agentsclimarketplace

Speq writing guardrails

Skill marconae/speq-skill/.claude/skills/speq-writing-guardrails

A light-weight and straightforward system for spec-driven development with Claude Code or OpenAI Codex. Written in Rust πŸ¦€

Install
npx -y skills add marconae/speq-skill --skill speq-writing-guardrails

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

What its author says it does

Copied from the file, not written here

Prose guardrails for speq artifacts and GitHub PRs, issues, and comments β€” front-loaded (BLUF), terse, unambiguous writing anchored in established style and requirements standards. Triggered by /speq-mission, /speq-audit, /speq-implement, /speq-plan-pr, /speq-implement-pr, planner-agent, plan-reviewer, and recorder-agent.

SKILL.md

3.7 KB, as published. Nobody here has run it

Writing Guardrails

Prose anchored in BLUF / Inverted Pyramid, Strunk & White, INCOSE GtWR, ISO/IEC/IEEE 29148, and RFC 2119 / 8174.

Scope

Governed (free prose β€” fix these):

  • plan.md β€” Summary, Goals, Non-Goals, Migration, Testing, Dead-Code, Impact prose
  • spec.md β€” the Feature-description line under # Feature:
  • mission.md β€” Problem, Architecture, Constraints prose
  • decision-log β€” Rationale and Finding prose
  • verification-report β€” Notes and Review-Summary bullets, and its condensed PR-comment excerpt
  • every GitHub PR, issue, and comment body and title, including advisory/assumption PR comments

Not governed (leave alone): Gherkin scenarios, Background bullets, tables, ASCII diagrams, delta markers, RFC-2119 keyword casing (validator-owned).

PR-facing content β€” invoke the writing:* skills

For content newly composed for a PR (a verification-report comment, an open-questions/advisory comment, or a plan's ## Impact section), the rules below are necessary but not sufficient β€” additionally invoke, via the Skill tool, before composing:

  • writing:clarity-editing
  • writing:evidence-and-credibility
  • writing:revision-and-qa-checklist

Target expert tier Γ— Evaluator stance per the Audience-Fit Ladder (writing:revision-and-qa-checklist): the reader is an architect deciding whether to approve β€” lead with tradeoffs and boundaries, cite the actual diff/file/test as evidence, never pad or hedge.

Structure first β€” BLUF / Inverted Pyramid

Anchored in NN/g F-pattern and Anthropic context-engineering.

  • Lead with the conclusion. State the decision or outcome in the first sentence.
  • One idea per paragraph. Split a paragraph that carries two.
  • Write statement headings that form a scan path. Prefer ## Rebuild the index over ## Migration.
  • Make each section self-contained. Agents retrieve sections out of order.

Cut verbosity β€” "Omit needless words"

Anchored in Strunk & White and Zinsser.

  • Cap sentences at 25 words; paragraphs at 3–7 lines. The Summary two-sentence cap is hard.
  • Start statements with a verb. Delete you can, there is, it is worth noting.
  • Cut adverbs, qualifiers, and hedges. Delete very, for now, and non-committal time hedges.
  • Ban filler: basically, simply, obviously, just, actually.

Kill ambiguity β€” INCOSE GtWR + ISO/IEC/IEEE 29148

Target: unambiguous, singular, verifiable, complete.

  • Quantify vague terms. Write within 2.0 s, not fast.
  • Delete escape clauses: as appropriate, where possible, etc., and the / in and/or.
  • Use one term per concept. Do not alternate names for one thing.
  • Use active voice and name the actor. Write speq record merges the delta.
  • Repeat the noun over a distant it or this. Pronouns lose their referent across sentences.

Word swaps

WriteNot
useutilize
toin order to
becausedue to the fact that
nowat this point in time
canhas the ability to
somea number of
ifin the event that

Register: descriptive vs normative β€” RFC 2119 / 8174

  • Descriptive prose β€” warm, concise, present tense.
  • Normative statements β€” ALL-CAPS MUST / SHOULD / MAY, used sparingly.
  • Lowercase should is non-binding. Only the ALL-CAPS keyword carries obligation.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.