GitHub Repo Search (No Clone)

Use this skill to locate code in a specific GitHub repo by URL, summarize matches, and deep-read a chosen file via the GitHub API.

Requirements

• gh CLI installed and authenticated (gh auth status).
• Network access to GitHub.

Inputs to Collect

• Repo URL (e.g. https://github.com/OWNER/REPO)
• Search query (the string or pattern to search)

If any input is missing, ask the user.

Workflow

1. Parse repo

- Extract OWNER and REPO from the URL.

1. Search code in the repo

- Run: ``bash gh search code --repo OWNER/REPO "<query>" ` - Prefer JSON output for structured parsing: `bash gh search code --repo OWNER/REPO "<query>" --json path,repository,sha,url,textMatches ``

1. Summarize results

- If zero results, report and ask for a refined query. - If multiple files match, list them with paths and brief snippets (from textMatches if present), then ask which file to Deep Read. - If a single file matches, proceed to Deep Read confirmation.

1. Deep Read chosen file

- Use the GitHub API to fetch content without cloning: ``bash gh api repos/OWNER/REPO/contents/<path> ` - The content field is base64. Decode it: `bash gh api repos/OWNER/REPO/contents/<path> --jq .content | base64 --decode ` - Fallback for large files (~1 MB limit on contents API): - Use the downloadurl from the contents response: `bash gh api repos/OWNER/REPO/contents/<path> --jq .downloadurl | xargs curl -L ` - Or fetch by blob SHA (from gh search code --json sha): `bash gh api repos/OWNER/REPO/git/blobs/<sha> --jq .content | base64 --decode ` - Or request raw content directly: `bash gh api repos/OWNER/REPO/contents/<path> -H "Accept: application/vnd.github.raw" ``

1. Present full file

- Show the full content and provide a short summary of the logic relevant to the query.

Usage Example

User request:

Find where rate limiting is implemented for DWS Processor API in https://github.com/PSPDFKit/PSPDFKit

Run:

gh search code --repo PSPDFKit/PSPDFKit "DWS Processor API rate limit" --json path,repository,sha,url,textMatches

If the user chooses a result like <path-from-results>, deep read it:

gh api repos/PSPDFKit/PSPDFKit/contents/<path-from-results> --jq .content | base64 --decode

Notes

• Do not clone the repository unless the user explicitly asks.
• If the file is large, ask whether to show the full file or specific sections.
• Use the default branch unless the user specifies a branch; for a specific ref:

``bash gh api repos/OWNER/REPO/contents/<path>?ref=<branch-or-sha> ``