sleap-support_skill

Overview

This skill handles SLEAP GitHub support workflow for issues and discussions related to talmolab/sleap. It scaffolds investigation folders, downloads post content and images, analyzes reports, and drafts friendly, copy-paste-ready responses. Use it when a user says “support”, provides an issue/discussion number like “#2512”, or asks to investigate a user report.

How this skill works

Given an issue or discussion number, the skill checks both issues and discussions, creates a dated investigation folder if none exists, and saves the post JSON. It extracts and downloads images, converts the post into a USER_POST.md, and creates an investigation README with a concise summary, key info, and next steps. The skill checks recent releases for fixes, can checkout specific tagged versions to reproduce, runs targeted analysis steps (reproduce, data checks, platform/GPU checks), and generates a polite RESPONSE_DRAFT.md with explicit troubleshooting commands.

When to use it

• User asks for support or uses the word “support”.
• A GitHub issue or discussion number is provided (e.g., “#2512”).
• You need a reproducible investigation folder and clear artifacts (post JSON, images).
• You want a step-by-step draft reply that non-technical users can follow.
• You need to check whether a reported problem was fixed in a later release.

Best practices

• Always check both issues and discussions since users post in either.
• Create a dated investigation folder and include USER_POST.md and images for traceability.
• Look for the user’s SLEAP version early; check recent releases before deep work.
• When asking users for info, provide copy-paste terminal commands and assume minimal terminal experience.
• Draft responses with a short summary, explicit next steps, and an extended technical section hidden behind details.
• Always obtain explicit approval from a developer before posting the response.

Example use cases

• User posts GUI display issue with screenshots and asks why labels aren’t showing.
• A bug report mentions a crash during inference; reproduce on the reported version and check GPU/CUDA compatibility.
• User reports tracking ID switches; analyze tracking settings and suggest sleap-track options.
• Someone uploads an SLP file with corrupted labels; request SLP via slp.sh and run sio show diagnostics.
• Feature request discussion where you summarize and mark for tracking with no immediate action.