Kartik-Doye/Ravengard-AI-Recruiter / .agents/skills/friction
friction
VerifiedDiagnoses and improves usability by walking real user tasks, finding hesitation, backtracking, errors, recovery gaps, accessibility barriers, and inefficient paths. Use for UX audits, onboarding, forms, navigation, error messages, or confusing flows.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 10.0 KB |
| references/research.md | markdown | 2.6 KB |
| references/usability-checklist.md | markdown | 4.1 KB |
references/usability-checklist.md
View on GitHub# Usability checklist: walk with a user's hands, not a linter
Grouped by where friction lives. Each: the question to answer on the real
flow, and the classic tell.
## Information architecture and navigation
- Can a first-time visitor identify the product and next action without
avoidable hesitation?
- Can users find critical actions without avoidable navigation? Tell:
settings split across three menus by internal team structure, not user
mental model.
- Labels name user goals ("Billing") not system concepts ("Subscription
Resource Management").
- Back/breadcrumb exists for anything buried; the user always knows where
they are.
## Forms and validation
- Inputs ask only for what's needed now; every field has a visible label
(placeholders are not labels: they vanish on type).
- Validation timing: inline on blur for format, on submit for
cross-field; never punish-as-you-type.
- Error messages: what happened + how to fix it, attached to the field:
"Email needs an @: like name@company.com", not "Invalid input".
- Constraints stated before the mistake: "Password: 12+ chars" under the
empty field, not after failing.
- Progress preserved on failure; back button never destroys form state.
## Error recovery and destructive safety
- Every error has a next step (retry, contact, alternate path): dead-end
screens are blockers.
- Destructive severity ladder: toggle → instant; delete draft → confirm
inline; delete account/project → typed confirmation + consequence
statement; prefer soft-delete with undo wherever possible.
- Undo beats confirmation beats nothing, in that order of preference.
- Session loss: drafts autosaved; timeout warns before expiry.
## Feedback and latency perception
- Every action acknowledges promptly relative to measured latency and user
expectations.
- Perceptible waits show progress honestly with treatment appropriate to the
operation.
- Success confirmations state what changed ("Moved to Archive", not "OK").
- Optimistic UI only where failure can roll back visibly.
## Keyboard, focus, and input
- Full task completion keyboard-only: logical tab order, visible focus
(WCAG 2.4.7), no focus traps without escape (2.1.2), Escape closes
dialogs.
- Enter submits the form it's in; Enter/Space activates the focused
control.
- Apply WCAG 2.2 target-size requirements with their documented exceptions.
Larger targets may be a product heuristic; no hover-only affordances on
touch devices.
## Accessibility anchors (WCAG 2.2, the recurring ones)
- Text alternatives for meaningful images (1.1.1); contrast 4.5:1 body /
3:1 large text and UI components (1.4.3 / 1.4.11).
- Reflow at 320px width without 2-D scroll (1.4.10); consistent help and
identification across pages (3.2.6).
- Accessible authentication: no cognitive-test-only login (3.3.8).
- Status messages announced without stealing focus (4.1.3).
- Run axe when available for the mechanical floor; judge the rest by hand.
## AI / Agentic UX (2026)
- **Planning Visibility:** Does the AI state its intended plan *before* executing destructive or long-running actions?
- **Audit Trails:** Are agent actions logged in plain language (e.g. "Emailed finance@...") rather than raw JSON/tool calls?
- **Graceful Degradation / Escalation:** When the AI gets stuck, is there a clear, immediate escalation path to a human, or does the user get trapped in a loop?
- **Selective Transparency:** Does the UI expose the 'why' behind an AI decision without overwhelming the user with full reasoning traces?
## Copy
- Sentence-case, plain words, no jargon the user didn't introduce.
- Buttons start with verbs and name the outcome ("Create project").
- Empty/error/success copy answers: what happened, so what, now what.
## Response patterns (novice vs expert)
- Novice: guidance where they'll land first-time (empty states with
actions, examples in placeholders where labels alone are unclear).
- Expert: defaults that skip steps, shortcuts that don't break the
guided path, bulk actions on repeated operations.
- Never make the expert route the only route, or the guided route a
maze for the expert.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| rea-analysis-optimizationmorluto/reaOptimize REA providers and shared analysis workflows for agent context, latency, CPU, and memory while preserving correctness and Evidence. Use for analysis bottlenecks and cache or query design; CI/test-suite optimization belongs to rea-ci-tests-audit. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| rea-ci-tests-auditmorluto/reaAudit REA CI and test suites for correctness, fixture drift, false passes, and measured performance; use for suite or workflow investigations and efficiency changes, rather than ordinary feature testing or a single test failure. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| rea-tool-designmorluto/reaAudit, investigate, design, or change REA investigation tools, CLI/MCP contracts, provider capabilities, and Evidence semantics. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-auditDietrichGebert/ponytailQuality audit of the whole repo: bugs, security, real load, missing tests, speed, and what to delete. Most important first. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |