Find agent skills by outcome

Before finalizing output

verify:

- For retryable failures

include a clear recovery action in UI (for example Retry/Refresh) with proper disabled/loading transitions.

- Distinguish states clearly: idle

Creative

- Error states must be first-class UI states

not plain text dumps: use a dedicated error container/card/banner that is consistent with the current design system.

- Build a clear hierarchy: header

control strip

- Use CSS variables for tokens: color

spacing

- For mutating commands (kill process

delete file

- When path-dependent behavior exists

display active working directory in UI/help text.

- If stdout is empty

return `{}` or `[]` consistently based on expected shape.

Creative

- Every error must be rendered in-design: error UI must follow the widget's visual language (color tokens

typography

Creative

- If truncation is reported

show partial output status and avoid false-success messaging.

- Validate and normalize user inputs (path

hostname

- Prefer allow-list style validation for sensitive parameters (e.g.

command mode

- In `parseJson` mode

ensure script returns exactly one JSON payload.

- For user text interpolated into PowerShell single-quoted literals

always escape `'` -> `''`.

Data

- For structured data

return JSON with:

- `setStatus(message

tone)` where `tone` supports at least: `info`

- `runPs(script

parseJson = false

- `pick(obj

camelKey

- `await window.HTWind.invoke(powershell.exec

{ script

- Quick edits: palette

refresh cadence

- Brief bullets: commands

safety decisions

Coding & Debugging

- One fenced `html` code block containing the full

self-contained single file.

- If the user asks for a specific language

follow that explicit instruction.