Routes GitHub URLs to efficient API calls instead of fetching heavy HTML to optimize context window usage.
- GitHub stars
- 49
- Files
- 5
- Last commit
- Nov 11, 2026
- License
- MIT
- Allowed tools
- —
What does the github-url-intercept skill do?
github-url-intercept routes GitHub URLs to efficient API calls instead of fetching heavy HTML to optimize context window usage.
Summary generated from the skill's SKILL.md.
About the github-url-intercept skill
Developers who paste GitHub links into an AI client can use this skill to prevent context window exhaustion. Instead of fetching heavy HTML pages that can consume millions of tokens, this tool intercepts any github.com URL and routes it to efficient API calls or specialized Python scripts. This process transforms massive web pages into small, structured JSON payloads ranging from 1 to 50KB.
When an agent detects a URL for a pull request, issue, commit, gist, or file, it automatically parses the path to identify the resource type. It then executes specific commands like get_pr_context.py or gh api calls to retrieve only the necessary data. This approach is designed for reading and analyzing existing content rather than performing mutations like posting comments or creating new pull requests.
What an agent can do with it
- 1Retrieve specific issue contextA developer can ask for information regarding a GitHub issue, which triggers a targeted API call to get the issue data.
- 2Examine code files and commitsYou can request a review of a specific file or commit by having the agent use the GitHub API to fetch the content directly.
- 3Fetch specific discussion commentsWhen a URL contains a fragment like a discussion ID, the skill extracts that ID to call the specific comment endpoint.
Written from the skill's SKILL.md. Every use case is checked against a quote from it.
How to install the github-url-intercept skill
Each command copies the skill folder into the client's skills directory. Restart the client or start a new session to load it.Install the github-url-intercept skill in Plugin
/plugin marketplace add rjmurillo/ai-agents /plugin install project-toolkit@ai-agents
Install the github-url-intercept skill in Claude Code
~/.claude/skills/github-url-intercept/SKILL.md
$ npx degit rjmurillo/ai-agents/src/claude/skills/github-url-intercept ~/.claude/skills/github-url-interceptInstall the github-url-intercept skill in Claude Code (project)
.claude/skills/github-url-intercept/SKILL.md
$ npx degit rjmurillo/ai-agents/src/claude/skills/github-url-intercept .claude/skills/github-url-interceptInstall the github-url-intercept skill in Codex
~/.agents/skills/github-url-intercept/SKILL.md
$ npx degit rjmurillo/ai-agents/src/claude/skills/github-url-intercept ~/.agents/skills/github-url-interceptInstall the github-url-intercept skill in Gemini CLI
~/.gemini/skills/github-url-intercept/SKILL.md
$ npx degit rjmurillo/ai-agents/src/claude/skills/github-url-intercept ~/.gemini/skills/github-url-interceptInstall the github-url-intercept skill in Claude.ai
npx degit rjmurillo/ai-agents/src/claude/skills/github-url-intercept github-url-intercept && zip -r github-url-intercept.zip github-url-intercept # claude.ai → Customize → Skills → + → Create skill → Upload a skill → github-url-intercept.zip
Automated checks
Ran Nov 14, 2026Full report →Passed our automated checks. That is not a security audit: read the scripts before you let an agent run them. How checks work
- Source repository linked and reachablerjmurillo/ai-agents/src/claude/skills/github-url-intercept is public and reachablegithub.com/rjmurillo/ai-agents/tree/HEAD/src/claude/skills/g ↗Pass
- Open source license (OSI)MITspdx.org/licenses/MIT.html ↗Pass
- Commit or release in the last yearLast commit 2026-11-11github.com/rjmurillo/ai-agents/commits ↗Pass
- No download-and-run or credential theft in scripts3 scripts read; no download-and-run, remote eval, base64 execution or credential uploadPass
- No hidden instructions in SKILL.mdNo override phrases, safeguard switches, credential requests or hidden charactersgithub.com/rjmurillo/ai-agents/blob/HEAD/src/claude/skills/g ↗Pass
- Pre-approved tools are scopedNo pre-approved tools; the agent asks before using onegithub.com/rjmurillo/ai-agents/blob/HEAD/src/claude/skills/g ↗Pass
- SKILL.md follows the Agent Skills formatValid name and description; says when to use it; 399 linesgithub.com/rjmurillo/ai-agents/blob/HEAD/src/claude/skills/g ↗Pass
SKILL.md
View on GitHubGitHub URL Intercept
CRITICAL: This skill activates AUTOMATICALLY when you see ANY
github.comURL. Do NOT useweb_fetch,curl, or any browser-based fetch on GitHub URLs. Doing so wastes 1-2.5 MILLION tokens on HTML that provides no useful data.
