Enrich pr-review postmortem KB from Confluence incident reports

[PLNech]

Context

The pr-review plugin ships with a convention-based postmortem knowledge base (.claude/knowledge/postmortems/). Currently each repo populates its own KB manually via /pr-review:postmortem-extract.

Opportunity

Algolia's Confluence space contains postmortem reports across all products. Mining these systematically would:

1. Bootstrap any repo's KB — new repos get cross-org learnings on day one
2. Surface cross-product patterns — an incident in Search infra may prevent one in Agent Studio
3. Scale institutional memory — learnings from 50+ incidents, not just 5

Proposed approach

• Audit Confluence postmortem space (identify structure, volume, access)
• Build extraction pipeline: Confluence → structured postmortem format (anti-patterns, search strategies, safe patterns)
• Deduplicate and categorize by domain (infra, data, API, multi-tenancy, etc.)
• Ship as optional knowledge/postmortems/ bundle in this plugin, or as a separate installable knowledge pack
• Automate: new Confluence postmortems trigger extraction PR via GitHub Actions

Open questions

• Access: who owns the Confluence postmortem space? Do we need specific permissions?
• Volume: how many postmortems exist? What's the signal-to-noise ratio?
• Privacy: any postmortems with sensitive customer data that can't be extracted?

/cc @pzaragoza93

[PLNech]

Moving to algolia/skills-private — this belongs in the private repo since it involves internal incident data.