direct sentences, no emdashes, no hype. These reports are working documents, not blog posts, so the blog's writing-style guide does not apply.