MANDATORY BEHAVIOR: Parse the URL → Route to API → Return structured JSON.
Quick Reference (Copy-Paste Commands)
When you see a GitHub URL, use these commands immediately:
SCRIPTS_DIR="${COPILOT_PLUGIN_ROOT:-${CLAUDE_PLUGIN_ROOT:-.claude}}/skills/github/scripts"
# PR URL → Use this
python3 "$SCRIPTS_DIR/pr/get_pr_context.py" --pull-request {n} --owner {owner} --repo {repo}
# Issue URL → Use this
python3 "$SCRIPTS_DIR/issue/get_issue_context.py" --issue {n} --owner {owner} --repo {repo}
# File/blob URL → Use this
gh api repos/{owner}/{repo}/contents/{path}?ref={ref}
# Commit URL → Use this
gh api repos/{owner}/{repo}/commits/{sha}
# Gist URL → Use this
gh api gists/{id}
# Comment fragment (#discussion_r{id}) → Use this
gh api repos/{owner}/{repo}/pulls/comments/{id}
Triggers
| Phrase | Action |
|---|---|
Any github.com URL in user input (even bare URL pasted alone) |
Parse URL type and route to API |
analyze / research / scan + GitHub URL |
Route based on URL type |
| URL + question (e.g., "...#r123 what's the tracking issue?") | Extract fragment, call specific API |
| Multiple GitHub URLs in one prompt | Process each URL, batch API calls |
When to Use
Use this skill when:
- Any
github.comURL appears in user input (automatic activation) - User pastes a PR, issue, commit, file, or actions URL
- User asks to "analyze" or "review" a GitHub link
Use github instead when:
- Performing mutations (creating PRs, posting comments, adding labels)
- Operating on the current repository without a URL
URL Patterns (Detailed Reference)
| Pattern | Example | Why Intercept |
|---|---|---|
github.com/.../pull/ |
https://github.com/owner/repo/pull/<number> |
PR HTML is 5-10MB |
github.com/.../pull/.../checks |
https://github.com/owner/repo/pull/<number>/checks?check_run_id=... |
CI checks page bloat |
github.com/.../pull/.../files or /changes |
https://github.com/owner/repo/pull/<number>/files#r123456 |
Diff view with comment fragment |
github.com/.../issues/ |
https://github.com/owner/repo/issues/<number> |
Issue HTML is 2-5MB |
github.com/.../actions/runs/ |
https://github.com/owner/repo/actions/runs/123/job/456 |
Workflow run page bloat |
github.com/.../blob/ |
https://github.com/owner/repo/blob/main/file.py |
File page has nav bloat |
github.com/.../tree/ |
https://github.com/owner/repo/tree/main/src |
Directory listing bloat |
github.com/.../commit/ |
https://github.com/owner/repo/commit/abc123 |
Commit page overhead |
github.com/.../compare/ |
https://github.com/owner/repo/compare/main...feat |
Diff page overhead |
gist.github.com/... |
https://gist.github.com/owner/0123456789abcdef0123 |
Gist HTML includes page chrome |
gist.githubusercontent.com/... |
https://gist.githubusercontent.com/owner/0123456789abcdef0123/raw/file.txt |
Raw gist content |
Fragment #discussion_r{id} |
Review comment ID in /changes or /files URL |
Extract ID, call API directly |
Fragment #issuecomment-{id} |
Issue comment ID | Extract ID, call API directly |
Fragment #pullrequestreview-{id} |
Review ID | Extract ID, call API directly |
Fragment #r{id} (short form) |
Review comment in /changes#r123 |
Same as #discussion_r{id} |
Decision Flow
GitHub URL detected in user input
│
├─ Is gist.github.com/{owner}/{id}?
