Weekly review

Input:

scoutjobs stats --json
scoutjobs stats --funnel --json   # what works: answer, interview and offer counts by source, fit band, month applied
scoutjobs job list --json --compact
scoutjobs decision summary --json
scoutjobs session list --limit 50 --json
scoutjobs learn list --json

Sections (write them to $(scoutjobs path)/reviews/<YYYY-WW>.md and show the user)

1. Numbers: applied this week vs target, interviews held, offers, response rate, average fit of applied jobs. Compare with last week's review file if it exists.
2. Funnel health: where jobs got stuck. Many ranked not drafted = application friction, suggest batching. Many applied with no response and avg fit under 65 = aim higher on fit, lower on volume. Interviews failing at one stage = point mock-interview there. What works: from the funnel, name the source and the fit band with the best answer rate (only groups with 3+ applications) and the one with the worst, with numbers; days is the median days to an answer. Suggest moving next week's applications toward the better one.
3. Close out: show scoutjobs job sweep (applied and interviewing jobs quiet past stale_days) and ask: close all, pick some, or skip; then scoutjobs job sweep --yes or scoutjobs job sweep <id>... --yes. List due follow-ups from scoutjobs job followup --json (next is due; each job gets at most 2).
4. Learn: run the learn-preferences skill now (it confirms with the user).
5. Skills: learning hours this week, items done, upskill suggestion if a gap caused a rejection.
6. Next week: three commitments, each with a number (for example "apply to 4 jobs with fit 70+", "two mock system design sessions", "3 hours Kubernetes").

Log: scoutjobs session log --kind review --topic "weekly review <YYYY-WW>" --notes "<the three commitments>".

Tone: factual, short. The user reads this on a Friday.