Exploring github

Exploring GitHub

Explore and understand GitHub repositories using the gh CLI — a zero-dependency alternative to MCP-based GitHub context servers like gitctx.

Prerequisites

• gh CLI installed and authenticated
• Works with any MCP client: Claude Code, Amp, Pi, or any agent with shell access

If gh is not found or not authenticated, stop and ask the user to install/authenticate it. If a command fails for a non-blocking reason (e.g., rate limit on a non-critical call), skip it and continue.

Context State

Maintain a mental "current repo" context. Once the user specifies a repo (e.g., rust-lang/rust), use -R owner/repo on all subsequent gh commands. Track:

• Repo: OWNER/REPO — always pass via -R
• Branch: default branch unless user switches — pass via --branch or @ref where applicable
• Path prefix: for scoped file navigation

Tool Reference

1. Repository Discovery & Metadata

Find/search repos:

gh search repos "<query>" --sort stars --limit 10 --json fullName,description,stargazersCount,language,updatedAt

View repo metadata:

gh repo view OWNER/REPO --json name,description,defaultBranchRef,isPrivate,diskUsage,forkCount,stargazerCount,languages,homepageUrl

List branches:

gh api repos/OWNER/REPO/branches --paginate --jq '.[].name'

2. Code Navigation

List directory contents (root or subdir):

gh api repos/OWNER/REPO/contents/PATH --jq '.[] | "\(.type)\t\(.name)\t\(.size // "")"' -H "Accept: application/vnd.github.v3+json"

For a specific branch:

gh api "repos/OWNER/REPO/contents/PATH?ref=BRANCH" --jq '.[] | "\(.type)\t\(.name)\t\(.size // "")"'

Read a file:

gh api repos/OWNER/REPO/contents/PATH -H "Accept: application/vnd.github.v3.raw" 2>&1 | head -500

For a specific branch:

gh api "repos/OWNER/REPO/contents/PATH?ref=BRANCH" -H "Accept: application/vnd.github.v3.raw" 2>&1 | head -500

Read multiple files (run in parallel): Read each file with the command above. Run multiple Bash calls in parallel for efficiency.

Get full repo tree (recursive):

gh api repos/OWNER/REPO/git/trees/BRANCH?recursive=1 --jq '.tree[] | select(.type=="blob") | .path' | head -200

Search code in a repo:

gh search code "<query>" --repo OWNER/REPO --limit 20 --json path,textMatches

Search code globally:

gh search code "<query>" --language LANG --limit 20 --json repository,path,textMatches

3. Issues

List/search issues:

gh issue list -R OWNER/REPO --state all --search "<query>" --limit 20 --json number,title,state,author,labels,createdAt,updatedAt

Get issue details:

gh issue view NUMBER -R OWNER/REPO --json number,title,state,body,author,labels,assignees,comments,createdAt,closedAt

List issue comments:

gh issue view NUMBER -R OWNER/REPO --json comments --jq '.comments[] | "[\(.author.login) \(.createdAt)]\n\(.body)\n---"'

4. Pull Requests

List/search PRs:

gh pr list -R OWNER/REPO --state all --search "<query>" --limit 20 --json number,title,state,author,labels,baseRefName,headRefName,createdAt,mergedAt

Get PR details:

gh pr view NUMBER -R OWNER/REPO --json number,title,state,body,author,labels,assignees,comments,reviews,additions,deletions,changedFiles,baseRefName,headRefName,mergedAt,mergedBy

Get PR diff:

gh pr diff NUMBER -R OWNER/REPO

List PR review comments:

gh api repos/OWNER/REPO/pulls/NUMBER/comments --jq '.[] | "[\(.user.login) \(.created_at) \(.path):\(.line // .original_line)]\n\(.body)\n---"'

5. Commits

List commits:

gh api repos/OWNER/REPO/commits --paginate -f sha=BRANCH -f per_page=20 --jq '.[] | "\(.sha[:8]) \(.commit.author.date) \(.commit.author.name): \(.commit.message | split("\n")[0])"'

With path filter:

gh api repos/OWNER/REPO/commits -f sha=BRANCH -f path=PATH -f per_page=20 --jq '.[] | "\(.sha[:8]) \(.commit.author.date) \(.commit.author.name): \(.commit.message | split("\n")[0])"'

With author filter:

gh api repos/OWNER/REPO/commits -f sha=BRANCH -f author=USERNAME -f per_page=20 --jq '.[] | "\(.sha[:8]) \(.commit.author.date): \(.commit.message | split("\n")[0])"'

Get commit details:

gh api repos/OWNER/REPO/commits/SHA --jq '{sha: .sha, message: .commit.message, author: .commit.author, files: [.files[] | {filename, status, additions, deletions, patch}]}'

