agentskills.codes
CR

critic-comms-completeness

by iampawanSource

Critic that ensures every behavior specifies empty, loading, success, and error states (UI copy + notifications). Invoked by contract-verify. Catches the "we forgot the error message" bug class.

Install

mkdir -p .claude/skills/critic-comms-completeness && curl -L -o skill.zip "https://agentskills.codes/api/skills/download/14424" && unzip -o skill.zip -d .claude/skills/critic-comms-completeness && rm skill.zip

Installs to .claude/skills/critic-comms-completeness

Activation

This is the description your AI agent reads to decide when to run this skill — the better it matches your request, the more reliably it fires.

Critic that ensures every behavior specifies empty, loading, success, and error states (UI copy + notifications). Invoked by contract-verify. Catches the "we forgot the error message" bug class.
194 charsno explicit “when” trigger

About this skill

Comms-completeness critic

Protocol: Follow reference/CRITIC-RULES.md (the compact runtime rules) at the plugin root for severity definitions (the closed enum blocker/warning/info), the output JSON schema, ID prefixes, and shared anti-patterns. This skill defines only WHAT to look for. Do not invent severities like "high"/"medium" — the validator rejects them. Do not re-check anything the deterministic validator (scripts/validate.mjs) already covers; add the judgment layer on top.

Most production "the feature doesn't work" bugs are actually "the feature works but the user can't tell" bugs. This critic catches them before the spec ships.

Contextual (handled by the validator). commsStates is required only for user-facing behaviors; server-only behaviors (platforms all server/backend) are exempt — don't flag them for missing states. Field presence for user-facing behaviors is the validator's job; yours is the judgment layer: is the error copy actionable, on-brand, and specific (not "Something went wrong")?

What to check

1. Every behavior has commsStates

Each ### B<N> should declare:

- commsStates:
    empty:    "<what user sees when state is empty>"
    loading:  "<copy or shimmer description for in-flight>"
    success:  "<confirmation copy or toast text>"
    error:    "<error message — must be actionable>"

Missing any field → blocker with a draft string the author can edit.

2. Error messages are actionable, not blame-y

Error copy should say what the user can DO, not what went wrong in abstract terms. Flag if the error text:

  • says "Something went wrong" with no recourse
  • blames the user without giving an out
  • exposes implementation details ("HTTP 500", "null pointer")

3. Notifications follow the behavior

If a behavior triggers a notification (push, email, in-app), there must be a notification fragment or sub-field describing:

  • Channel (push / email / in-app / SMS)
  • Trigger condition
  • Body copy
  • Deep link target

If the behavior obviously needs a notification but none is specified (e.g., "user receives invitation" with no email copy), emit a blocker.

4. Tone consistency

If the contract or referenced style guide specifies a brand voice (e.g., tone: friendly, or a CLAUDE.md in the repo declaring style), scan the copy fields against it. Flag tonal mismatches as info with suggested rewrites.

5. Internationalization readiness

If the contract's platforms includes platforms that ship in multiple locales, every copy string should be intended for translation (no concatenation, no embedded variables without ICU/i18n markers). Flag as warning.

Severity

  • blocker — missing empty/loading/success/error state, missing notification when behavior implies one.
  • warning — vague error copy, blame-y tone, i18n readiness.
  • info — tone nits, suggested rephrasings.

Output

Same JSON array shape as other critics.

Anti-patterns

  • Don't write the final copy yourself; propose drafts the author edits.
  • Don't enforce a tone that isn't documented somewhere — speculation is worse than no review.

Search skills

Search the agent skills registry