card-helper

Guides users through open-loop kanban cards one step at a time with exact links and commands.

1|Updated Nov 21, 2015
One-click install
npx skills add https://github.com/mark-brannan/dotfiles --skill card-helper-mark-brannan
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: card-helper
Source: https://github.com/mark-brannan/dotfiles/tree/main/.claude/skills/card-helper
Command: npx skills add https://github.com/mark-brannan/dotfiles --skill card-helper-mark-brannan

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Some tasks on a kanban board can only be completed by a human in a UI or account an agent cannot reach. This Skill walks the user through those open-loop cards interactively, supplying exact instructions while the user supplies the clicks. ## Core Features & Use Cases - One Step Per Message: Delivers a single URL, click path, copy-paste command, or question per message, then waits for the user to act. - Ground-Truth Recovery: Treats user reports of missing menus or buttons as ground truth, re-derives instructions, and never repeats a failed step. - Card Lifecycle Management: Reads linked context before starting, records stall findings back into the card, and deletes and commits the card when closed. - Use Case: A card on kanban.md says a GitHub setting must be toggled in the repo UI. The Skill gives the exact Settings navigation path, waits for confirmation, then moves to the next step until the loop is closed. ## Quick Start Ask the assistant to walk you through card 3 from your kanban board one step at a time.

Frequently Asked Questions about card-helper

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I get step-by-step help completing a kanban card?▼

Ask the assistant to walk you through a specific card by number or name. It reads the card's linked context first, then gives one step per message — a URL, click path, or command — and waits for your confirmation before continuing.

What kinds of tasks is interactive card walking suited for?▼

It suits open loops only a human can close: changing settings in web UIs, managing accounts, or clicking through dashboards an agent cannot access. It is not for work the agent could do autonomously in the repository.

What happens when an instruction doesn't match what I see on screen?▼

Your report is treated as ground truth. The assistant acknowledges the error in one line, asks for exactly what would resolve it (page title, visible menus, actual JSON), and re-derives the next step from what you report.

Can the assistant complete these UI tasks for me instead?▼

No. This workflow exists precisely for loops in UIs and accounts the agent cannot reach. The assistant supplies knowledge and exact instructions; you supply the clicks in your own browser or account.

What happens if a walk stalls partway through a card?▼

The assistant writes what it learned into the card or issue before ending: the step that failed, what the UI actually showed, and what would unblock it. This prevents repeated half-finished attempts with nothing recorded.