│ Yes → gh api gists/{id}
│
├─ Has fragment (#pullrequestreview-, #discussion_r, #issuecomment-)?
│ Yes → Extract ID, use gh api for specific comment/review
│
├─ Is /pull/{n}?
│ Yes → get_pr_context.py --pull-request {n} --owner {o} --repo {r}
│ (or get_pr_review_comments.py / get_pr_review_threads.py for comments)
│
├─ Is /issues/{n}?
│ Yes → get_issue_context.py --issue {n} --owner {o} --repo {r}
│
├─ Is /blob/{ref}/{path} or /tree/{ref}/{path}?
│ Yes → gh api repos/{o}/{r}/contents/{path}?ref={ref}
│
├─ Is /commit/{sha}?
│ Yes → gh api repos/{o}/{r}/commits/{sha}
│
└─ Is /compare/{base}...{head}?
Yes → gh api repos/{o}/{r}/compare/{base}...{head}
Process
Phase 1: URL Detection and Parsing
| Step | Action | Verification |
|---|---|---|
| 1.1 | Detect github.com URL in user input | URL pattern matched |
| 1.2 | Extract owner/repo, or gist owner/id, from path | Required values non-empty |
| 1.3 | Identify URL type (gist, pull, issues, blob, tree, commit, compare) | Type classified |
| 1.4 | Extract fragment ID if present | Fragment parsed or null |
Phase 2: Route Selection
| Step | Action | Verification |
|---|---|---|
| 2.1 | Check if github skill script exists for URL type | Script path resolved |
| 2.2 | If script exists → use github skill (primary route) | Script invocation planned |
| 2.3 | If no script → use gh api (fallback route) | API command constructed |
| 2.4 | For fragments → always use gh api with specific endpoint | Endpoint includes ID |
Phase 3: Execution
| Step | Action | Verification |
|---|---|---|
| 3.1 | Execute selected command | Command runs without error |
| 3.2 | Receive structured JSON response | Success: true for scripts |
| 3.3 | Parse relevant fields for user query | Response processed |
URL Routing Table
Primary: Use GitHub Skill Scripts
| URL Pattern | Script | Parameters |
|---|---|---|
/pull/{n} |
get_pr_context.py |
--pull-request {n} --owner {o} --repo {r} |
/pull/{n} (with diff) |
get_pr_context.py |
--pull-request {n} --include-diff |
/pull/{n} (review comments) |
get_pr_review_comments.py |
--pull-request {n} |
/pull/{n} (review threads) |
get_pr_review_threads.py |
--pull-request {n} |
/pull/{n}/checks |
get_pr_checks.py |
--pull-request {n} --owner {o} --repo {r} |
/issues/{n} |
get_issue_context.py |
--issue {n} --owner {o} --repo {r} |
Script location: .claude/skills/github/scripts/
Fallback: Raw gh Commands
Use only when no script exists for the operation:
| URL Pattern | API Call |
|---|---|
/pull/{n}#pullrequestreview-{id} |
gh api repos/{o}/{r}/pulls/{n}/reviews/{id} |
/pull/{n}#discussion_r{id} |
gh api repos/{o}/{r}/pulls/comments/{id} |
/pull/{n}/files#r{id} or /changes#r{id} |
gh api repos/{o}/{r}/pulls/comments/{id} |
/pull/{n}#issuecomment-{id} |
gh api repos/{o}/{r}/issues/comments/{id} |
/pull/{n}/checks |
gh api repos/{o}/{r}/check-runs?head_sha=... or use get_pr_checks.py --owner {o} --repo {r} |
/issues/{n}#issuecomment-{id} |
gh api repos/{o}/{r}/issues/comments/{id} |
/actions/runs/{run_id} |
gh api repos/{o}/{r}/actions/runs/{run_id} |
/actions/runs/{run_id}/job/{job_id} |
gh api repos/{o}/{r}/actions/jobs/{job_id} |
/blob/{ref}/{path} |
gh api repos/{o}/{r}/contents/{path}?ref={ref} |
/tree/{ref}/{path} |
gh api repos/{o}/{r}/contents/{path}?ref={ref} |
/commit/{sha} |
gh api repos/{o}/{r}/commits/{sha} |
/compare/{base}...{head} |
gh api repos/{o}/{r}/compare/{base}...{head} |
gist.github.com/{owner}/{id} |
gh api gists/{id} |
URL Parsing Pattern
Extract owner, repo, and resource from GitHub URLs:
https://github.com/{owner}/{repo}/pull/{number}
https://github.com/{owner}/{repo}/issues/{number}
https://github.com/{owner}/{repo}/blob/{ref}/{path}
https://github.com/{owner}/{repo}/tree/{ref}/{path}
https://github.com/{owner}/{repo}/commit/{sha}
https://github.com/{owner}/{repo}/compare/{base}...{head}
https://gist.github.com/{owner}/{id}
https://gist.github.com/{id}
https://gist.github.com/{owner}/{id}.js
https://gist.githubusercontent.com/{owner}/{id}/raw/{revision}/{path}
Fragment extraction (when present):
#pullrequestreview-{id}→ Review ID#discussion_r{id}→ Discussion comment ID#issuecomment-{id}→ Issue comment ID
Why This Matters (CRITICAL)
Fetching GitHub HTML is catastrophic for your context window:
| Method | Response Size | Token Cost | Time | Usability |
|---|---|---|---|---|
| ❌ HTML fetch | 5-10 MB | 1-2.5M tokens | 10-30s | UNUSABLE - HTML noise, no structured data |
| ✅ API call | 1-50 KB | 250-12K tokens | 0.5-2s | Clean JSON with exactly what you need |
| ✅ Script | 1-50 KB | 250-12K tokens | 0.5-2s | Structured output, error handling |
Impact: 100-1000x reduction in token consumption.
If you fetch GitHub HTML directly, you will:
- Consume your entire context window on ONE page
- Get no useful structured data (just HTML soup)
- Be unable to process subsequent user requests
- Need to start a new conversation
ALWAYS use this skill when you see a GitHub URL.
Examples
Bare URL Pasted (Most Common!)
Input: "https://github.com/owner/repo/pull/<number>/checks?check_run_id=59355308734"
Action:
1. Parse: owner=rjmurillo, repo=ai-agents, pr=735, type=checks
2. Route: python3 "$SCRIPTS_DIR/pr/get_pr_checks.py" --pull-request 735 --owner rjmurillo --repo ai-agents
Gist URL
Input: "https://gist.github.com/owner/0123456789abcdef0123"
Action:
1. Parse: owner=owner, gist=0123456789abcdef0123, type=gist
2. Route: gh api "gists/0123456789abcdef0123"
Ownerless URLs, query strings, file fragments, .js embed URLs, raw-content
URLs, revision SHAs, and /revisions or /raw suffixes route through the gist
API. Pinned revisions stay pinned. Raw file URLs return only the requested file.
URL with Question After It
Input: "https://github.com/owner/repo/pull/<number>/changes#r2656144507 are the graph refactoring items part of Issue <N>?"
Action:
1. Extract fragment: r2656144507 (review comment ID)
2. Call: gh api "repos/owner/repo/pulls/comments/2656144507"
3. Answer user's question using the comment content
Multiple URLs in One Prompt
Input: "https://github.com/owner/repo/pull/<number>#discussion_r123 https://github.com/owner/repo/pull/<number>#discussion_r456"
Action:
1. Parse each URL
2. Batch: gh api "repos/owner/repo/pulls/comments/123"
3. Batch: gh api "repos/owner/repo/pulls/comments/456"
Research/Analyze Pattern
Input: "analyze https://github.com/modu-ai/moai-adk for insights"
Action:
1. Use deepwiki MCP or gh api to get repo info
2. Call: gh api "repos/modu-ai/moai-adk" (NOT web_fetch!)
CI/Actions Run URL
Input: "https://github.com/owner/repo/actions/runs/20675405338/job/59362398542?pr=740"
Action:
1. Extract: run_id=20675405338, job_id=59362398542
2. Call: gh api "repos/owner/repo/actions/jobs/59362398542"
PR URL → Script
Input: "Review this: https://github.com/owner/repo/pull/<number>"
Action:
python3 "$SCRIPTS_DIR/pr/get_pr_context.py" --pull-request 123 --owner owner --repo repo
File URL → API
Input: "would a hook like https://github.com/ruvnet/claude-flow/blob/main/.claude/settings.json help?"
