reskill

To reduce agent context overhead, you can audit agent charters and histories to identify repeated boilerplate, duplicated domain knowledge, and recurring learnings, then extract those repeated patterns into shared skills. This shrinks agent charters and makes coordination leaner.

The best way to extract recurring learnings is to measure byte sizes and select extraction candidates by overlap thresholds, such as content appearing across 2 or 3 agents. You then update `.squad/skills/{skill-name}/SKILL.md` with the required structure to promote these shared patterns out of individual agents.

The target size limits for trimming agent charters and histories are charters at 1.5KB or less and histories at 8KB or less. Enforcing these context trimming rules ensures per-agent context bloat is minimized across the team.

You should perform a team-wide charter and history audit whenever charter or history growth is suspected across multiple agents. If repeated content occurs in several places, this maintenance pass flags boilerplate and promotes entries appearing across 2 or more agents into shared skills.

Extracting shared skills generates a savings report showing before and after sizes per agent and the overall reduction. This before/after savings table helps quantify the context optimization achieved by moving duplicated domain knowledge out of individual agent charters and histories.

Agent history grows with duplicated domain knowledge when recurring learnings and repeated boilerplate are not promoted into shared skills. Running a periodic charter and history audit identifies these repeated patterns and extracts them into `.squad/skills/` to prevent ongoing context bloat.