Runbook — review and triage findings
Owner: security analyst · Frequency: after every completed scan ·
Last updated: 2026-08-14 · Last run: procedure verified against
../../modules/report.md and the source of strix/report/writer.py/sarif.py.
Purpose
Turn a finished scan's raw output into triaged, actionable items.
Prerequisites
- [ ] A completed (or budget-stopped) scan under
strix_runs/<run-name>/.
Procedure
Step 1 — Read the executive report first
cat strix_runs/<run-name>/penetration_test_report.md
Expected result: a narrative summary (executive summary, methodology, technical
analysis, recommendations) — this is exactly the four fields the root agent supplied
to finish_scan (../../modules/tools.md).
Step 2 — Walk the findings index, highest severity first
cat strix_runs/<run-name>/vulnerabilities.csv # sorted by severity, then timestamp
Expected result: one row per vuln-NNNN, with severity and title — use this to
prioritize which individual reports to read in full.
Step 3 — Read individual findings
cat strix_runs/<run-name>/vulnerabilities/vuln-0001.md
Expected result: full structured finding — CVSS breakdown, proof-of-concept
script, evidence, code locations. Every finding here already passed
report/dedupe.py's LLM-judged duplicate check, so this list should not contain
near-duplicates of the same root cause.
Step 4 — Wire into CI/CD if relevant
cat strix_runs/<run-name>/findings.sarif
Expected result: a SARIF 2.1.0 document consumable by GitHub code scanning or similar tooling, for gating a pull request automatically.
Step 5 — Browse interactively instead of reading raw files
strix view <run-name>
Verification
- [ ] Every finding you plan to act on has a working proof-of-concept attached — per STRIX — Overview's design principle, Strix findings are meant to already be validated, not signature matches requiring separate confirmation. If a finding looks unvalidated, treat that as worth re-checking, not as normal.
Troubleshooting
| Symptom | Likely cause | Fix |
|---|---|---|
| Two findings look like the same bug | Dedup judged them as different root causes/assets | Re-read both PoCs — if genuinely the same issue, file feedback; this is an LLM judgment call, not guaranteed perfect. |
vulnerabilities.json and vulnerabilities.csv seem out of sync mid-scan |
Both are fully rewritten on every new finding while --resume reloads them; a scan mid-flight can show a partial set |
Re-check after the scan reaches completed status. |
Rollback
Not applicable — this runbook is read-only.
Escalation
| Issue | Contact |
|---|---|
| A finding suggests active exploitation already occurred against the target | Incident response process, immediately |
| Disagreement on a severity rating | Whoever owns the CVSS scoring standard for the team |
History
| Date | Change |
|---|---|
| 2026-08-14 | Runbook authored against usestrix/strix commit 8ca0c4a9. |