[quickbeard]

Summary

Follow-up to #373. The analysis web app is switching its reports from log-derived analytics to the session stats pushed at launch (oadtq/z8l-ai-hub#9, plus the reports PR). Four report breakdowns relied on fields the stats didn't carry, so each SessionStats record now includes them as optional { name: count } maps:

Field	Contents	Reports that use it
agents	Assistant messages per agent (build, plan, …), top-level session only	AI usage "by mode"
skills	Skill uses by name: skill tool calls, plus reads of <name>/SKILL.md	Top skills, proficiency skill diversity
skills_created	Skills whose SKILL.md the session wrote (completed writes only)	"Skills created"
file_extensions	Distinct files read or modified, by lowercased extension	File-type chart, proficiency language breadth

• These match what ai-hub-worker's skills / file_paths / model_mode enrichers derived from markdown.
• Subagent sessions count toward skills and files but not toward agents, since a subagent's turns aren't the session's mode.
• The maps are additive to payload v1. The backend validator treats them as optional, so this doesn't need to ship in lockstep with it.
• No DIVERGENCES.md row: packages/codev-upload is fork-only. No CHANGELOG entry, matching feat(upload): push session usage metrics at launch; logs upload only via codev upload --logs #373.

Test plan

• bun test in packages/codev-upload (58 tests, including a new one covering all four maps: subagent handling, denied writes, case-folded and extensionless files)
• bun typecheck in packages/codev-upload
• End-to-end against a local Supabase running the ai-hub-web reports branch: 52 real sessions re-pushed with the maps filled in (agents on all 52, file extensions on 24, skills on 2), and the rebuilt codev_sessions view exposes them as mode, skills_invoked, file_extensions

🤖 Generated with Claude Code