knowledge-base（知识库）

Knowledge Base - Confluence Integration

Authentication

IMPORTANT: Credentials are injected automatically by a proxy layer. Do NOT check for CONFLUENCE_API_TOKEN in environment variables - it won't be visible to you. Just run the scripts directly; authentication is handled transparently.

Why Confluence Matters During Incidents

Before diving deep into technical investigation:

• Is there a runbook? Find documented procedures for this service/alert
• Has this happened before? Search for related post-mortems
• What's the architecture? Find service documentation
• What's the mitigation? Look for incident response guides

Available Scripts

All scripts are in .claude/skills/knowledge-base/scripts/

search_pages.py - General Search

Search across all Confluence pages for any topic.

python .claude/skills/knowledge-base/scripts/search_pages.py --query SEARCH_QUERY [--space SPACE_KEY] [--limit N]

# Examples:
python .claude/skills/knowledge-base/scripts/search_pages.py --query "payment service timeout"
python .claude/skills/knowledge-base/scripts/search_pages.py --query "database connection" --space SRE
python .claude/skills/knowledge-base/scripts/search_pages.py --query "api error" --limit 20

find_runbooks.py - Find Runbooks

Find runbooks for a specific service or alert. Searches for pages labeled as runbooks, playbooks, or SOPs.

python .claude/skills/knowledge-base/scripts/find_runbooks.py [--service SERVICE] [--alert ALERT_NAME] [--space SPACE_KEY] [--limit N]

# Examples:
python .claude/skills/knowledge-base/scripts/find_runbooks.py --service payment
python .claude/skills/knowledge-base/scripts/find_runbooks.py --alert "HighErrorRate"
python .claude/skills/knowledge-base/scripts/find_runbooks.py --service checkout --space OPS

find_postmortems.py - Find Post-mortems

Find incident post-mortems to understand historical patterns.

python .claude/skills/knowledge-base/scripts/find_postmortems.py [--service SERVICE] [--days N] [--space SPACE_KEY] [--limit N]

# Examples:
python .claude/skills/knowledge-base/scripts/find_postmortems.py --service payment
python .claude/skills/knowledge-base/scripts/find_postmortems.py --service payment --days 180
python .claude/skills/knowledge-base/scripts/find_postmortems.py --space SRE

get_page.py - Read Full Page Content

Get the full content of a specific Confluence page.

python .claude/skills/knowledge-base/scripts/get_page.py --page-id PAGE_ID
python .claude/skills/knowledge-base/scripts/get_page.py --title "Page Title" --space SPACE_KEY

# Examples:
python .claude/skills/knowledge-base/scripts/get_page.py --page-id 123456789
python .claude/skills/knowledge-base/scripts/get_page.py --title "Payment Service Runbook" --space SRE

search_cql.py - Advanced CQL Search

Use Confluence Query Language for advanced searches with filters.

python .claude/skills/knowledge-base/scripts/search_cql.py --cql "CQL_QUERY" [--limit N]

# Examples:
python .claude/skills/knowledge-base/scripts/search_cql.py --cql 'type = page AND label = "runbook"'
python .claude/skills/knowledge-base/scripts/search_cql.py --cql 'space = "SRE" AND lastModified >= now("-30d")'
python .claude/skills/knowledge-base/scripts/search_cql.py --cql 'text ~ "payment" AND label = "postmortem"'

Common CQL Patterns

Common Workflows

1. Find Runbook for Alert

# Step 1: Search for runbook by alert name
python find_runbooks.py --alert "HighErrorRate"

# Step 2: If found, read the full runbook
python get_page.py --page-id 123456789

2. Investigate Service Issues

# Step 1: Find service documentation
python search_pages.py --query "payment service architecture"

# Step 2: Check for runbooks
python find_runbooks.py --service payment

# Step 3: Look for historical incidents
python find_postmortems.py --service payment --days 90

3. Learn from Past Incidents

# Step 1: Find similar post-mortems
python find_postmortems.py --service checkout --days 180

# Step 2: Read specific post-mortem
python get_page.py --page-id 987654321

# Step 3: Search for related issues
python search_cql.py --cql 'space = "SRE" AND text ~ "checkout timeout" AND lastModified >= now("-180d")'

4. Check for Known Issues

# Search for known issues documentation
python search_pages.py --query "known issues" --space SRE

# Look for troubleshooting guides
python search_cql.py --cql 'title ~ "troubleshooting" OR label = "troubleshooting"'

Quick Commands Reference

Best Practices

When to Use Knowledge Base

1. Start of investigation - Check for existing runbooks before deep diving
2. Unknown service - Find architecture docs to understand the system
3. Recurring alerts - Look for post-mortems about similar incidents
4. Before remediation - Verify documented procedures exist

Search Strategy

1. Start broad - Use general search first (search_pages.py)
2. Narrow down - Use specific tools (find_runbooks.py, find_postmortems.py)
3. Read details - Get full page content only when needed
4. Check recency - Look for recent post-mortems to find patterns

Labels to Look For

Common Confluence labels in SRE/Ops teams:

• runbook, playbook, sop - Operational procedures
• postmortem, post-mortem, incident-review - Incident analysis
• architecture, design-doc - System design
• troubleshooting, debugging - Diagnostic guides

Anti-Patterns to Avoid

1. ❌ Reading full pages first - Start with search/find, then read details
2. ❌ Ignoring runbooks - Check knowledge base before manual investigation
3. ❌ Not learning from history - Post-mortems prevent repeated mistakes
4. ❌ Searching without space filters - Narrow results with --space when possible

Integration with Other Skills

Combine knowledge base with other investigation tools:

# 1. Find runbook
python find_runbooks.py --alert "HighMemoryUsage"

# 2. Follow runbook instructions, e.g., check pod resources
python .claude/skills/infrastructure/kubernetes/scripts/describe_pod.py payment-xxx -n otel-demo

# 3. Document findings for future post-mortem
# (Manual step - create post-mortem page after incident resolution)

Tips for Effective Searches

• Use service names - Include the service name in queries
• Check multiple spaces - Try SRE, OPS, Engineering spaces
• Look for patterns - Post-mortems reveal recurring issues
• Verify freshness - Recent docs are more accurate
• Follow links - Runbooks often link to related documentation