trust-and-recovery

📁 bfmcneill/agi-marketplace 📅 13 days ago
1
总安装量
1
周安装量
#43063
全站排名
安装命令
npx skills add https://github.com/bfmcneill/agi-marketplace --skill trust-and-recovery

Agent 安装分布

openclaw 1
opencode 1
claude-code 1

Skill 文档

Trust and Recovery

Trust is built through predictability and tested through failure. Users trust systems that behave consistently and recover gracefully when things go wrong.

Evidence Tiers

[Research]   — Peer-reviewed studies, controlled experiments
[Expert]     — Nielsen Norman Group, recognized UX authorities
[Case Study] — Documented examples from major products
[Convention] — Industry practice, limited formal validation

Multiple tags = stronger evidence: [Research][Expert]
Mixed findings noted as: [Research — Mixed]

Research Foundations

Peak-End Rule

[Research][Expert] Daniel Kahneman’s research (Nobel Prize in Economics, 2002) established that people judge experiences based on:

  1. The peak moment (most intense, positive or negative)
  2. The end (how it concluded)

They do not average the entire experience.

UX implication: A single graceful recovery can redeem an otherwise frustrating experience. Don’t let the last interaction be an error.

Source: Kahneman, D. (1999). Objective happiness. In Well-being: Foundations of hedonic psychology.

Loss Aversion

[Research][Expert] Kahneman & Tversky’s Prospect Theory showed losses feel approximately 2x as painful as equivalent gains feel good.

UX implication: Users are highly motivated to avoid losing their work. Auto-save, undo, and data preservation are disproportionately important.

Source: Kahneman, D. & Tversky, A. (1979). Prospect Theory: An Analysis of Decision under Risk.


Undo vs. Confirmation Dialogs

[Expert] Nielsen Norman Group and multiple UX authorities recommend undo over confirmation dialogs in most cases.

Why Confirmation Often Fails

[Expert] From NNg and practitioner observation:

  • Users habitually click “OK” without reading
  • Frequent confirmations train users to ignore them
  • Confirmations interrupt flow

When to Use Each

Approach Use When Evidence
Undo Action is reversible [Expert] NNg
Confirmation Action is truly irreversible AND destructive [Expert] NNg
Neither Routine, low-risk actions [Convention]

[Case Study] Google Drive: No confirmation for moving files to trash (reversible). Confirmation required for emptying trash (irreversible).

Pattern: Undo Toast

[Convention]

[User clicks delete]
[Item disappears immediately]
[Toast: "Item deleted" [Undo] — auto-dismisses in 10s]

Caution: No controlled studies directly comparing undo vs. confirmation outcomes found. This is strong expert consensus, not validated research.

Source: Nielsen Norman – Confirmation Dialogs


Error Message Design

[Expert] Nielsen’s Heuristic #9: “Help users recognize, diagnose, and recover from errors.”

The Three Questions

Every error message should answer:

  1. What happened? (Clear description)
  2. Why? (Cause, if helpful)
  3. What now? (Recovery path)

Useless:

Error 500: Internal Server Error

Actionable:

Couldn't save your changes — the server is temporarily
unavailable. Your draft has been saved locally.

[Try again] [Continue editing]

[Research] Cognitive Load Theory (Sweller) supports this: vague errors increase extraneous cognitive load.

Source: Nielsen Norman – Error Message Guidelines


Core Patterns

trust-1: Confirm Destructive, Not Routine

[Expert] Only interrupt for truly irreversible actions.

Over-confirming (trains users to ignore):

"Are you sure you want to save?"
"Are you sure you want to go back?"

Appropriate confirmation:

"Delete 47 files permanently? This cannot be undone."
[Cancel] [Delete]

trust-2: Preserve Data Aggressively

[Research] Loss aversion (Kahneman & Tversky) explains why losing work is disproportionately frustrating.

Trust-breaking:

[User writes long comment]
[Accidentally navigates away]
[Returns — comment gone]

Trust-building:

[User writes long comment]
[Accidentally navigates away]
[Returns — draft restored]

Auto-save drafts. Preserve form state. Cache locally.

trust-3: Degrade Gracefully

[Convention] Isolate failures. Don’t let one problem cascade.

Brittle:

[One image fails to load]
[Entire page shows error]

Graceful:

[One image fails to load]
[Placeholder shown with retry option]
[Rest of page works fine]

trust-4: Show System Status

[Expert] Nielsen’s Heuristic #1: “Visibility of system status.”

Opaque:

[User clicks Submit]
[Nothing happens for 3 seconds]
[Suddenly: "Submitted!"]

Transparent:

[User clicks Submit]
[Button shows spinner: "Submitting..."]
[Button changes: "✓ Submitted"]

Recovery Patterns

Pattern: Optimistic UI with Rollback

[Convention]

1. User takes action
2. UI updates immediately (optimistic)
3. Server request in background
4. If success: done
5. If failure: rollback UI + show error + offer retry

Pattern: Forgiving Input

[Expert] Postel’s Law: “Be liberal in what you accept.”

// Rigid
Phone: [Must be exactly ###-###-####]

// Forgiving
Phone: [Accepts any format, normalizes internally]
"5551234567" → displays as "(555) 123-4567"

Pattern: Graceful Timeout

[Convention]

[Operation takes too long]
"This is taking longer than expected.
 You can keep waiting or try again."
[Keep waiting] [Cancel and retry]

Don’t make users guess if something is frozen.


Anti-Patterns

Pattern Why It Breaks Trust Evidence
Silent failures User doesn’t know something went wrong [Expert] NNg
Generic errors No path to recovery [Expert] NNg
Lost form data Punishes user for system failure [Research] Loss aversion
Inconsistent behavior Can’t build mental model [Expert] Jakob’s Law
Hidden data usage Feels deceptive [Convention]

Key Sources