agentleFS
danicatdanicat/speedgrapher/skills/draft/SKILL.md

draft

Skill51 starsChanged 17 days agoApache-2.0View on GitHub

Adaptive-voice technical writing workflow that drafts and expands engineering blog posts, tutorials, READMEs, and documentation matching the author's natural style.

  • Commits and pushes

We split this skill into the rules it gives an agent

13 rules. 0 are common, 9 are distinctive, and 4 are very specific to this project.

  • 0

    Common

    Most of the 20 edited versions we read keep these.

  • 9

    Distinctive

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Follow this five-stage workflow whenever authoring or expanding a technical document:

    line 23
  2. Distinctive:

    Before writing the first paragraph, inspect the workspace for existing posts, README.md files, or style guidelines to capture the author's …

    line 40
  3. Distinctive:

    Peer-to-Peer Tone: Write as an experienced engineer sharing hard-won context with a colleague.

    line 43
  4. Distinctive:

    Every draft must serve both time-constrained human readers (*The Scanner*) and AI answer engines (Generative Engine Optimization / GEO):

    line 46
  5. Distinctive:

    Self-Contained & Idiomatic Snippets: Keep examples focused, idiomatic for the target language, and runnable in principle.

    line 54
  6. Distinctive:

    Highlight the Critical Mechanism: After the code block, direct the reader's eye to the specific struct field, concurrency primitive, or …

    line 55
  7. Very specific:

    Never Print Bare URLs in Prose: Do not expose raw URLs (https://...) or unformatted domain names inside narrative sentences.

    line 62
  8. Distinctive:

    All technical drafts must comply with strict brand, trademark, and legal safety rules:

    line 88
  9. Very specific:

    Before presenting a completed draft or expanded section to the user, run the speedgrapher MCP tools on the target file or text:

    line 103
  10. Very specific:

    If any paragraph exceeds 16.0, split compound clauses over 30 words and replace multi-syllable nominalizations with active verbs.

    line 107
  11. Very specific:

    Always include file.md in your response so the author can read, comment, or edit the draft in the side panel.

    line 115
  12. Distinctive:

    Never Fabricate Technical Facts: Do not invent CLI flags, package names, API signatures, or benchmark numbers.

    line 121
  13. Distinctive:

    No Git Operations: Never execute git commit, git push, or other git state changes.

    line 122

“Common” is measured: kept by at least 6 in 10 of the 20 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Draft: Adaptive-Voice Technical Writing

draft transforms structured outlines, interview briefs, and raw engineering notes into clear, authoritative, human-grade technical prose. You never impose a generic machine persona; instead, you fingerprint and replicate the author's natural voice while enforcing strict structural, pedagogical, citation, and brand-safety standards from the very first paragraph.


1. End-to-End Drafting Workflow

Follow this five-stage workflow whenever authoring or expanding a technical document:

flowchart LR
    Voice["1. Voice Fingerprinting\nInspect workspace style & posts"] --> Framing["2. Outcome-First Framing\n150-word lead & GEO section leads"]
    Framing --> Pedagogy["3. Grounded Drafting\nWhy-before-how code & citations"]
    Pedagogy --> Safety["4. Brand, Slop & Jargon Guardrails\nSentence case & trademark rules"]
    Safety --> Gate{"5. Pre-Delivery Gate\nRun `fog`, `slop` & `vale`"}
    Gate -->|Out of Bounds| Pedagogy
    Gate -->|Clean| Deliver["Deliver Draft + Speedgrapher Pill"]

2. Core Editorial Pillars

A. Voice Fingerprinting

Before writing the first paragraph, inspect the workspace for existing posts, README.md files, or style guidelines to capture the author's voice fingerprint:

  • Pronoun & Perspective Stance: Match the author's established stance consistently—whether conversational first-person singular ("I tracked the memory leak to..."), collaborative engineering plural ("We benchmarked three column readers..."), or direct second-person instructional ("Configure the listener...").
  • Sentence Cadence: Mirror the author's natural mix of crisp declarative statements (5–10 words) and analytical clauses (20–28 words).
  • Peer-to-Peer Tone: Write as an experienced engineer sharing hard-won context with a colleague. Never use patronizing minimizers (simply, merely, just, obviously, trivial) or synthetic teacher-mode fillers ("Let's break this down", "Here's the kicker").

B. Built-In Inverted Pyramid & GEO-Ready Section Leads

Every draft must serve both time-constrained human readers (The Scanner) and AI answer engines (Generative Engine Optimization / GEO):

  1. 150-Word Outcome Lead: Open the document by stating the exact technical problem, the architectural fix, and the measurable outcome (or primary copy-paste command) within the first 150 words. Never open with throat-clearing truisms ("In today's fast-paced cloud-native world...").
  2. Self-Contained Section Leads: Open every H2 and H3 section with a 1–2 sentence factual statement that directly answers the heading's implied question before diving into code or background narrative.
  3. Explicit Version Pinning: State exact language, runtime, and library versions (for example, Go 1.27, Vale v3.13.1, MCP 2025-03-26) so readers and retrieval engines know the exact contract.

C. Grounded Code Pedagogy (Why Before How)

Code blocks are primary engineering evidence, not decorative padding:

  1. Explain Why Before How: Introduce every code snippet by stating the constraint, failure mode, or design decision it solves before the code block appears. Never drop an unintroduced code block immediately below a heading.
  2. Self-Contained & Idiomatic Snippets: Keep examples focused, idiomatic for the target language, and runnable in principle. Strip repetitive boilerplate, but never omit error handling when error handling is part of the lesson.
  3. Highlight the Critical Mechanism: After the code block, direct the reader's eye to the specific struct field, concurrency primitive, or syscall that makes the pattern work rather than narrating every line mechanically.

D. Canonical Citation Hygiene

Anchor every tool, library, standard, and specification to its authoritative upstream source:

  • Hyperlink Canonical Names on First Mention:
    • ✅ We validated the schema against the [Agent Skills specification](https://agentskills.io) and exposed the tools over the [Model Context Protocol](https://modelcontextprotocol.io).
    • ✅ Run [Vale](https://vale.sh) in your pre-commit hook.
  • Never Print Bare URLs in Prose: Do not expose raw URLs (https://...) or unformatted domain names inside narrative sentences.
  • No Vague Authority Appeals: Replace unnamed attributions ("experts suggest", "industry reports show") with links to specific RFCs, benchmarks, papers, or source files.

E. Sentence-Case Headings & Forward Momentum

  • Strict Sentence Case: Format all headings (H1–H4) in sentence case—capitalize only the first word and proper nouns or acronyms (for example, ## Benchmarking JSON serialization in Go 1.27, not ## Benchmarking JSON Serialization In Go 1.27).
  • Zero Narrative Re-Litigation: Once the opening lead establishes why the problem matters, keep implementation sections moving forward without re-litigating the motivation.
  • Zero Signposted Outros: End on concrete benchmarks, operational trade-offs, or next steps—never on a redundant "In conclusion, we have seen..." paragraph.

3. First-Draft Anti-Slop & Anti-Jargon Guardrails

Prevent synthetic phrasing and status-signaling jargon during initial composition so the draft requires minimal surgical cleanup later:

GuardrailForbidden First-Draft Habit (❌)Clean Engineering Prose (✅)
Zero Inflated Verbs & Nounsdelve, tapestry, landscape, paradigm, synergy, testament, beacon, catalyst, leverage, utilize, harnessUse exact system nouns and direct verbs (use, build, measure, parse, route).
Zero Unverified Adjectivesrobust, seamless, transformative, cutting-edgeState the verifiable mechanism (retries expired tokens automatically, processes 1.4 GB/s).
Zero Negative Parallelisms"It's not X — it's Y", "Not X. Not Y. Just Z.", "The result? Devastating."Write a single affirmative cause-and-effect clause.
Functional Precision vs. JargonUndefined acronyms ("E2E SLOs via eBPF CIL"), noun stacks ("orchestration layer validation methodology")Expand unfamiliar acronyms on first use and convert noun stacks into active verbs ("how the coordinator validates requests").
Controlled Punctuation & ListsMultiple em-dashes (—) per paragraph or mechanical **Bold Keyword**: Description listiclesUse commas, parentheses, or periods; use prose paragraphs or multi-column comparison tables.

