ge-context

name	ge-context
description	Use when you need to prepare, read, or refresh full context for a GitHub issue, PR, or discussion — fetching data, downloading media, and writing structured context documents.

ge-context: Prepare Full Context Documents

Use this skill when you need to prepare or refresh a structured context document for a GitHub issue, PR, or discussion. This is the data-preparation layer — for the full "work on an issue" workflow, use the main ge skill instead.

Prerequisites

• ge package installed
• gh CLI authenticated
• ffmpeg (optional, for video frame extraction)
• anthropic package + ANTHROPIC_API_KEY (optional, for AI image descriptions)
• ImageMagick (optional, for clipboard montage via copy_images_to_clipboard)

Preparing Context

Auto-detect from URL

import ge

ctx = ge.prepare('https://github.com/owner/repo/issues/42')
# Also works with PR and discussion URLs

Explicit type

ctx = ge.prepare_issue('owner/repo', 42)
ctx = ge.prepare_pr('owner/repo', 7, include_diff=True)
ctx = ge.prepare_discussion('owner/repo', 5)

Parameters

All prepare_* functions accept:

• output_dir (str, default None) — Where to write context files. Defaults to ~/.cache/ge/<owner>/<repo>/<kind>_<number>/
• download_media_flag (bool, default True) — Whether to download images/videos
• describe_media (bool, default True) — Whether to describe images via Claude API. Requires anthropic package and ANTHROPIC_API_KEY. Fails silently if unavailable.

prepare_pr additionally accepts:

• include_diff (bool, default True) — Whether to include the full diff

Output Files

After calling prepare_*, two files are written to the output directory (returned in ctx['output_dir']):

File	Format	Contents
issue_42_context.md	Markdown	Human/agent-readable context document
issue_42_context.json	JSON	Structured data for programmatic access

The naming pattern is {kind}_{number}_context.{ext} where kind is issue, pr, or discussion.

Media files are saved to a media/ subdirectory. File types are auto-detected from content (magic bytes), and correct extensions are added (.png, .jpg, .mp4, etc.). Videos get a directory named after their UUID for extracted frames:

media/
├── <uuid>.png              ← image (with correct extension)
├── <uuid>.mp4              ← video source
└── <uuid>/                 ← video frames directory
    ├── scene_001.jpg
    └── scene_002.jpg

Context Document Structure

The markdown context document contains these sections:

1. Header — Title, state, author, dates
2. Body — Issue/PR description with media paths rewritten to local files
3. Comments — All comments in chronological order
4. Reviews (PRs only) — Review comments and verdicts
5. Diff (PRs only) — Full diff if include_diff=True
6. Analysis — Freshness assessment, signals, recommendation
7. Media Files — Manifest of downloaded images and extracted video frames
8. Image Descriptions — AI-generated descriptions of all visual files (when available)

Return Value

The prepare_* functions return a dict containing:

{
    'kind': 'issue',           # or 'pr', 'discussion'
    'number': 42,
    'title': 'Bug in auth flow',
    'body': '...',             # With local media paths
    'comments': [...],
    'analysis': {
        'recommendation': 'proceed',
        'signals': [...],
        ...
    },
    'media': {
        'images': [...],
        'video_frames': {...},
        'all_visual_files': [...],
        'manifest': [...],
        'image_descriptions': '...',  # AI-generated (or None)
    },
    'context_md': '...',       # The rendered markdown
}

Image Analysis Tools

Beyond automatic descriptions during prepare_*, two standalone tools are available:

Describe images (Claude API)

from ge.media import describe_images

# Describe one or more images
text = describe_images('screenshot.png', 'error.jpg')
text = describe_images('frame1.jpg', 'frame2.jpg',
                       prompt="What changed between these frames?")

Copy images to clipboard (montage)

from ge.media import copy_images_to_clipboard

# Create montage and copy to clipboard for pasting into Claude Code
path = copy_images_to_clipboard('img1.png', 'img2.png', 'img3.png')
# Then Cmd+V in Claude Code

Refreshing Context

To refresh context (e.g., after new comments), simply call prepare_* again — it overwrites the existing files.

Standalone Media Processing

To download media from any markdown text (not just GitHub context):

from ge.media import process_all_media

result = process_all_media(markdown_text, output_dir='/tmp/my_media')
# Returns: {'images': [...], 'video_frames': {...}, 'manifest': [...], 'rewritten_markdown': '...'}