Composio powers 1000+ toolkits, tool search, context management, authentication, and a sandboxed workbench to help you build AI agents that turn intent into action.

next
md 3.69 KB

Audit Process

Procedure

  1. Read the good-docs-writing skill first so you audit against its current rules (Voice, Structure, Terminology, Punctuation, Code examples, Formatting). Those rules are the rubric; this skill is the process.
  2. Read the target file(s) in full with line numbers. If the user named a directory or glob, enumerate the docs and audit each. If no target is given, ask which file(s) to audit.
  3. Walk each rule category against the prose. Note every concrete violation with its line number.
  4. Write the report in the format below. Keep suggestions concrete: give the actual rewrite, not "consider revising."
  5. Stop and present the report. Offer to apply fixes; apply only if asked.

Prioritized checklist

Scan for these in roughly this order. The top items most damage the voice:

  1. Hedging & filler: "it might be the case that," "in order to," "basically," "simply," "just." Cut or commit.
  2. Passive voice where active is clearer: "the function is invoked by the client" → "the client invokes the function."
  3. Marketing fluff / vague intensifiers: "seamlessly," "powerful," "robust," "blazing-fast," "cutting-edge," "world-class." Replace with a concrete claim or delete.
  4. Em-dashes: flag every em-dash; the house style bans them. Suggest a period, comma, colon, or parentheses in its place.
  5. Third-person distance: "the user," "one," "developers can" where "you" is meant.
  6. Jargon without context: an undefined acronym or term-of-art on first use with no grounding.
  7. Inconsistent terminology: the same concept named two ways, or core nouns (App, Function, Image, Volume, Secret) miscapitalized or capitalized inconsistently.
  8. Title-case drift: heading case that switches between sentence case and title case within one doc.
  9. Missing backticks: code identifiers, commands, params, paths, or filenames in plain text.
  10. Buried warnings: gotchas, version notes, or gated-feature notes in prose that should be callouts.
  11. Weak openings: page or section that doesn't lead with the concept or benefit; throat-clearing before the point.
  12. Fragment / unrunnable code: snippets that can't be pasted and run, or missing the expected-output / footgun note.

Report format

Output a summary line, then findings grouped by severity. Each finding uses this shape:

- file.md:42 · [Punctuation] Em-dash banned
  Offending: "Modal is fast — really fast — and it scales automatically."
  Rewrite:   "Modal is fast, really fast, and it scales automatically."

Structure the full report as:

Audit:

Summary:

High (breaks the voice / reader-facing clarity)

  • path:line · [Category] · Offending: "…" · Rewrite: "…"

Medium (consistency and polish)

  • path:line · [Category] · Offending: "…" · Rewrite: "…"

Low (nits)

  • path:line · [Category] · Offending: "…" · Rewrite: "…"

Patterns:

Rules

  • Always cite file:line so findings are actionable. Use the line numbers from your Read.
  • Quote the exact offending text; don't paraphrase the problem away.
  • Give a real rewrite, in the target voice, for every finding, not generic advice.
  • Map each finding to a style-guide category (Voice, Structure, Terminology, Punctuation, Code, Formatting) in brackets.
  • Don't invent violations. If the prose already matches the voice, say so. A short report is a good outcome.
  • Don't edit by default. Report, then offer to apply. Only write to files when explicitly asked.