Compare commits/refs:

gh api repos/OWNER/REPO/compare/BASE...HEAD --jq '{ahead_by, behind_by, total_commits: .total_commits, files: [.files[] | {filename, status, additions, deletions}], commits: [.commits[] | {sha: .sha[:8], message: .commit.message | split("\n")[0]}]}'

Blame a file:

gh api graphql -f query='
  query {
    repository(owner: "OWNER", name: "REPO") {
      object(expression: "BRANCH:PATH") {
        ... on Blob {
          blame {
            ranges {
              startingLine
              endingLine
              commit {
                abbreviatedOid
                message
                author { name date }
              }
            }
          }
        }
      }
    }
  }
' --jq '.data.repository.object.blame.ranges[] | "L\(.startingLine)-\(.endingLine) \(.commit.abbreviatedOid) \(.commit.author.name) (\(.commit.author.date)): \(.commit.message | split("\n")[0])"'

6. Releases

List releases:

gh release list -R OWNER/REPO --limit 20 --json tagName,name,publishedAt,isPrerelease,isDraft,isLatest

Get release details:

gh release view TAG -R OWNER/REPO --json tagName,name,body,publishedAt,author,assets,isPrerelease,isDraft

Compare releases (commits between two tags):

gh api repos/OWNER/REPO/compare/TAG1...TAG2 --jq '{total_commits: .total_commits, commits: [.commits[] | {sha: .sha[:8], message: .commit.message | split("\n")[0]}], files_changed: [.files[] | .filename]}'

7. Repository Insights

Top contributors:

gh api repos/OWNER/REPO/contributors -f per_page=20 --jq '.[] | "\(.login): \(.contributions) commits"'

Repository stats (commit activity):

gh api repos/OWNER/REPO/stats/participation --jq '"Owner commits (last 52 weeks): \(.owner | add)\nAll commits (last 52 weeks): \(.all | add)"'

Language breakdown:

gh api repos/OWNER/REPO/languages

Dependency graph (if available):

gh api repos/OWNER/REPO/dependency-graph/sbom --jq '.sbom.packages[:30][] | "\(.name) \(.versionInfo // "unknown")"'

Usage Recommendation

This skill is best used via a subagent (e.g. Claude Code's Task tool, or any agent that supports delegated subtasks). GitHub exploration generates a lot of intermediate output — file trees, code snippets, API responses — that is useful for answering the question but clutters the main conversation. Delegating to a subagent keeps that noise isolated and returns only the synthesized answer.

Exploration Strategy

Follow a broad-then-narrow approach:

1. Orient — start with repo metadata (gh repo view) and file tree to understand project shape
2. Survey — read key entry points: README, config files (package.json, Cargo.toml, go.mod), main entry files
3. Target — use code search and directory listing to locate the specific area asked about
4. Deep-read — read the relevant files thoroughly; do not skim or summarize code you have not actually read
5. Cross-reference — connect findings across concerns (trace a function from definition to usage to tests to related PRs/issues)

Run independent operations in parallel whenever possible. For example, read multiple files at once, or search code while listing a directory. When exploring unfamiliar territory, adjust your investigation based on what you discover — do not follow a rigid script.

Codebase Understanding

1. gh repo view — get overview, default branch, languages
2. gh api .../git/trees/BRANCH?recursive=1 — get file tree
3. Read key files: README, config files, entry points
4. gh search code — find specific patterns or implementations
5. Synthesize architecture understanding

Issue/PR Triage

1. gh issue list / gh pr list with filters
2. gh issue view / gh pr view for details
3. gh pr diff for code changes
4. Cross-reference with commits and blame

Release Analysis

1. gh release list — recent releases
2. gh api .../compare/TAG1...TAG2 — diff between releases
3. Check linked issues/PRs for changelog context

Output Guidelines

• Always use --json + --jq for structured, token-efficient output
• Pipe large outputs through head -N to avoid flooding context
• For file contents, limit to 500 lines unless the user requests more
• Lead with the answer — state the key finding first, then support with details
• Include evidence — reference specific files, line ranges, commit SHAs, issue/PR numbers
• Use Mermaid diagrams when explaining architecture, data flow, or component relationships
• Quantify when possible — "47 files changed across 3 packages" not "many files changed"
• When mentioning files, link to them on GitHub: [src/auth.rs](https://github.com/OWNER/REPO/blob/BRANCH/src/auth.rs#L10-L25)

Error Handling

• Rate-limited? Use gh api -H "Accept: ..." --cache 1h for cacheable requests
• 404 errors likely mean private repo without access or wrong path — verify repo name
• Large repos: use tree + targeted reads instead of recursive listing