review

Review Unreviewed Reports

Review reports that have not been reviewed yet (review_status: pending). Walks through findings for each report using the same review flow as /self-care:run stages 4, 5, and 6.

Stage 0: Parse Arguments

Parse $ARGUMENTS to extract:

• count: Optional positive integer — the number of reports to review. Defaults to all pending reports if not specified.

Example inputs:

• `` (empty) → review all pending reports
• 3 → review the 3 most recent pending reports
• 1 → review only the most recent pending report

---

Stage 1: Find Pending Reports

Use Bash to find all reports with review_status: pending in ./.self-care/reports/:

grep -l 'review_status: pending' ./.self-care/reports/*-triage.md 2>/dev/null | sort -r

This returns report files sorted newest-first (filenames start with YYYY-MM-DD).

For each report file found, verify that a corresponding data sidecar exists. The data file has the same name but with -data.json instead of -triage.md:

• Report: ./.self-care/reports/2026-04-01-abc12345-triage.md
• Data: ./.self-care/reports/2026-04-01-abc12345-data.json

Skip any report whose data sidecar is missing (output a warning: ⚠ Skipping <report_file>: data file not found).

If count is specified, take only the first count reports from the sorted list.

If no pending reports are found, output:

No unreviewed reports found in ./.self-care/reports/

And stop.

Otherwise, output:

Found <N> unreviewed report(s). Reviewing <M> report(s)...

Where <N> is the total pending and <M> is the number being reviewed (limited by count if specified).

---

Stage 2: Review Each Report

For each pending report (newest first):

2a. Load Report Data

Read the data sidecar JSON file. Extract:

• format: trace format (otel or claude-code)
• metadata: trace metadata
• cases: the full case array (type, severity, description, evidence, classification, proposedFix)
• agentContext: optional user-provided context