Morning Report

Morning Report

After each scan, Sentinel generates a markdown report at .sentinel/report-<run-id>.md. This is designed to be scannable in under 2 minutes.

Report structure

# Sentinel Report — Run #5
Generated: 2026-04-10 07:00:00
Repository: /home/user/my-project
Scope: full | Findings: 12

## Summary
  3 medium, 9 low

## Findings by Severity

### Medium (3)
  1. [dep-audit] CVE-2024-3651 in idna 3.4
  2. [docs-drift] Stale link in README.md:45
  3. [complexity] Complex function: parse_config (CC=16)

### Low (9)
  4. [todo-scanner] TODO in src/auth.py:23 (14 months old)
  ...

Finding detail

Each finding includes:

• Detector name and category
• Severity (critical, high, medium, low)
• Confidence (0.0–1.0)
• Title — one-line summary
• Description — what was found and why it matters
• Evidence — code snippets, config excerpts, git history
• File path and line numbers
• Recurrence badge — first-seen vs. recurring (with occurrence count)
• Judge verdict — ✅ confirmed, ❌ rejected, ❓ unverified (when judge runs)

Noise reduction

Multiple mechanisms reduce noise:

1. Fingerprint deduplication — identical findings across runs are tracked, not duplicated
2. Fuzzy fingerprints — findings survive file renames
3. Confidence threshold — min_confidence config filters low-confidence findings from the report
4. Severity grouping — high-severity findings first, easy to stop reading when you hit LOW
5. Finding cluster synthesis — (standard+ capability) LLM collapses related findings into root-cause groups
6. Suppression — false positives suppressed via sentinel suppress are permanently excluded

JSON output

For machine consumption:

sentinel scan /repo --json-output

Returns structured JSON with run, findings, and report_path.

Viewing past reports

sentinel findings                          # latest run
sentinel findings --run 3                  # specific run
sentinel findings --severity medium        # filter by severity
sentinel findings --detector todo-scanner  # filter by detector

Or use the Web UI for a browser-based dashboard.