Agent Skills: Approachability Audit

Approachability Audit

Read the code like a smart but newly onboarded TypeScript developer.

Related skills: use post-implementation-review for the full second-read ritual after implementation, refactoring for caller counting and inlining mechanics, and cohesive-clean-breaks when a readability issue points to the wrong public boundary.

The goal is not "make it shorter" or "make it more abstract." The goal is:

• make the first read cheaper
• make ownership boundaries obvious
• remove tricks that only exist for the type system
• keep real domain or validation boundaries intact

What to Look For

• Names that imply a more important concept than the code actually represents
• Types that pretend to be runtime truths but are only inference shims
• Client-only hacks living in server packages, or server concerns leaking into browser/shared code
• Single-use helpers that hide simple control flow
• Tiny wrappers that add vocabulary but not clarity
• Multiple files for one concept when readers really need one place to look
• "Smart" generic helpers where direct code would be easier to trust
• Ad-hoc as assertions outside a clear parse or interop boundary
• JSDoc that restates code instead of explaining why the boundary exists

What Not to "Fix"

• Real parse boundaries at JSON, file, or network edges
• Runtime validation that protects unsafe input
• Shared contracts that genuinely belong in one place
• Repetition that is cheaper than an extra abstraction

Review Method

1. Start with the entrypoint a caller uses first.
2. Trace the minimum path needed to understand the behavior.
3. Count the hops:
   • if understanding one field or behavior requires jumping across multiple files, ask whether that indirection earns its keep
4. Mark each abstraction as one of:
   • earns its keep
   • probably inlineable
   • wrong ownership boundary
   • misleading name
   • type-system workaround
5. Prefer fixes that:
   • move code to the right owner
   • rename things honestly
   • collapse fake layers
   • replace cleverness with one explicit boundary

Output Shape

When reporting findings, prioritize:

1. The one or two biggest readability or ownership smells
2. Why a new TypeScript developer would stumble there
3. The greenfield correction that would make the code easiest to trust

When editing code:

• keep commits surgical
• add or improve JSDoc on public boundaries
• do not add a new abstraction unless it reduces total cognitive load
• prefer explicit names like *Bridge, *Contract, *Parser, *Factory, *State

Heuristics

Good smell

export type SessionContract = { ... };

Portable type, honest name, obvious job.

Bad smell

export type AppAuth = {
  options: { plugins: [...] }
};

If this is not the real auth type, do not name it like it is.

Good workaround

export type CustomSessionClientBridge = { ... };

If a library forces a type trick, keep it local and name it as a bridge.

Go-to-Definition Is the First Read

A new developer's first read of a foreign symbol is Go-to-Def. Treat it as the canonical readability tool: pressing it from a call site should land on the actual source of truth in as few hops as possible. Counting hops (review method, step 3) means counting Go-to-Def hops.

The smell catalog that bloats the hop count (re-export chains, destructure-re-exports, no-op adapters, identity-obscuring annotations) lives in typescript "Go-to-Definition Awareness". Each extra hop has to earn its keep: if a layer doesn't own a real invariant, name non-obvious domain behavior, or isolate unsafe input, collapse it.

Success Criteria

The code should leave a new teammate thinking:

• "I know which file owns this concept."
• "I know which type is a real contract and which one is just library glue."
• "I don't have to reverse-engineer the architecture from naming accidents."