/insights undercounts sessions — excludes subagents and conductor workspaces

Problem

The /insights command significantly undercounts actual usage by excluding subagent sessions from its totals. For power users who heavily leverage parallel multi-agent workflows (Task tool, teams, conductor workspaces), the reported numbers don't reflect real usage.

My actual numbers vs. what /insights reports:

| Metric | /insights reports | Actual |
|---|---|---|
| Total sessions | 375 | 1,321 |
| Subagent sessions | excluded entirely | 946 |
| Messages (JSONL lines) | 2,408 | 210,033 |
| Conductor workspace sessions | partially counted | 610 across 102 workspaces |

The report says "375 sessions total · 308 analyzed" but there are 1,321 .jsonl files across 125 distinct project directories.

How I verified

# Top-level sessions only (what /insights counts)
find ~/.claude/projects -name "*.jsonl" -not -path "*/subagents/*" | wc -l
# 375

# ALL sessions including subagents
find ~/.claude/projects -name "*.jsonl" | wc -l
# 1321

# Subagent sessions excluded
find ~/.claude/projects -name "*.jsonl" -path "*/subagents/*" | wc -l
# 946

# Total JSONL lines
find ~/.claude/projects -name "*.jsonl" -print0 | xargs -0 wc -l | tail -1
# 210033

Expected behavior

/insights should:

1. Count subagent sessions in the headline total (or at minimum show them as a separate line like "375 sessions + 946 subagent sessions")
2. Analyze subagent sessions for the usage narrative — subagent work is real work that reveals usage patterns (parallel reviews, batch content ops, team coordination)
3. Accurately count messages across all session types
4. Include conductor workspace sessions in project area breakdowns — these are full autonomous agent sessions, not throwaway temp work

Context

I use Claude Code heavily with parallel multi-agent workflows — spawning 8-20 agents for code reviews, blog audits, and codebase exploration. The conductor workspaces (named after cities like AS001-kingston, AS001-douala, etc.) represent significant autonomous work sessions. Excluding them makes the report miss ~72% of actual sessions.

🤖 Generated with Claude Code