Agentic Coding UX Designer

Design developer-agent interfaces that feel fast, trustworthy, interruptible, and worth returning to.

Use This For

• Prompt-to-plan-to-diff flows for AI coding assistants.
• Agent consoles where multiple tasks run in parallel.
• Trust and recovery surfaces: checkpoints, revert, named snapshots, staged diffs, approvals, and proof logs.
• "Magic" context flows: selected text, terminal output, browser screenshot, issue/PR mention, or failed test automatically becomes agent input.
• Port Daddy Agent Harbor, FleetBar, Fleet Control Center, pd-console, and swarm invocation surfaces.

Do Not Use This For

• Marketing homepages without the working product as the first viewport.
• Pure visual styling without workflow mechanics.
• Replacing human review with hidden automation.

Design Loop

flowchart TD
  A[Pick a high-frequency user job] --> B[Name the first input gesture]
  B --> C[Show agent intent and plan]
  C --> D[Expose progress and interruptions]
  D --> E[Make review cheap]
  E --> F[Make rollback obvious]
  F --> G[Leave a durable receipt]
  G --> H[Score the flow]

1. Start from one user job and one first gesture. Good first gestures are already in the user's hand: selected code, failed test, terminal block, issue link, PR review, screenshot, or voice note.
2. Show agent intent before action: model, repo/worktree, files, tools, budget, permissions, and stop condition.
3. Keep progress legible: current step, queued messages, running commands, changed files, test state, and blockers.
4. Make review local to the work: inline diffs, grouped hunks, comments, accept/reject, and "why this changed."
5. Make rollback a visible primitive, not a hidden Git trick: checkpoints, snapshots, worktree isolation, and named receipts.
6. End with a receipt: transcript, commands, tests, diff summary, unresolved risks, and next action.

Output Contract

Return:

• flow: entrypoint, context captured, agent actions, review point, rollback point, and receipt.
• scorecard: friction, context, trust, rollback, progress visibility, and comeback trigger scores.
• risks: places where the flow hides state, encourages over-trust, or creates spend/security surprises.
• artifacts: wireframe paths, screenshots, GIFs, or review notes when applicable.

Use scripts/magic_progression_score.mjs to score a proposed flow.

Anti-Patterns

Chat Box With Secret Hands

Novice: "Put an AI chat next to the code and let it do things." Expert: Agentic UX must show what context it sees, what it intends to change, what it already did, and how to undo it. Detection: The UI has no plan, command log, diff, checkpoint, or stop affordance.

Infinite Spinner Trust Fall

Novice: "Background agents can be invisible until they finish." Expert: Background work needs a heartbeat, current step, spend meter, cancel button, and inspectable transcript. Silence reads as failure. Detection: A user cannot tell whether the agent is blocked, expensive, stale, or unsafe.

Review After The Damage

Novice: "Let the agent modify everything, then ask for review at the end." Expert: Review belongs at decision points: before broad file writes, before dependency installs, before PR creation, and before merge. Detection: The first human checkpoint is after a large diff has already landed.

References

File	Load When
references/magic-progressions.md	Need concrete low-friction patterns and comeback loops.
references/design-rules.md	Need product rules and anti-patterns for agentic developer tools.
examples/expected-output.md	Need a finished flow critique example.
templates/output-template.md	Need a reusable UX flow template.
schemas/magic-flow.schema.json	Need to validate a flow score input.
scripts/magic_progression_score.mjs	Need deterministic scoring for a proposed flow.
agents/openai.yaml	Need a subagent descriptor for delegated UX design review.

Layout QA gate (mechanical — run before shipping)

Before calling any rendered page, artifact, dashboard, deck, or component done, run the mechanical overflow/collision checker. It renders the page headlessly and flags text-vs-text collisions, clipped/ellipsis-truncated elements, text escaping its container, and horizontal page scroll — the visual defects a screenshot hides and that only appear at a specific width or in one theme.

Resolve layout-overflow-guard from the active skill catalog before running it. The command below shows the standard Claude install path; use the path reported by your harness. If the skill is absent, install or sync it instead of skipping this gate.

python3 ~/.claude/skills/layout-overflow-guard/scripts/check_layout.py <file-or-url> \
  --widths 1280,1100,860,720,390 --themes light,dark

You do not need to read check_layout.py — invoke it with the Bash tool and act on its report and exit code (non-zero = a defect). The script's source never enters your context; only its findings do. Drive it to zero violations across every width and both themes before you ship. Full detail: the layout-overflow-guard skill.

<!-- BEGIN BUNDLE INDEX (auto: index_references.py) -->

Skill Bundle Index

Every file in this skill, and when to open it. Auto-generated by the repo skill-architect indexer.

root

• CHANGELOG.md — Agentic Coding Ux Designer — Changelog — - Initial skill creation - Core process defined - Reference files added
• README.md — Agentic Coding UX Designer — Design guidance for low-friction, inspectable, recoverable AI coding assistant flows.

agents/

• agents/openai.yaml — openai (data/schema)

examples/

• examples/expected-output.md — Example Output: Agentic Coding UX Designer — Name: summon skeptical reviewer 1.

references/

• references/design-rules.md — Design Rules For AI Coding Apps — Use this when critiquing or specifying a screen.
• references/magic-progressions.md — Magic Progressions For Agentic Coding UX — Use this when designing the flow that makes users return.

schemas/

• schemas/magic-flow.schema.json — magic flow.schema (data/schema)

scripts/

• scripts/magic_progression_score.mjs

templates/

• templates/output-template.md — Agentic Coding UX Flow Spec — [The operator intent this flow should satisfy.] - Audience: [who is using it] - Starting surface: [editor / terminal / FleetBar / dashboard

<!-- END BUNDLE INDEX -->