version: "1.0.1" name: dx-audit description: Audits the developer-facing surface of a library, CLI, SDK, or npm package across API and SDK ergonomics, the error messages developers see, CLI UX, type ergonomics, install and first-run onboarding, and config ergonomics. 35 prefix-dispatched rules; reports findings by file and location with impact tiers and concrete fixes. Use when asked to "audit my CLI", "make a CLI agent-friendly", "is this API ergonomic", "review the developer experience", "improve these error messages", "is this SDK easy to adopt", "is this package easy to install", or "review my library's DX". For end-user UI quality and React/Next UX use ui-audit, for agentic-app UX use ax-audit, for documentation prose use docs-writing, for a README use readme-creator, for AGENTS.md use agents-md, and for repo structure use define-architecture.

DX Audit

Audit the developer-facing surface of a library, CLI, SDK, or package. Load only the rule categories the scoped surfaces need; report findings with file:line and a concrete fix each.

• IS: a quality audit of what other developers touch when they adopt your code: the public API and SDK shape, the errors they read, the CLI commands they run, the types their editor surfaces, the install and first-run path, and the config they write.
• IS NOT: an end-user UI or React/Next UX audit (ui-audit), an agentic-app pattern review (ax-audit), a docs prose review (docs-writing; readme-creator for a README, agents-md for AGENTS.md), or repo structure design (define-architecture).

Audit Workflow

Track this checklist during the audit:

Audit progress:
- [ ] Step 1: Scope. List developer surfaces and the prefixes they need
- [ ] Step 2: Load rules. Read rules/<prefix>-*.md for selected prefixes only
- [ ] Step 3: CRITICAL pass. api-, err- against every scoped file
- [ ] Step 4: HIGH/MEDIUM pass. cli-, types-, onboard-, config- as scoped
- [ ] Step 5: Report. Findings per file with rule id, impact, fix; clean files pass

1. Scope. Default to the changed public surface only; a full-package sweep needs an explicit request. Map each surface to the prefixes it can violate: a public entry point or index.ts needs api-, types-, err-; a bin/ entry or command file needs cli-, err-; a package.json, README quickstart, or postinstall needs onboard-; a config loader or schema needs config-.
2. Load rules. Read rules/_sections.md for the category map, then only the rules/<prefix>-*.md files for the selected prefixes.
3. CRITICAL first. Run api- and err- before anything else; a polished CLI flag is moot while the SDK returns T | T[] | null or an error reads undefined is not a function. For api-stable-contract, first establish the prior public contract to diff against (last release tag, published .d.ts, or git diff main); with no prior version, fall back to the weaker signal that rule names.
4. HIGH/MEDIUM next. cli-, types-, onboard-, config- as scoped.
5. Report and verify. Emit the output contract below. After fixes, rerun the same rule subset on touched files before marking them pass; the rerun output is the evidence the audit is done.

Rule Categories by Priority

35 rules. Per-rule frontmatter may override the category impact; report the rule's own impact, not the category's.

Reference Files

Load on condition:

• rules/_sections.md: category map with impact rationale. Read at Step 2.
• rules/_template.md: rule-file shape. Read only when adding or editing a rule.
• rules/<prefix>-*.md: rule guidance and examples. Read only the Step 1 prefixes.
• references/dx-principles.md: the ten DX principles each rule serves. Read to ground a finding's why or judge a borderline call.

Review Output Contract

Report findings in this format:

## DX Audit Findings
 
### src/client.ts
-[CRITICAL] `api-predictable-return-shape` (line 42): `find()` returns `User | User[] | null` depending on input.
  -Principle: Predictable by design.
  -Fix: Split into `findOne(): User | null` and `findMany(): User[]`; never overload the return shape on the argument.
-[HIGH] `types-no-leaked-any` (line 58): the public `parse()` returns `any`, erasing autocomplete downstream.
  -Principle: No detail is too small.
  -Fix: Return `ParseResult` and constrain the generic to the caller's input type.
 
### bin/cli.ts
-[CRITICAL] `err-suggest-the-fix` (line 31): config-not-found error states the failure but no remedy.
  -Principle: Fight uncertainty, never leave the developer hanging.
  -Fix: Append `Run \`mytool init\` to create one, or pass --config <path>.`
 
### src/clean-file.ts
-✓ pass

• Group findings by file; include file:line when available.
• Every finding states the issue, the DX principle (from references/dx-principles.md), and a concrete fix, not just "improve the API".
• Use the rule's own impact from its frontmatter.
• Include every scoped file; clean ones as ✓ pass.

Gotchas

• Do not load all 35 rule files for a scoped audit; the context cost flattens finding quality. Load only the Step 1 prefixes: a CLI-only audit needs cli- and err-, not types- or config-.
• Do not invent rule ids. Cite only filenames under rules/; describe id-less issues in prose.
• Do not audit internal or private code as public surface. A T | T[] return on an unexported helper is not a DX finding; scope is what consumers import, run, or configure.
• Do not widen scope unprompted. A full sweep when one entry point changed buries the real findings; it requires an explicit request.
• Do not reorder priorities. Reporting a config-default nit while the SDK throws raw stacks (err-no-raw-stack-as-message) inverts the table's load-bearing order; CRITICAL categories always run first.
• Do not mark ✓ pass on a file not read against the loaded rules.
• Not an end-user product audit. When the surface is buttons, pages, and flows rather than functions, flags, and types, route to ui-audit or ax-audit.

Related Skills

• ui-audit: end-user frontend audit, rendered quality plus accessibility and diff-aware React/Next.js UX (state coverage, focus, optimistic UI).
• ax-audit: agentic application patterns and trust design.
• docs-writing: documentation prose, API references, and tutorials.
• readme-creator: a project's README from scratch.
• agents-md: AGENTS.md and CLAUDE.md agent-instruction files.
• define-architecture: folder structure, module contracts, and repository layout.