name: "agent-history-hygiene" description: "Keep agent histories as durable context stores by summarizing stable knowledge and removing session chatter." domain: "team-optimization" confidence: "high"

#source: "manual audit across squad histories during reskill"

#Context

Histories should help the next session start smarter, not longer. Repeated project intros, rollout broadcasts, timestamps, and requester names inflate context without improving recall.

#Patterns

• Use Core Context for stable ownership and operating context.
• Keep Learnings for durable technical or editorial lessons that will matter again.
• Merge duplicate learnings into one stronger statement.
• Promote team-wide patterns into .squad/skills/ once they recur across several agents.
• Prefer summaries of outcomes over transcript-style chronology.

#Examples

• Good: summarize a whole migration as "automation now writes through the branch-protection workflow skill".
• Good: condense repeated onboarding updates into one Core Context bullet.
• Good: keep one canonical lesson about a pipeline contract instead of storing the same insight in multiple dated entries.

#Anti-Patterns

• Repeating "PRD now available" or milestone broadcasts in multiple histories after the team has absorbed them.
• Keeping long timestamped session logs inside an agent history.
• Storing branch names, requester names, or one-off PR trivia that does not change future decisions.