4. Google Developer Brand & Legal Safety

All technical drafts must comply with strict brand, trademark, and legal safety rules:

RuleForbidden Pattern (❌)Compliant Pattern (✅)
No Unverifiable Superlatives"Gemini is the fastest and best model for code generation.""Gemini 2.5 Flash reduces time-to-first-token by 35% on our benchmark suite."
No Possessives on Brand/Product Names"Google's Cloud Run's autoscaler" / "Go's compiler's escape analysis""The Cloud Run autoscaler" / "Escape analysis in the Go compiler"
Product Names as Adjectives (or Exact Nouns)"Spin up three Cloud Runs to handle the queue.""Deploy three Cloud Run services to process the queue."
No Unsubstantiated Claims"This architecture eliminates all production outages.""This retry budget prevented cascading timeouts during our regional failover test."
No Future Roadmap Promises"Support for streaming gRPC is coming soon in v3.0."Document only capabilities that exist in the current released version.
Respect Copyright & Fair UseCopying verbatim paragraphs from third-party blogs or proprietary docs.Write original analysis and link directly to primary documentation.

5. Mandatory Pre-Delivery Verification Gate

Before presenting a completed draft or expanded section to the user, run the speedgrapher MCP tools on the target file or text:

  1. fog (Gunning Fog Index):
    • Target range: $11.0\text{–}15.0$ across the document and individual paragraphs.
    • If any paragraph exceeds 16.0, split compound clauses over 30 words and replace multi-syllable nominalizations with active verbs.
  2. slop (AI Cliché & Cadence Detector):
    • Target score: $< 20$ (never deliver a draft with a Slop Score $\ge 25$).
    • Purge any flagged lexical tells, structural clichés, or monotonous sentence-length runs.
  3. vale (Static Style & Grammar Linter):
    • Target: 0 errors and 0 warnings.
    • Fix passive voice clusters, weasel words, title-case headings, and spelling/style violations while preserving intentional technical terminology.
  4. Surface the Speedgrapher Pill:
    • Always include [file.md](sidecar://speedgrapher/panel/?file=/absolute/path/to/file.md) in your response so the author can read, comment, or edit the draft in the side panel.

Operational Constraints

  • Never Fabricate Technical Facts: Do not invent CLI flags, package names, API signatures, or benchmark numbers. Verify every symbol against the workspace or official documentation.
  • No Git Operations: Never execute git commit, git push, or other git state changes.

Download now (ZIP)

  • Other versions23 projects use a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 23

    1. anthropics/claude-for-legal★ 9.5ksame name
    2. jigripokri/POHA★ 121same name
    3. tance-mang/chinese-webnovel-skills★ 65same name
    4. richard-kim-79/archora-skills★ 49same name
    5. drafthq/draft★ 40same name
    6. alirezarezvani/gaios★ 37same name
    7. coolxeo/grok-bot-skills★ 7same name
    8. vercel-labs/workflow-campaign-demos★ 6same name
    9. iwe-org/marketing-workspace★ 5same name
    10. imankulov/skills★ 4same name

    13 more not listed

    How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.

  • Used alongsideProjects with it also use onboard, status and 2 more.

    Other files people use alongside this one.

    Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
    • SentrySkill

      onboard

      Onboard a service to sentry-options. Use when integrating a new service, creating schemas, or setting up the multi-repo sentry-options workflow.

      Also in 3 projects that use this

    • AnthropicSkill

      status

      Case status summary by audience — client-facing (plain language), internal (for the professor), or court-ready (formal caption format per local rules). Same facts, different framing and depth. Use when a student needs to update the client, brief the professor, or prepare a court status report.

      Also in 3 projects that use this

    • VercelSkill

      setup

      Set up Vercel CLI and project configuration. Use when the user says "set up Vercel", "configure Vercel", "link to Vercel", or "vercel init".

      Also in 4 projects that use this

    • SentrySkill

      review

      Fixture review skill.

      Also in 5 projects that use this

  • In this repository6 other files it gives its agents.

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.