Flows category icon
Harness Engineering
Intermediate

Design the Permission Approval UX

Design approval prompts users actually read: diffs, risk framing, scoped grants, and pacing that prevents approval fatigue.

4 steps12 verify checks60-90 minutesWorks with: emergent · chatgpt · claude · cursor
Start Guided Walkthrough

The Route

0/4 verified

Context Pack

Paste this first. It briefs the AI on requirements, constraints, and the Definition of Done before your first build prompt.

Context Pack
PROJECT CONTEXT:
My agent asks permission constantly, users stop reading and click yes to everything - the safety layer trained them out of safety. The leading agent UX pattern is diff-first approvals, risk-proportional friction, and scoped grants.

GOAL:
Rebuild the approval experience: informative diff-first prompts, risk-tiered presentation, scoped grant options, and measured approval fatigue.

REQUIREMENTS:
- File-change approvals rendered as diffs, not descriptions
- Command approvals showing the exact command, cwd, and a plain-language risk note
- Risk-proportional presentation: routine asks are compact, dangerous asks are unmissable
- Grant scoping in the prompt: once / session / this-directory choices
- Fatigue metrics: approvals per task and time-to-decision tracked

CONSTRAINTS:
- The approval prompt must contain everything needed to decide - no digging
- Dangerous approvals must not be visually confusable with routine ones

DEFINITION OF DONE:
- An edit approval shows the actual diff with additions/deletions highlighted
- A dangerous command approval is visually distinct and states blast radius
- Scoped grants reduce repeat prompts measurably within a session
- Approvals-per-task drops after the redesign without loosening the rules

COMMON FAILURES TO AVOID:
- Approval prompts that say 'Agent wants to edit file.js - Allow?' with no diff
- Every ask styled identically, so rm approval looks like ls approval
- No session grants, generating twenty prompts for twenty edits of one file
- Unmeasured fatigue, discovered only when users disable safety wholesale

Paste this into your AI builder first. It teaches the AI what you want before you give it the build prompt.

Related routes

More Harness Engineering flows that share ground with this one.

Flows category icon
Harness Engineering

Hook a Permission Layer onto Dangerous Tools

Intercept dangerous tool calls with a hook layer: pattern rules, approval gates, and blocks that the model cannot talk its way past.

01Define the danger rule set
02Implement the hook point
03Harden the patterns against evasion

+1 more steps to Done

Best forbuilders who have shipped a basic app before

4 steps60-90 minutesIntermediate
Flows category icon
Harness Engineering

Verification Loops: Never Trust an Unrun Edit

Wire verification into your agent so every edit is checked by machines - syntax, tests, behavior - before it counts as done.

01Verify at the write boundary
02Add the post-change verify step
03Gate completion on evidence

+1 more steps to Done

Best forbuilders who have shipped a basic app before

4 steps60-100 minutesIntermediate
Flows category icon
Harness Engineering

Assemble Tool Prompts Dynamically per Environment

Stop shipping one static mega-prompt: generate tool prompts from environment conditions, inject blocks only when features are on, and dedupe config to save tokens.

01Identify conditional blocks
02Build the prompt generator
03Dedupe and budget every block

+1 more steps to Done

Best forproduction-grade builds with strict verification

4 steps90-120 minutesAdvanced

Made with Emergent