secret-scanning（秘密扫描）

Secret Scanning

This skill provides procedural guidance for configuring GitHub secret scanning — detecting leaked credentials, preventing secret pushes, defining custom patterns, and managing alerts.

When to Use This Skill

Use this skill when the request involves:

• Enabling or configuring secret scanning for a repository or organization
• Setting up push protection to block secrets before they reach the repository
• Defining custom secret patterns with regular expressions
• Resolving a blocked push from the command line
• Triaging, dismissing, or remediating secret scanning alerts
• Configuring delegated bypass for push protection
• Excluding directories from secret scanning via secret_scanning.yml
• Understanding alert types (user, partner, push protection)
• Enabling validity checks or extended metadata checks
• Scanning local code changes for secrets before committing (via MCP / AI coding agent) — see the Pre-Commit Scanning via AI Coding Agents section below for the recommended plugin

How Secret Scanning Works

Secret scanning automatically detects exposed credentials across:

• Entire Git history on all branches
• Issue descriptions, comments, and titles (open and closed)
• Pull request titles, descriptions, and comments
• GitHub Discussions titles, descriptions, and comments
• Wikis and secret gists

Availability

Core Workflow — Enable Secret Scanning

Step 1: Enable Secret Protection

1. Navigate to repository Settings → Advanced Security
2. Click Enable next to "Secret Protection"
3. Confirm by clicking Enable Secret Protection

For organizations, use security configurations to enable at scale:

• Settings → Advanced Security → Global settings → Security configurations

Step 2: Enable Push Protection

Push protection blocks secrets during the push process — before they reach the repository.

1. Navigate to repository Settings → Advanced Security
2. Enable "Push protection" under Secret Protection

Push protection blocks secrets in:

• Command line pushes
• GitHub UI commits
• File uploads
• REST API requests
• REST API content creation endpoints

Step 3: Configure Exclusions (Optional)

Create .github/secret_scanning.yml to auto-close alerts for specific directories:

paths-ignore:
  - "docs/**"
  - "test/fixtures/**"
  - "**/*.example"

Limits:

• Maximum 1,000 entries in paths-ignore
• File must be under 1 MB
• Excluded paths also skip push protection checks

Best practices:

• Be as specific as possible with exclusion paths
• Add comments explaining why each path is excluded
• Review exclusions periodically — remove stale entries
• Inform the security team about exclusions

Step 4: Enable Additional Features (Optional)

Non-provider patterns — detect private keys, connection strings, generic API keys:

• Settings → Advanced Security → enable "Scan for non-provider patterns"

AI-powered generic secret detection — uses Copilot to detect unstructured secrets like passwords:

• Settings → Advanced Security → enable "Use AI detection"

Validity checks — verify if detected secrets are still active:

• Settings → Advanced Security → enable "Validity checks"
• GitHub periodically tests detected credentials against provider APIs
• Status shown in alert: active, inactive, or unknown

Extended metadata checks — additional context about who owns a secret:

• Requires validity checks to be enabled first
• Helps prioritize remediation and identify responsible teams

Core Workflow — Resolve Blocked Pushes

When push protection blocks a push from the command line:

Option A: Remove the Secret

If the secret is in the latest commit:

# Remove the secret from the file
# Then amend the commit
git commit --amend --all
git push

If the secret is in an earlier commit:

# Find the earliest commit containing the secret
git log

# Start interactive rebase before that commit
git rebase -i <COMMIT-ID>~1

# Change 'pick' to 'edit' for the offending commit
# Remove the secret, then:
git add .
git commit --amend
git rebase --continue
git push

Option B: Bypass Push Protection

1. Visit the URL returned in the push error message (as the same user)
2. Select a bypass reason:

- It's used in tests — alert created and auto-closed - It's a false positive — alert created and auto-closed - I'll fix it later — open alert created

1. Click Allow me to push this secret
2. Re-push within 3 hours

Option C: Request Bypass Privileges

If delegated bypass is enabled and you lack bypass privileges:

1. Visit the URL from the push error
2. Add a comment explaining why the secret is safe
3. Click Submit request
4. Wait for email notification of approval/denial
5. If approved, push the commit; if denied, remove the secret

For detailed bypass and delegated bypass workflows, search references/push-protection.md.

Custom Patterns

Define organization-specific secret patterns using regular expressions.

Quick Setup

1. Settings → Advanced Security → Custom patterns → New pattern
2. Enter pattern name and regex for secret format
3. Add a sample test string
4. Click Save and dry run to test (up to 1,000 results)
5. Review results for false positives
6. Click Publish pattern
7. Optionally enable push protection for the pattern

Scopes

Custom patterns can be defined at:

• Repository level — applies to that repo only
• Organization level — applies to all repos with secret scanning enabled
• Enterprise level — applies across all organizations

Copilot-Assisted Pattern Generation

Use Copilot secret scanning to generate regex from a text description of the secret type, including optional example strings.

For detailed custom pattern configuration, search references/custom-patterns.md.

Alert Management

Alert Types

Alert Lists

• Default alerts — supported provider patterns and custom patterns
• Generic alerts — non-provider patterns and AI-detected secrets (limited to 5,000 per repo)

Remediation Priority

1. Rotate the credential immediately — this is the critical action
2. Review the alert for context (location, commit, author)
3. Check validity status: active (urgent), inactive (lower priority), unknown
4. Remove from Git history if needed (time-intensive, often unnecessary after rotation)

Dismissing Alerts

Dismiss with a documented reason:

• False positive — detected string is not a real secret
• Revoked — credential has already been revoked
• Used in tests — secret is only in test code

For detailed alert types, validity checks, and REST API, search references/alerts-and-remediation.md.

Pre-Commit Scanning via AI Coding Agents

For scanning code changes for secrets inside an AI coding agent before committing, install the Advanced Security plugin which provides the run_secret_scanning MCP tool and a dedicated scanning skill.

GitHub Copilot CLI:

/plugin install advanced-security@copilot-plugins

Visual Studio Code:

• In Copilot Chat, open Chat: Plugins (or use @agentPlugins) and install the advanced-security plugin
• Then run /secret-scanning in Copilot Chat

See: Advanced Security Plugin — Secret Scanning Skill

Announced in Secret scanning in AI coding agents via the GitHub MCP Server (March 2026)

Reference Files

For detailed documentation, load the following reference files as needed:

• references/push-protection.md — Push protection mechanics, bypass workflow, delegated bypass, user push protection

- Search patterns: bypass, delegated, bypass request, command line, REST API, user push protection

• references/custom-patterns.md — Custom pattern creation, regex syntax, dry runs, Copilot regex generation, scopes

- Search patterns: custom pattern, regex, dry run, publish, organization, enterprise, Copilot

• references/alerts-and-remediation.md — Alert types, validity checks, extended metadata, generic alerts, secret removal, REST API

- Search patterns: user alert, partner alert, validity, metadata, generic, remediation, git history, REST API