Browser automation via agent-browser. Use when the user wants to browse, interact with, or inspect web pages.
Review agent-facing knowledge (rules, AGENTS.md, CONVENTIONS.md, README.md, domain docs) for staleness and accuracy. Use when the user asks to audit, clean, or prune knowledge. Also use when they're confused by a rule/doc that seems wrong — treat that as the starting point for a broader audit.
Reflect on session, update rules, commit with jj.
Adversarial design interview.
Ingest sources into the wiki.
Wiki health checks.
Query the personal wiki.
Read-only code review of jj changes.