Action:
gh api "repos/ruvnet/claude-flow/contents/.claude/settings.json?ref=main"
Anti-Patterns (NEVER DO THESE)
| ❌ NEVER | Why It's Catastrophic | ✅ Do This Instead |
|---|---|---|
web_fetch("https://github.com/...") |
5-10 MB HTML, 1-2.5M tokens WASTED | Parse URL, use script or gh api |
curl https://github.com/... |
Same catastrophic result | Use gh CLI for authentication + JSON |
fetch / requests.get on GitHub URLs |
Same catastrophic result | Route through this skill |
gh pr view without --json |
Unstructured text output | Use get_pr_context.py for structured JSON |
| Fetching full page to find one comment | Fetches 5MB to read 500 bytes | Extract fragment ID (#discussion_r...), call specific endpoint |
| Ignoring GitHub URLs in user input | User expects you to understand the link | ALWAYS parse and route |
| Hardcoding owner/repo in commands | Breaks when user shares fork/different repo | Extract from URL path |
RED FLAG PHRASES - If you're about to do any of these, STOP:
- "Let me fetch that page..."
- "I'll retrieve the content from that URL..."
- "Accessing the GitHub page..."
These indicate you're about to waste millions of tokens. Use this skill instead.
Scripts
test_url_routing.py
Routes GitHub URLs to appropriate API access methods with CWE-78 command injection protection.
python3 "${COPILOT_PLUGIN_ROOT:-${CLAUDE_PLUGIN_ROOT:-.claude}}/skills/github-url-intercept/scripts/test_url_routing.py" --url <github-url>
Related Skills
| Skill | When to Use |
|---|---|
| github | Full PR/issue operations (mutations, reactions, labels) |
| pr-comment-responder | Systematic PR review response |
Verification
After routing a GitHub URL, run the bundled validator and require exit 0:
python3 "${COPILOT_PLUGIN_ROOT:-${CLAUDE_PLUGIN_ROOT:-.claude}}/skills/github-url-intercept/scripts/test_url_routing.py" --url "$GITHUB_URL"
echo "exit=$?" # must be 0; exit 1 means invalid URL or no routing available
-
test_url_routing.pyexited 0 (non-zero = invalid URL or unroutable type; routing is not valid) - Extracted owner/repo or gist owner/id from URL path
- Identified URL type (gist, PR, issue, blob, commit, compare)
- Extracted fragment ID if present (#discussion_r, #issuecomment-, #pullrequestreview-)
- Selected appropriate github skill script (primary) or gh command (fallback)
- Did NOT use web_fetch, curl, or browser-based fetch on the URL
- Received structured JSON response with
success: true
Files in this skill
5github-url-intercept skill questions
How do I install the github-url-intercept skill in Claude Code?
Run npx degit rjmurillo/ai-agents/src/claude/skills/github-url-intercept ~/.claude/skills/github-url-intercept. Claude Code loads skills from ~/.claude/skills (all projects) or .claude/skills (one project) and uses a skill when its description matches the task.
Which clients can use the github-url-intercept skill?
Any client that reads the Agent Skills format: Claude Code, Codex, Gemini CLI and claude.ai (upload as a ZIP). Scripts in the skill run only where the client allows code execution.
Why should I not use web_fetch on GitHub URLs?
Fetching GitHub HTML directly is extremely inefficient, consuming 5-10MB and up to 2.5 million tokens, which can exhaust your context window.
When should I use the standard github skill instead of this intercept skill?
Use the standard github skill when you need to perform mutations like creating pull requests, posting comments, or adding labels.
What types of GitHub URLs does this skill handle?
It handles a wide range of patterns including pulls, issues, blobs, trees, commits, compares, and gists.